diff --git a/artiq/coredevice/ad9910.py b/artiq/coredevice/ad9910.py
index c7e1b49a1..9bdfd37fb 100644
--- a/artiq/coredevice/ad9910.py
+++ b/artiq/coredevice/ad9910.py
@@ -1,13 +1,27 @@
 from numpy import int32, int64
 
-from artiq.language.core import kernel, delay, portable
-from artiq.language.units import us, ns, ms
+from artiq.language.core import (
+    kernel, delay, portable, delay_mu, now_mu, at_mu)
+from artiq.language.units import us, ms
 
 from artiq.coredevice import spi2 as spi
 from artiq.coredevice import urukul
+# Work around ARTIQ-Python import machinery
 urukul_sta_pll_lock = urukul.urukul_sta_pll_lock
+urukul_sta_smp_err = urukul.urukul_sta_smp_err
 
 
+__all__ = [
+    "AD9910",
+    "PHASE_MODE_CONTINUOUS", "PHASE_MODE_ABSOLUTE", "PHASE_MODE_TRACKING"
+]
+
+
+_PHASE_MODE_DEFAULT = -1
+PHASE_MODE_CONTINUOUS = 0
+PHASE_MODE_ABSOLUTE = 1
+PHASE_MODE_TRACKING = 2
+
 _AD9910_REG_CFR1 = 0x00
 _AD9910_REG_CFR2 = 0x01
 _AD9910_REG_CFR3 = 0x02
@@ -49,12 +63,21 @@ class AD9910:
         Urukul CPLD instance).
     :param pll_cp: DDS PLL charge pump setting.
     :param pll_vco: DDS PLL VCO range selection.
+    :param sync_delay_seed: SYNC_IN delay tuning starting value.
+        To stabilize the SYNC_IN delay tuning, run :meth:`tune_sync_delay` once
+        and set this to the delay tap number returned (default: -1 to signal no
+        synchronization and no tuning during :meth:`init`).
+    :param io_update_delay: IO_UPDATE pulse alignment delay.
+        To align IO_UPDATE to SYNC_CLK, run :meth:`tune_io_update_delay` and
+        set this to the delay tap number returned.
     """
     kernel_invariants = {"chip_select", "cpld", "core", "bus",
-            "ftw_per_hz", "pll_n", "pll_cp", "pll_vco"}
+                         "ftw_per_hz", "pll_n", "io_update_delay",
+                         "sysclk_per_mu"}
 
     def __init__(self, dmgr, chip_select, cpld_device, sw_device=None,
-            pll_n=40, pll_cp=7, pll_vco=5):
+                 pll_n=40, pll_cp=7, pll_vco=5, sync_delay_seed=-1,
+                 io_update_delay=0):
         self.cpld = dmgr.get(cpld_device)
         self.core = self.cpld.core
         self.bus = self.cpld.bus
@@ -68,14 +91,64 @@ class AD9910:
         assert self.cpld.refclk/4 <= 60e6
         sysclk = self.cpld.refclk*pll_n/4  # Urukul clock fanout divider
         assert sysclk <= 1e9
-        self.ftw_per_hz = 1./sysclk*(int64(1) << 32)
+        self.ftw_per_hz = (1 << 32)/sysclk
+        self.sysclk_per_mu = int(round(sysclk*self.core.ref_period))
+        assert self.sysclk_per_mu == sysclk*self.core.ref_period
         assert 0 <= pll_vco <= 5
         vco_min, vco_max = [(370, 510), (420, 590), (500, 700),
-                (600, 880), (700, 950), (820, 1150)][pll_vco]
+                            (600, 880), (700, 950), (820, 1150)][pll_vco]
         assert vco_min <= sysclk/1e6 <= vco_max
         self.pll_vco = pll_vco
         assert 0 <= pll_cp <= 7
         self.pll_cp = pll_cp
+        self.sync_delay_seed = sync_delay_seed
+        self.io_update_delay = io_update_delay
+        self.phase_mode = PHASE_MODE_CONTINUOUS
+
+    @kernel
+    def set_phase_mode(self, phase_mode):
+        """Set the default phase mode.
+
+        for future calls to :meth:`set` and
+        :meth:`set_mu`. Supported phase modes are:
+
+        * :const:`PHASE_MODE_CONTINUOUS`: the phase accumulator is unchanged
+          when changing frequency or phase. The DDS phase is the sum of the
+          phase accumulator and the phase offset. The only discontinuous
+          changes in the DDS output phase come from changes to the phase
+          offset. This mode is also knows as "relative phase mode".
+          :math:`\phi(t) = q(t^\prime) + p + (t - t^\prime) f`
+
+        * :const:`PHASE_MODE_ABSOLUTE`: the phase accumulator is reset when
+          changing frequency or phase. Thus, the phase of the DDS at the
+          time of the change is equal to the specified phase offset.
+          :math:`\phi(t) = p + (t - t^\prime) f`
+
+        * :const:`PHASE_MODE_TRACKING`: when changing frequency or phase,
+          the phase accumulator is cleared and the phase offset is offset
+          by the value the phase accumulator would have if the DDS had been
+          running at the specified frequency since a given fiducial
+          time stamp. This is functionally equivalent to
+          :const:`PHASE_MODE_ABSOLUTE`. The only difference is the fiducial
+          time stamp. This mode is also known as "coherent phase mode".
+          The default fiducial time stamp is 0.
+          :math:`\phi(t) = p + (t - T) f`
+
+        Where:
+
+        * :math:`\phi(t)`: the DDS output phase
+        * :math:`q(t) = \phi(t) - p`: DDS internal phase accumulator
+        * :math:`p`: phase offset
+        * :math:`f`: frequency
+        * :math:`t^\prime`: time stamp of setting :math:`p`, :math:`f`
+        * :math:`T`: fiducial time stamp
+        * :math:`t`: running time
+
+        .. warning:: This setting may become inconsistent when used as part of
+            a DMA recording. When using DMA, it is recommended to specify the
+            phase mode explicitly when calling :meth:`set` or :meth:`set_mu`.
+        """
+        self.phase_mode = phase_mode
 
     @kernel
     def write32(self, addr, data):
@@ -85,10 +158,10 @@ class AD9910:
         :param data: Data to be written
         """
         self.bus.set_config_mu(urukul.SPI_CONFIG, 8,
-                urukul.SPIT_DDS_WR, self.chip_select)
+                               urukul.SPIT_DDS_WR, self.chip_select)
         self.bus.write(addr << 24)
         self.bus.set_config_mu(urukul.SPI_CONFIG | spi.SPI_END, 32,
-                urukul.SPIT_DDS_WR, self.chip_select)
+                               urukul.SPIT_DDS_WR, self.chip_select)
         self.bus.write(data)
 
     @kernel
@@ -98,11 +171,11 @@ class AD9910:
         :param addr: Register address
         """
         self.bus.set_config_mu(urukul.SPI_CONFIG, 8,
-                urukul.SPIT_DDS_WR, self.chip_select)
+                               urukul.SPIT_DDS_WR, self.chip_select)
         self.bus.write((addr | 0x80) << 24)
-        self.bus.set_config_mu(urukul.SPI_CONFIG | spi.SPI_END
-                | spi.SPI_INPUT, 32,
-                urukul.SPIT_DDS_RD, self.chip_select)
+        self.bus.set_config_mu(
+            urukul.SPI_CONFIG | spi.SPI_END | spi.SPI_INPUT,
+            32, urukul.SPIT_DDS_RD, self.chip_select)
         self.bus.write(0)
         return self.bus.read()
 
@@ -115,13 +188,13 @@ class AD9910:
         :param data_low: Low (LSB) 32 data bits
         """
         self.bus.set_config_mu(urukul.SPI_CONFIG, 8,
-                urukul.SPIT_DDS_WR, self.chip_select)
+                               urukul.SPIT_DDS_WR, self.chip_select)
         self.bus.write(addr << 24)
         self.bus.set_config_mu(urukul.SPI_CONFIG, 32,
-                urukul.SPIT_DDS_WR, self.chip_select)
+                               urukul.SPIT_DDS_WR, self.chip_select)
         self.bus.write(data_high)
         self.bus.set_config_mu(urukul.SPI_CONFIG | spi.SPI_END, 32,
-                urukul.SPIT_DDS_WR, self.chip_select)
+                               urukul.SPIT_DDS_WR, self.chip_select)
         self.bus.write(data_low)
 
     @kernel
@@ -136,7 +209,7 @@ class AD9910:
         """
         # Set SPI mode
         self.write32(_AD9910_REG_CFR1, 0x00000002)
-        self.cpld.io_update.pulse(2*us)
+        self.cpld.io_update.pulse(1*us)
         delay(1*ms)
         if not blind:
             # Use the AUX DAC setting to identify and confirm presence
@@ -145,60 +218,121 @@ class AD9910:
                 raise ValueError("Urukul AD9910 AUX_DAC mismatch")
             delay(50*us)  # slack
         # Configure PLL settings and bring up PLL
-        self.write32(_AD9910_REG_CFR2, 0x01400020)
-        self.cpld.io_update.pulse(2*us)
+        # enable amplitude scale from profiles
+        # read effective FTW
+        # sync timing validation disable (enabled later)
+        self.write32(_AD9910_REG_CFR2, 0x01010020)
+        self.cpld.io_update.pulse(1*us)
         cfr3 = (0x0807c100 | (self.pll_vco << 24) |
                 (self.pll_cp << 19) | (self.pll_n << 1))
         self.write32(_AD9910_REG_CFR3, cfr3 | 0x400)  # PFD reset
-        self.cpld.io_update.pulse(100*us)
+        self.cpld.io_update.pulse(1*us)
         self.write32(_AD9910_REG_CFR3, cfr3)
-        self.cpld.io_update.pulse(100*us)
+        self.cpld.io_update.pulse(1*us)
         if blind:
             delay(100*ms)
-            return
-        # Wait for PLL lock, up to 100 ms
-        for i in range(100):
-            sta = self.cpld.sta_read()
-            lock = urukul_sta_pll_lock(sta)
-            delay(1*ms)
-            if lock & (1 << self.chip_select - 4):
-                return
-        raise ValueError("PLL lock timeout")
+        else:
+            # Wait for PLL lock, up to 100 ms
+            for i in range(100):
+                sta = self.cpld.sta_read()
+                lock = urukul_sta_pll_lock(sta)
+                delay(1*ms)
+                if lock & (1 << self.chip_select - 4):
+                    break
+                if i >= 100 - 1:
+                    raise ValueError("PLL lock timeout")
+        if self.sync_delay_seed >= 0:
+            self.tune_sync_delay(self.sync_delay_seed)
+        delay(1*ms)
 
     @kernel
-    def set_mu(self, ftw, pow=0, asf=0x3fff):
+    def power_down(self, bits=0b1111):
+        """Power down DDS.
+
+        :param bits: power down bits, see datasheet
+        """
+        self.write32(_AD9910_REG_CFR1, 0x00000002 | (bits << 4))
+        self.cpld.io_update.pulse(1*us)
+
+    @kernel
+    def set_mu(self, ftw, pow=0, asf=0x3fff, phase_mode=_PHASE_MODE_DEFAULT,
+               ref_time=-1):
         """Set profile 0 data in machine units.
 
+        This uses machine units (FTW, POW, ASF). The frequency tuning word
+        width is 32, the phase offset word width is 16, and the amplitude
+        scale factor width is 12.
+
         After the SPI transfer, the shared IO update pin is pulsed to
         activate the data.
 
+        .. seealso: :meth:`set_phase_mode` for a definition of the different
+            phase modes.
+
         :param ftw: Frequency tuning word: 32 bit.
         :param pow: Phase tuning word: 16 bit unsigned.
         :param asf: Amplitude scale factor: 14 bit unsigned.
+        :param phase_mode: If specified, overrides the default phase mode set
+            by :meth:`set_phase_mode` for this call.
+        :param ref_time: Fiducial time used to compute absolute or tracking
+            phase updates. In machine units as obtained by `now_mu()`.
+        :return: Resulting phase offset word after application of phase
+            tracking offset. When using :const:`PHASE_MODE_CONTINUOUS` in
+            subsequent calls, use this value as the "current" phase.
         """
+        if phase_mode == _PHASE_MODE_DEFAULT:
+            phase_mode = self.phase_mode
+        # Align to coarse RTIO which aligns SYNC_CLK
+        at_mu(now_mu() & ~0xf)
+        if phase_mode != PHASE_MODE_CONTINUOUS:
+            # Auto-clear phase accumulator on IO_UPDATE.
+            # This is active already for the next IO_UPDATE
+            self.write32(_AD9910_REG_CFR1, 0x00002002)
+            if phase_mode == PHASE_MODE_TRACKING and ref_time < 0:
+                # set default fiducial time stamp
+                ref_time = 0
+            if ref_time >= 0:
+                # 32 LSB are sufficient.
+                # Also no need to use IO_UPDATE time as this
+                # is equivalent to an output pipeline latency.
+                dt = int32(now_mu()) - int32(ref_time)
+                pow += dt*ftw*self.sysclk_per_mu >> 16
         self.write64(_AD9910_REG_PR0, (asf << 16) | pow, ftw)
-        self.cpld.io_update.pulse(10*ns)
+        delay_mu(int64(self.io_update_delay))
+        self.cpld.io_update.pulse_mu(8)  # assumes 8 mu > t_SYSCLK
+        at_mu(now_mu() & ~0xf)
+        if phase_mode != PHASE_MODE_CONTINUOUS:
+            self.write32(_AD9910_REG_CFR1, 0x00000002)
+            # future IO_UPDATE will activate
+        return pow
 
     @portable(flags={"fast-math"})
     def frequency_to_ftw(self, frequency):
-        """Returns the frequency tuning word corresponding to the given
+        """Return the frequency tuning word corresponding to the given
         frequency.
         """
         return int32(round(self.ftw_per_hz*frequency))
 
     @portable(flags={"fast-math"})
     def turns_to_pow(self, turns):
-        """Returns the phase offset word corresponding to the given phase
+        """Return the phase offset word corresponding to the given phase
         in turns."""
         return int32(round(turns*0x10000))
 
     @portable(flags={"fast-math"})
     def amplitude_to_asf(self, amplitude):
-        """Returns amplitude scale factor corresponding to given amplitude."""
+        """Return amplitude scale factor corresponding to given amplitude."""
         return int32(round(amplitude*0x3ffe))
 
+    @portable(flags={"fast-math"})
+    def pow_to_turns(self, pow):
+        """Return the phase in turns corresponding to a given phase offset
+        word."""
+        return pow/0x10000
+
     @kernel
-    def set(self, frequency, phase=0.0, amplitude=1.0):
+    def set(self, frequency, phase=0.0, amplitude=1.0,
+            phase_mode=_PHASE_MODE_DEFAULT, ref_time=-1):
         """Set profile 0 data in SI units.
 
         .. seealso:: :meth:`set_mu`
@@ -206,10 +340,13 @@ class AD9910:
         :param ftw: Frequency in Hz
         :param pow: Phase tuning word in turns
         :param asf: Amplitude in units of full scale
+        :param phase_mode: Phase mode constant
+        :param ref_time: Fiducial time stamp in machine units
+        :return: Resulting phase offset in turns
         """
-        self.set_mu(self.frequency_to_ftw(frequency),
-                    self.turns_to_pow(phase),
-                    self.amplitude_to_asf(amplitude))
+        return self.pow_to_turns(self.set_mu(
+            self.frequency_to_ftw(frequency), self.turns_to_pow(phase),
+            self.amplitude_to_asf(amplitude), phase_mode, ref_time))
 
     @kernel
     def set_att_mu(self, att):
@@ -240,3 +377,148 @@ class AD9910:
         :param state: CPLD CFG RF switch bit
         """
         self.cpld.cfg_sw(self.chip_select - 4, state)
+
+    @kernel
+    def set_sync(self, in_delay, window):
+        """Set the relevant parameters in the multi device synchronization
+        register. See the AD9910 datasheet for details. The SYNC clock
+        generator preset value is set to zero, and the SYNC_OUT generator is
+        disabled.
+
+        :param in_delay: SYNC_IN delay tap (0-31) in steps of ~75ps
+        :param window: Symmetric SYNC_IN validation window (0-15) in
+            steps of ~75ps for both hold and setup margin.
+        """
+        self.write32(_AD9910_REG_MSYNC,
+                     (window << 28) |  # SYNC S/H validation delay
+                     (1 << 27) |  # SYNC receiver enable
+                     (0 << 26) |  # SYNC generator disable
+                     (0 << 25) |  # SYNC generator SYS rising edge
+                     (0 << 18) |  # SYNC preset
+                     (0 << 11) |  # SYNC output delay
+                     (in_delay << 3))  # SYNC receiver delay
+
+    @kernel
+    def clear_smp_err(self):
+        """Clear the SMP_ERR flag and enables SMP_ERR validity monitoring.
+
+        Violations of the SYNC_IN sample and hold margins will result in
+        SMP_ERR being asserted. This then also activates the red LED on
+        the respective Urukul channel.
+
+        Also modifies CFR2.
+        """
+        self.write32(_AD9910_REG_CFR2, 0x01010020)  # clear SMP_ERR
+        self.cpld.io_update.pulse(1*us)
+        self.write32(_AD9910_REG_CFR2, 0x01010000)  # enable SMP_ERR
+        self.cpld.io_update.pulse(1*us)
+
+    @kernel
+    def tune_sync_delay(self, sync_delay_seed):
+        """Find a stable SYNC_IN delay.
+
+        This method first locates the smallest SYNC_IN validity window at
+        minimum window size and then increases the window a bit to provide some
+        slack and stability.
+
+        It starts scanning delays around `sync_delay_seed` (see the
+        device database arguments and :class:`AD9910`) at maximum validation
+        window size and decreases the window size until a valid delay is found.
+
+        :param sync_delay_seed: Start value for valid SYNC_IN delay search.
+        :return: Tuple of optimal delay and window size.
+        """
+        dt = 14  # 1/(f_SYSCLK*75ps) taps per SYSCLK period
+        max_delay = dt  # 14*75ps > 1ns
+        max_window = dt//4 + 1  # 2*75ps*4 = 600ps high > 1ns/2
+        min_window = max(0, max_window - 2)  # 2*75ps hold, 2*75ps setup
+        for window in range(max_window - min_window + 1):
+            window = max_window - window
+            for in_delay in range(max_delay):
+                # alternate search direction around seed_delay
+                if in_delay & 1:
+                    in_delay = -in_delay
+                in_delay = sync_delay_seed + (in_delay >> 1)
+                if in_delay < 0:
+                    in_delay = 0
+                elif in_delay > 31:
+                    in_delay = 31
+                self.set_sync(in_delay, window)
+                self.clear_smp_err()
+                # integrate SMP_ERR statistics for a few hundred cycles
+                delay(10*us)
+                err = urukul_sta_smp_err(self.cpld.sta_read())
+                err = (err >> (self.chip_select - 4)) & 1
+                delay(40*us)  # slack
+                if not err:
+                    window -= min_window  # add margin
+                    self.set_sync(in_delay, window)
+                    self.clear_smp_err()
+                    delay(40*us)  # slack
+                    return in_delay, window
+        raise ValueError("no valid window/delay")
+
+    @kernel
+    def measure_io_update_alignment(self, io_up_delay):
+        """Use the digital ramp generator to locate the alignment between
+        IO_UPDATE and SYNC_CLK.
+
+        The ramp generator is set up to a linear frequency ramp
+        (dFTW/t_SYNC_CLK=1) and started at a RTIO time stamp.
+
+        After scanning the alignment, an IO_UPDATE delay midway between two
+        edges should be chosen.
+
+        :return: odd/even SYNC_CLK cycle indicator
+        """
+        # set up DRG
+        # DRG ACC autoclear and LRR on io update
+        self.write32(_AD9910_REG_CFR1, 0x0000c002)
+        # DRG -> FTW, DRG enable
+        self.write32(_AD9910_REG_CFR2, 0x01090000)
+        # no limits
+        self.write64(_AD9910_REG_DRAMPL, -1, 0)
+        # DRCTL=0, dt=1 t_SYNC_CLK
+        self.write32(_AD9910_REG_DRAMPR, 0x00010000)
+        # dFTW = 1, (work around negative slope)
+        self.write64(_AD9910_REG_DRAMPS, -1, 0)
+        at_mu(now_mu() + 0x10 & ~0xf)  # align to RTIO/2
+        self.cpld.io_update.pulse_mu(8)
+        # disable DRG autoclear and LRR on io_update
+        self.write32(_AD9910_REG_CFR1, 0x00000002)
+        # stop DRG
+        self.write64(_AD9910_REG_DRAMPS, 0, 0)
+        at_mu((now_mu() + 0x10 & ~0xf) + io_up_delay)  # delay
+        self.cpld.io_update.pulse_mu(32 - io_up_delay)  # realign
+        ftw = self.read32(_AD9910_REG_FTW)  # read out effective FTW
+        delay(100*us)  # slack
+        # disable DRG
+        self.write32(_AD9910_REG_CFR2, 0x01010000)
+        self.cpld.io_update.pulse_mu(8)
+        return ftw & 1
+
+    @kernel
+    def tune_io_update_delay(self):
+        """Find a stable IO_UPDATE delay alignment.
+
+        Scan through increasing IO_UPDATE delays until a delay is found that
+        lets IO_UPDATE be registered in the next SYNC_CLK cycle. Return a
+        IO_UPDATE delay that is midway between two such SYNC_CLK transitions.
+
+        This method assumes that the IO_UPDATE TTLOut device has one machine
+        unit resolution (SERDES) and that the ratio between fine RTIO frequency
+        (RTIO time machine units) and SYNC_CLK is 4.
+
+        :return: Stable IO_UPDATE delay to be passed to the constructor
+            :class:`AD9910` via the device database.
+        """
+        period = 4  # f_RTIO/f_SYNC = 4
+        max_delay = 8  # mu, 1 ns
+        d0 = self.io_update_delay
+        t0 = int32(self.measure_io_update_alignment(d0))
+        for i in range(max_delay - 1):
+            t = self.measure_io_update_alignment(
+                (d0 + i + 1) & (max_delay - 1))
+            if t != t0:
+                return (d0 + i + period//2) & (period - 1)
+        raise ValueError("no IO_UPDATE-SYNC_CLK alignment edge found")
diff --git a/artiq/coredevice/ttl.py b/artiq/coredevice/ttl.py
index 2548bab70..f7bf48bb9 100644
--- a/artiq/coredevice/ttl.py
+++ b/artiq/coredevice/ttl.py
@@ -430,14 +430,15 @@ class TTLClockGen:
     The time cursor is not modified by any function in this class.
 
     :param channel: channel number
+    :param acc_width: accumulator width in bits
     """
     kernel_invariants = {"core", "channel", "acc_width"}
 
-    def __init__(self, dmgr, channel, core_device="core"):
+    def __init__(self, dmgr, channel, acc_width=24, core_device="core"):
         self.core = dmgr.get(core_device)
         self.channel = channel
 
-        self.acc_width = numpy.int64(24)
+        self.acc_width = numpy.int64(acc_width)
 
     @portable
     def frequency_to_ftw(self, frequency):
diff --git a/artiq/coredevice/urukul.py b/artiq/coredevice/urukul.py
index fe9c1a68c..9f96a92ed 100644
--- a/artiq/coredevice/urukul.py
+++ b/artiq/coredevice/urukul.py
@@ -1,4 +1,4 @@
-from artiq.language.core import kernel, delay, portable
+from artiq.language.core import kernel, delay, portable, at_mu, now_mu
 from artiq.language.units import us, ms
 
 from numpy import int32
@@ -61,7 +61,7 @@ def urukul_cfg(rf_sw, led, profile, io_update, mask_nu,
             (io_update << CFG_IO_UPDATE) |
             (mask_nu << CFG_MASK_NU) |
             ((clk_sel & 0x01) << CFG_CLK_SEL0) |
-            ((clk_sel & 0x02) << (CFG_CLK_SEL1-1)) |
+            ((clk_sel & 0x02) << (CFG_CLK_SEL1 - 1)) |
             (sync_sel << CFG_SYNC_SEL) |
             (rst << CFG_RST) |
             (io_rst << CFG_IO_RST))
@@ -109,12 +109,22 @@ class _RegIOUpdate:
         self.cpld.cfg_write(cfg)
 
 
+class _DummySync:
+    def __init__(self, cpld):
+        self.cpld = cpld
+
+    @kernel
+    def set_mu(self, ftw):
+        pass
+
+
 class CPLD:
     """Urukul CPLD SPI router and configuration interface.
 
     :param spi_device: SPI bus device name
     :param io_update_device: IO update RTIO TTLOut channel name
     :param dds_reset_device: DDS reset RTIO TTLOut channel name
+    :param sync_device: AD9910 SYNC_IN RTIO TTLClockGen channel name
     :param refclk: Reference clock (SMA, MMCX or on-board 100 MHz oscillator)
         frequency in Hz
     :param clk_sel: Reference clock selection. For hardware revision >= 1.3
@@ -122,20 +132,25 @@ class CPLD:
         internal MMCX. For hardware revision <= v1.2 valid options are: 0 -
         either XO or MMCX dependent on component population; 1 SMA. Unsupported
         clocking options are silently ignored.
-    :param sync_sel: SYNC clock selection. 0 corresponds to SYNC clock over EEM
-        from FPGA. 1 corresponds to SYNC clock from DDS0.
+    :param sync_sel: SYNC_IN selection. 0 corresponds to SYNC_IN over EEM
+        from FPGA. 1 corresponds to SYNC_IN from DDS0.
     :param rf_sw: Initial CPLD RF switch register setting (default: 0x0).
         Knowledge of this state is not transferred between experiments.
     :param att: Initial attenuator setting shift register (default:
         0x00000000). See also: :meth:`set_all_att_mu`. Knowledge of this state
         is not transferred between experiments.
+    :param sync_div: SYNC_IN generator divider. The ratio between the coarse
+        RTIO frequency and the SYNC_IN generator frequency (default: 2 if
+        `sync_device` was specified).
     :param core_device: Core device name
     """
     kernel_invariants = {"refclk", "bus", "core", "io_update"}
 
     def __init__(self, dmgr, spi_device, io_update_device=None,
-                 dds_reset_device=None, sync_sel=0, clk_sel=0, rf_sw=0,
-                 refclk=125e6, att=0x00000000, core_device="core"):
+                 dds_reset_device=None, sync_device=None,
+                 sync_sel=0, clk_sel=0, rf_sw=0,
+                 refclk=125e6, att=0x00000000, sync_div=None,
+                 core_device="core"):
 
         self.core = dmgr.get(core_device)
         self.refclk = refclk
@@ -147,11 +162,20 @@ class CPLD:
             self.io_update = _RegIOUpdate(self)
         if dds_reset_device is not None:
             self.dds_reset = dmgr.get(dds_reset_device)
+        if sync_device is not None:
+            self.sync = dmgr.get(sync_device)
+            if sync_div is None:
+                sync_div = 2
+        else:
+            self.sync = _DummySync(self)
+            assert sync_div is None
+            sync_div = 0
 
         self.cfg_reg = urukul_cfg(rf_sw=rf_sw, led=0, profile=0,
                                   io_update=0, mask_nu=0, clk_sel=clk_sel,
                                   sync_sel=sync_sel, rst=0, io_rst=0)
         self.att_reg = att
+        self.sync_div = sync_div
 
     @kernel
     def cfg_write(self, cfg):
@@ -207,6 +231,9 @@ class CPLD:
                 raise ValueError("Urukul proto_rev mismatch")
         delay(100*us)  # reset, slack
         self.cfg_write(cfg)
+        if self.sync_div:
+            at_mu(now_mu() & ~0xf)  # align to RTIO/2
+            self.set_sync_div(self.sync_div)  # 125 MHz/2 = 1 GHz/16
         delay(1*ms)  # DDS wake up
 
     @kernel
@@ -289,3 +316,20 @@ class CPLD:
                                SPIT_ATT_RD, CS_ATT)
         self.bus.write(self.att_reg)
         return self.bus.read()
+
+    @kernel
+    def set_sync_div(self, div):
+        """Set the SYNC_IN AD9910 pulse generator frequency
+        and align it to the current RTIO timestamp.
+
+        The SYNC_IN signal is derived from the coarse RTIO clock
+        and the divider must be a power of two two.
+        Configure ``sync_sel == 0``.
+
+        :param div: SYNC_IN frequency divider. Must be a power of two.
+            Minimum division ratio is 2. Maximum division ratio is 16.
+        """
+        ftw_max = 1 << 4
+        ftw = ftw_max//div
+        assert ftw*div == ftw_max
+        self.sync.set_mu(ftw)
diff --git a/artiq/examples/kasli_basic/device_db_hub.py b/artiq/examples/kasli_basic/device_db_hub.py
index dc2fc2d59..ea364ecfc 100644
--- a/artiq/examples/kasli_basic/device_db_hub.py
+++ b/artiq/examples/kasli_basic/device_db_hub.py
@@ -86,13 +86,19 @@ for j in range(3):
             "type": "local",
             "module": "artiq.coredevice.spi2",
             "class": "SPIMaster",
-            "arguments": {"channel": 27 + 2*j}
+            "arguments": {"channel": 27 + 3*j}
+        },
+        "ttl_urukul{}_sync".format(j): {
+            "type": "local",
+            "module": "artiq.coredevice.ttl",
+            "class": "TTLClockGen",
+            "arguments": {"channel": 28 + 3*j, "acc_width": 4}
         },
         "ttl_urukul{}_io_update".format(j): {
             "type": "local",
             "module": "artiq.coredevice.ttl",
             "class": "TTLOut",
-            "arguments": {"channel": 28 + 2*j}
+            "arguments": {"channel": 29 + 3*j}
         },
         "urukul{}_cpld".format(j): {
             "type": "local",
@@ -100,6 +106,7 @@ for j in range(3):
             "class": "CPLD",
             "arguments": {
                 "spi_device": "spi_urukul{}".format(j),
+                "sync_device": "ttl_urukul{}_sync".format(j),
                 "io_update_device": "ttl_urukul{}_io_update".format(j),
                 "refclk": 100e6,
                 "clk_sel": 0
@@ -126,13 +133,13 @@ device_db.update({
         "type": "local",
         "module": "artiq.coredevice.ttl",
         "class": "TTLOut",
-        "arguments": {"channel": 33}
+        "arguments": {"channel": 36}
     },
     "led1": {
         "type": "local",
         "module": "artiq.coredevice.ttl",
         "class": "TTLOut",
-        "arguments": {"channel": 34}
+        "arguments": {"channel": 37}
     }
 })
 
@@ -142,19 +149,19 @@ device_db.update({
         "type": "local",
         "module": "artiq.coredevice.spi2",
         "class": "SPIMaster",
-        "arguments": {"channel": 35}
+        "arguments": {"channel": 38}
     },
     "ttl_zotino0_ldac": {
         "type": "local",
         "module": "artiq.coredevice.ttl",
         "class": "TTLOut",
-        "arguments": {"channel": 36}
+        "arguments": {"channel": 39}
     },
     "ttl_zotino0_clr": {
         "type": "local",
         "module": "artiq.coredevice.ttl",
         "class": "TTLOut",
-        "arguments": {"channel": 37}
+        "arguments": {"channel": 40}
     },
     "zotino0": {
         "type": "local",
diff --git a/artiq/examples/kasli_basic/device_db_luh.py b/artiq/examples/kasli_basic/device_db_luh.py
new file mode 100644
index 000000000..9efdbd190
--- /dev/null
+++ b/artiq/examples/kasli_basic/device_db_luh.py
@@ -0,0 +1,183 @@
+core_addr = "staging.ber.quartiq.de"
+
+device_db = {
+    "core": {
+        "type": "local",
+        "module": "artiq.coredevice.core",
+        "class": "Core",
+        "arguments": {"host": core_addr, "ref_period": 1e-9}
+    },
+    "core_log": {
+        "type": "controller",
+        "host": "::1",
+        "port": 1068,
+        "command": "aqctl_corelog -p {port} --bind {bind} " + core_addr
+    },
+    "core_cache": {
+        "type": "local",
+        "module": "artiq.coredevice.cache",
+        "class": "CoreCache"
+    },
+    "core_dma": {
+        "type": "local",
+        "module": "artiq.coredevice.dma",
+        "class": "CoreDMA"
+    },
+
+    "i2c_switch0": {
+        "type": "local",
+        "module": "artiq.coredevice.i2c",
+        "class": "PCA9548",
+        "arguments": {"address": 0xe0}
+    },
+    "i2c_switch1": {
+        "type": "local",
+        "module": "artiq.coredevice.i2c",
+        "class": "PCA9548",
+        "arguments": {"address": 0xe2}
+    },
+}
+
+
+device_db.update({
+    "ttl" + str(i): {
+        "type": "local",
+        "module": "artiq.coredevice.ttl",
+        "class": "TTLInOut" if i < 4 else "TTLOut",
+        "arguments": {"channel": i},
+    } for i in range(24)
+})
+
+
+device_db.update({
+    "spi_sampler0_adc": {
+        "type": "local",
+        "module": "artiq.coredevice.spi2",
+        "class": "SPIMaster",
+        "arguments": {"channel": 24}
+    },
+    "spi_sampler0_pgia": {
+        "type": "local",
+        "module": "artiq.coredevice.spi2",
+        "class": "SPIMaster",
+        "arguments": {"channel": 25}
+    },
+    "spi_sampler0_cnv": {
+        "type": "local",
+        "module": "artiq.coredevice.ttl",
+        "class": "TTLOut",
+        "arguments": {"channel": 26},
+    },
+    "sampler0": {
+        "type": "local",
+        "module": "artiq.coredevice.sampler",
+        "class": "Sampler",
+        "arguments": {
+            "spi_adc_device": "spi_sampler0_adc",
+            "spi_pgia_device": "spi_sampler0_pgia",
+            "cnv_device": "spi_sampler0_cnv"
+        }
+    }
+})
+
+for j in range(2):
+    device_db.update({
+        "spi_urukul{}".format(j): {
+            "type": "local",
+            "module": "artiq.coredevice.spi2",
+            "class": "SPIMaster",
+            "arguments": {"channel": 27 + 3*j}
+        },
+        "ttl_urukul{}_sync".format(j): {
+            "type": "local",
+            "module": "artiq.coredevice.ttl",
+            "class": "TTLClockGen",
+            "arguments": {"channel": 28 + 3*j, "acc_width": 4}
+        },
+        "ttl_urukul{}_io_update".format(j): {
+            "type": "local",
+            "module": "artiq.coredevice.ttl",
+            "class": "TTLOut",
+            "arguments": {"channel": 29 + 3*j}
+        },
+        "urukul{}_cpld".format(j): {
+            "type": "local",
+            "module": "artiq.coredevice.urukul",
+            "class": "CPLD",
+            "arguments": {
+                "spi_device": "spi_urukul{}".format(j),
+                "sync_device": "ttl_urukul{}_sync".format(j),
+                "io_update_device": "ttl_urukul{}_io_update".format(j),
+                "refclk": 100e6,
+                "clk_sel": 0
+            }
+        }
+    })
+
+    device_db.update({
+        "urukul{}_ch{}".format(j, i): {
+            "type": "local",
+            "module": "artiq.coredevice.ad9910",
+            "class": "AD9910",
+            "arguments": {
+                "pll_n": 40,
+                "chip_select": 4 + i,
+                "cpld_device": "urukul{}_cpld".format(j)
+            }
+        } for i in range(4)
+    })
+
+device_db.update({
+    "grabber0": {
+        "type": "local",
+        "module": "artiq.coredevice.grabber",
+        "class": "grabber",
+        "arguments": {"channel_base": 33}
+    }
+})
+
+device_db.update({
+    "led0": {
+        "type": "local",
+        "module": "artiq.coredevice.ttl",
+        "class": "TTLOut",
+        "arguments": {"channel": 35}
+    },
+    "led1": {
+        "type": "local",
+        "module": "artiq.coredevice.ttl",
+        "class": "TTLOut",
+        "arguments": {"channel": 36}
+    }
+})
+
+device_db.update({
+    "spi_zotino0": {
+        "type": "local",
+        "module": "artiq.coredevice.spi2",
+        "class": "SPIMaster",
+        "arguments": {"channel": 37}
+    },
+    "ttl_zotino0_ldac": {
+        "type": "local",
+        "module": "artiq.coredevice.ttl",
+        "class": "TTLOut",
+        "arguments": {"channel": 38}
+    },
+    "ttl_zotino0_clr": {
+        "type": "local",
+        "module": "artiq.coredevice.ttl",
+        "class": "TTLOut",
+        "arguments": {"channel": 39}
+    },
+    "zotino0": {
+        "type": "local",
+        "module": "artiq.coredevice.zotino",
+        "class": "Zotino",
+        "arguments": {
+            "spi_device": "spi_zotino0",
+            "ldac_device": "ttl_zotino0_ldac",
+            "clr_device": "ttl_zotino0_clr"
+        }
+    }
+})
diff --git a/artiq/examples/kasli_basic/device_db_opticlock.py b/artiq/examples/kasli_basic/device_db_opticlock.py
index 068aba9fa..70c4d22d6 100644
--- a/artiq/examples/kasli_basic/device_db_opticlock.py
+++ b/artiq/examples/kasli_basic/device_db_opticlock.py
@@ -198,7 +198,7 @@ device_db = {
         "class": "TTLOut",
         "arguments": {"channel": 25}
     },
-    "novogorny0" : {
+    "novogorny0": {
         "type": "local",
         "module": "artiq.coredevice.novogorny",
         "class": "Novogorny",
@@ -313,23 +313,94 @@ device_db = {
         "arguments": {"channel": 33}
     },
 
+    "spi_urukul1": {
+        "type": "local",
+        "module": "artiq.coredevice.spi2",
+        "class": "SPIMaster",
+        "arguments": {"channel": 34}
+    },
+    "ttl_urukul1_sync": {
+        "type": "local",
+        "module": "artiq.coredevice.ttl",
+        "class": "TTLClockGen",
+        "arguments": {"channel": 35, "acc_width": 4}
+    },
+    "ttl_urukul1_io_update": {
+        "type": "local",
+        "module": "artiq.coredevice.ttl",
+        "class": "TTLOut",
+        "arguments": {"channel": 36}
+    },
+    "urukul1_cpld": {
+        "type": "local",
+        "module": "artiq.coredevice.urukul",
+        "class": "CPLD",
+        "arguments": {
+            "spi_device": "spi_urukul1",
+            "sync_device": "ttl_urukul1_sync",
+            "io_update_device": "ttl_urukul1_io_update",
+            "refclk": 100e6,
+            "clk_sel": 1
+        }
+    },
+    "urukul1_ch0": {
+        "type": "local",
+        "module": "artiq.coredevice.ad9910",
+        "class": "AD9910",
+        "arguments": {
+            "pll_n": 40,
+            "chip_select": 4,
+            "cpld_device": "urukul1_cpld"
+        }
+    },
+    "urukul1_ch1": {
+        "type": "local",
+        "module": "artiq.coredevice.ad9910",
+        "class": "AD9910",
+        "arguments": {
+            "pll_n": 40,
+            "chip_select": 5,
+            "cpld_device": "urukul1_cpld"
+        }
+    },
+    "urukul1_ch2": {
+        "type": "local",
+        "module": "artiq.coredevice.ad9910",
+        "class": "AD9910",
+        "arguments": {
+            "pll_n": 40,
+            "chip_select": 6,
+            "cpld_device": "urukul1_cpld"
+        }
+    },
+    "urukul1_ch3": {
+        "type": "local",
+        "module": "artiq.coredevice.ad9910",
+        "class": "AD9910",
+        "arguments": {
+            "pll_n": 40,
+            "chip_select": 7,
+            "cpld_device": "urukul1_cpld"
+        }
+    },
+
     "spi_zotino0": {
         "type": "local",
         "module": "artiq.coredevice.spi2",
         "class": "SPIMaster",
-        "arguments": {"channel": 36}
+        "arguments": {"channel": 37}
     },
     "ttl_zotino0_ldac": {
         "type": "local",
         "module": "artiq.coredevice.ttl",
         "class": "TTLOut",
-        "arguments": {"channel": 37}
+        "arguments": {"channel": 38}
     },
     "ttl_zotino0_clr": {
         "type": "local",
         "module": "artiq.coredevice.ttl",
         "class": "TTLOut",
-        "arguments": {"channel": 38}
+        "arguments": {"channel": 39}
     },
     "zotino0": {
         "type": "local",
diff --git a/artiq/examples/kasli_basic/device_db_ptb2.py b/artiq/examples/kasli_basic/device_db_ptb2.py
new file mode 100644
index 000000000..ed6d4722f
--- /dev/null
+++ b/artiq/examples/kasli_basic/device_db_ptb2.py
@@ -0,0 +1,240 @@
+core_addr = "staging.ber.quartiq.de"
+
+device_db = {
+    "core": {
+        "type": "local",
+        "module": "artiq.coredevice.core",
+        "class": "Core",
+        "arguments": {"host": core_addr, "ref_period": 1e-9}
+    },
+    "core_log": {
+        "type": "controller",
+        "host": "::1",
+        "port": 1068,
+        "command": "aqctl_corelog -p {port} --bind {bind} " + core_addr
+    },
+    "core_cache": {
+        "type": "local",
+        "module": "artiq.coredevice.cache",
+        "class": "CoreCache"
+    },
+    "core_dma": {
+        "type": "local",
+        "module": "artiq.coredevice.dma",
+        "class": "CoreDMA"
+    },
+
+    "i2c_switch0": {
+        "type": "local",
+        "module": "artiq.coredevice.i2c",
+        "class": "PCA9548",
+        "arguments": {"address": 0xe0}
+    },
+    "i2c_switch1": {
+        "type": "local",
+        "module": "artiq.coredevice.i2c",
+        "class": "PCA9548",
+        "arguments": {"address": 0xe2}
+    },
+}
+
+
+device_db.update({
+    "ttl" + str(i): {
+        "type": "local",
+        "module": "artiq.coredevice.ttl",
+        "class": "TTLInOut" if i < 4 else "TTLOut",
+        "arguments": {"channel": i},
+    } for i in range(24)
+})
+
+
+device_db.update({
+    "spi_sampler0_adc": {
+        "type": "local",
+        "module": "artiq.coredevice.spi2",
+        "class": "SPIMaster",
+        "arguments": {"channel": 24}
+    },
+    "spi_sampler0_pgia": {
+        "type": "local",
+        "module": "artiq.coredevice.spi2",
+        "class": "SPIMaster",
+        "arguments": {"channel": 25}
+    },
+    "spi_sampler0_cnv": {
+        "type": "local",
+        "module": "artiq.coredevice.ttl",
+        "class": "TTLOut",
+        "arguments": {"channel": 26},
+    },
+    "sampler0": {
+        "type": "local",
+        "module": "artiq.coredevice.sampler",
+        "class": "Sampler",
+        "arguments": {
+            "spi_adc_device": "spi_sampler0_adc",
+            "spi_pgia_device": "spi_sampler0_pgia",
+            "cnv_device": "spi_sampler0_cnv"
+        }
+    }
+})
+
+device_db.update({
+    "spi_urukul0": {
+        "type": "local",
+        "module": "artiq.coredevice.spi2",
+        "class": "SPIMaster",
+        "arguments": {"channel": 27}
+    },
+    "ttl_urukul0_io_update": {
+        "type": "local",
+        "module": "artiq.coredevice.ttl",
+        "class": "TTLOut",
+        "arguments": {"channel": 28}
+    },
+    "ttl_urukul0_sw0": {
+        "type": "local",
+        "module": "artiq.coredevice.ttl",
+        "class": "TTLOut",
+        "arguments": {"channel": 29}
+    },
+    "ttl_urukul0_sw1": {
+        "type": "local",
+        "module": "artiq.coredevice.ttl",
+        "class": "TTLOut",
+        "arguments": {"channel": 30}
+    },
+    "ttl_urukul0_sw2": {
+        "type": "local",
+        "module": "artiq.coredevice.ttl",
+        "class": "TTLOut",
+        "arguments": {"channel": 31}
+    },
+    "ttl_urukul0_sw3": {
+        "type": "local",
+        "module": "artiq.coredevice.ttl",
+        "class": "TTLOut",
+        "arguments": {"channel": 32}
+    },
+    "urukul0_cpld": {
+        "type": "local",
+        "module": "artiq.coredevice.urukul",
+        "class": "CPLD",
+        "arguments": {
+            "spi_device": "spi_urukul0",
+            "io_update_device": "ttl_urukul0_io_update",
+            "refclk": 100e6,
+            "clk_sel": 0
+        }
+    }
+})
+
+device_db.update({
+    "urukul0_ch" + str(i): {
+        "type": "local",
+        "module": "artiq.coredevice.ad9912",
+        "class": "AD9912",
+        "arguments": {
+            "pll_n": 10,
+            "chip_select": 4 + i,
+            "cpld_device": "urukul0_cpld",
+            "sw_device": "ttl_urukul0_sw" + str(i)
+        }
+    } for i in range(4)
+})
+
+
+device_db.update({
+    "spi_urukul1": {
+        "type": "local",
+        "module": "artiq.coredevice.spi2",
+        "class": "SPIMaster",
+        "arguments": {"channel": 33}
+    },
+    "ttl_urukul1_sync": {
+        "type": "local",
+        "module": "artiq.coredevice.ttl",
+        "class": "TTLClockGen",
+        "arguments": {"channel": 34, "acc_width": 4}
+    },
+    "ttl_urukul1_io_update": {
+        "type": "local",
+        "module": "artiq.coredevice.ttl",
+        "class": "TTLOut",
+        "arguments": {"channel": 35}
+    },
+    "urukul1_cpld": {
+        "type": "local",
+        "module": "artiq.coredevice.urukul",
+        "class": "CPLD",
+        "arguments": {
+            "spi_device": "spi_urukul1",
+            "sync_device": "ttl_urukul1_sync",
+            "io_update_device": "ttl_urukul1_io_update",
+            "refclk": 125e6,
+            "clk_sel": 0
+        }
+    }
+})
+
+device_db.update({
+    "urukul1_ch" + str(i): {
+        "type": "local",
+        "module": "artiq.coredevice.ad9910",
+        "class": "AD9910",
+        "arguments": {
+            "pll_n": 32,
+            "chip_select": 4 + i,
+            "cpld_device": "urukul1_cpld"
+        }
+    } for i in range(4)
+})
+
+
+device_db.update({
+    "led0": {
+        "type": "local",
+        "module": "artiq.coredevice.ttl",
+        "class": "TTLOut",
+        "arguments": {"channel": 36}
+    },
+    "led1": {
+        "type": "local",
+        "module": "artiq.coredevice.ttl",
+        "class": "TTLOut",
+        "arguments": {"channel": 37}
+    }
+})
+
+
+device_db.update({
+    "spi_zotino0": {
+        "type": "local",
+        "module": "artiq.coredevice.spi2",
+        "class": "SPIMaster",
+        "arguments": {"channel": 38}
+    },
+    "ttl_zotino0_ldac": {
+        "type": "local",
+        "module": "artiq.coredevice.ttl",
+        "class": "TTLOut",
+        "arguments": {"channel": 39}
+    },
+    "ttl_zotino0_clr": {
+        "type": "local",
+        "module": "artiq.coredevice.ttl",
+        "class": "TTLOut",
+        "arguments": {"channel": 40}
+    },
+    "zotino0": {
+        "type": "local",
+        "module": "artiq.coredevice.zotino",
+        "class": "Zotino",
+        "arguments": {
+            "spi_device": "spi_zotino0",
+            "ldac_device": "ttl_zotino0_ldac",
+            "clr_device": "ttl_zotino0_clr"
+        }
+    }
+})
diff --git a/artiq/examples/kasli_tester/device_db.py b/artiq/examples/kasli_tester/device_db.py
index e9779af5d..027fbe2e9 100644
--- a/artiq/examples/kasli_tester/device_db.py
+++ b/artiq/examples/kasli_tester/device_db.py
@@ -57,36 +57,42 @@ device_db.update(
         "class": "SPIMaster",
         "arguments": {"channel": 8}
     },
-    ttl_urukul0_io_update={
+    ttl_urukul0_sync={
         "type": "local",
         "module": "artiq.coredevice.ttl",
-        "class": "TTLOut",
-        "arguments": {"channel": 9}
+        "class": "TTLClockGen",
+        "arguments": {"channel": 9, "acc_width": 4}
     },
-    ttl_urukul0_sw0={
+    ttl_urukul0_io_update={
         "type": "local",
         "module": "artiq.coredevice.ttl",
         "class": "TTLOut",
         "arguments": {"channel": 10}
     },
-    ttl_urukul0_sw1={
+    ttl_urukul0_sw0={
         "type": "local",
         "module": "artiq.coredevice.ttl",
         "class": "TTLOut",
         "arguments": {"channel": 11}
     },
-    ttl_urukul0_sw2={
+    ttl_urukul0_sw1={
         "type": "local",
         "module": "artiq.coredevice.ttl",
         "class": "TTLOut",
         "arguments": {"channel": 12}
     },
-    ttl_urukul0_sw3={
+    ttl_urukul0_sw2={
         "type": "local",
         "module": "artiq.coredevice.ttl",
         "class": "TTLOut",
         "arguments": {"channel": 13}
     },
+    ttl_urukul0_sw3={
+        "type": "local",
+        "module": "artiq.coredevice.ttl",
+        "class": "TTLOut",
+        "arguments": {"channel": 14}
+    },
     urukul0_cpld={
         "type": "local",
         "module": "artiq.coredevice.urukul",
@@ -94,6 +100,7 @@ device_db.update(
         "arguments": {
             "spi_device": "spi_urukul0",
             "io_update_device": "ttl_urukul0_io_update",
+            "sync_device": "ttl_urukul0_sync",
             "refclk": 125e6,
             "clk_sel": 0
         }
@@ -119,19 +126,19 @@ device_db["spi_sampler0_adc"] = {
     "type": "local",
     "module": "artiq.coredevice.spi2",
     "class": "SPIMaster",
-    "arguments": {"channel": 14}
+    "arguments": {"channel": 15}
 }
 device_db["spi_sampler0_pgia"] = {
     "type": "local",
     "module": "artiq.coredevice.spi2",
     "class": "SPIMaster",
-    "arguments": {"channel": 15}
+    "arguments": {"channel": 16}
 }
 device_db["spi_sampler0_cnv"] = {
     "type": "local",
     "module": "artiq.coredevice.ttl",
     "class": "TTLOut",
-    "arguments": {"channel": 16},
+    "arguments": {"channel": 17},
 }
 device_db["sampler0"] = {
     "type": "local",
@@ -150,19 +157,19 @@ device_db["spi_zotino0"] = {
     "type": "local",
     "module": "artiq.coredevice.spi2",
     "class": "SPIMaster",
-    "arguments": {"channel": 17}
+    "arguments": {"channel": 18}
 }
 device_db["ttl_zotino0_ldac"] = {
     "type": "local",
     "module": "artiq.coredevice.ttl",
     "class": "TTLOut",
-    "arguments": {"channel": 18}
+    "arguments": {"channel": 19}
 }
 device_db["ttl_zotino0_clr"] = {
     "type": "local",
     "module": "artiq.coredevice.ttl",
     "class": "TTLOut",
-    "arguments": {"channel": 19}
+    "arguments": {"channel": 20}
 }
 device_db["zotino0"] = {
     "type": "local",
@@ -181,7 +188,7 @@ device_db["grabber0"] = {
     "type": "local",
     "module": "artiq.coredevice.grabber",
     "class": "Grabber",
-    "arguments": {"channel_base": 20}
+    "arguments": {"channel_base": 21}
 }
 
 
@@ -191,13 +198,19 @@ device_db.update(
         "type": "local",
         "module": "artiq.coredevice.spi2",
         "class": "SPIMaster",
-        "arguments": {"channel": 22}
+        "arguments": {"channel": 23}
+    },
+    ttl_urukul1_sync={
+        "type": "local",
+        "module": "artiq.coredevice.ttl",
+        "class": "TTLClockGen",
+        "arguments": {"channel": 24, "acc_width": 4}
     },
     ttl_urukul1_io_update={
         "type": "local",
         "module": "artiq.coredevice.ttl",
         "class": "TTLOut",
-        "arguments": {"channel": 23}
+        "arguments": {"channel": 25}
     },
     urukul1_cpld={
         "type": "local",
@@ -206,6 +219,7 @@ device_db.update(
         "arguments": {
             "spi_device": "spi_urukul1",
             "io_update_device": "ttl_urukul1_io_update",
+            "sync_device": "ttl_urukul1_sync",
             "refclk": 100e6,
             "clk_sel": 1
         }
@@ -231,7 +245,7 @@ for i in range(8):
         "type": "local",
         "module": "artiq.coredevice.ttl",
         "class": "TTLOut",
-        "arguments": {"channel": 24 + i},
+        "arguments": {"channel": 26 + i},
     }
 
 
@@ -241,7 +255,7 @@ for i in range(8):
         "type": "local",
         "module": "artiq.coredevice.ttl",
         "class": "TTLOut",
-        "arguments": {"channel": 32 + i},
+        "arguments": {"channel": 34 + i},
     }
 
 
@@ -250,19 +264,19 @@ device_db["spi_sampler1_adc"] = {
     "type": "local",
     "module": "artiq.coredevice.spi2",
     "class": "SPIMaster",
-    "arguments": {"channel": 40}
+    "arguments": {"channel": 42}
 }
 device_db["spi_sampler1_pgia"] = {
     "type": "local",
     "module": "artiq.coredevice.spi2",
     "class": "SPIMaster",
-    "arguments": {"channel": 41}
+    "arguments": {"channel": 43}
 }
 device_db["spi_sampler1_cnv"] = {
     "type": "local",
     "module": "artiq.coredevice.ttl",
     "class": "TTLOut",
-    "arguments": {"channel": 42},
+    "arguments": {"channel": 44},
 }
 device_db["sampler1"] = {
     "type": "local",
@@ -281,19 +295,19 @@ device_db["spi_zotino1"] = {
     "type": "local",
     "module": "artiq.coredevice.spi2",
     "class": "SPIMaster",
-    "arguments": {"channel": 43}
+    "arguments": {"channel": 45}
 }
 device_db["ttl_zotino1_ldac"] = {
     "type": "local",
     "module": "artiq.coredevice.ttl",
     "class": "TTLOut",
-    "arguments": {"channel": 44}
+    "arguments": {"channel": 46}
 }
 device_db["ttl_zotino1_clr"] = {
     "type": "local",
     "module": "artiq.coredevice.ttl",
     "class": "TTLOut",
-    "arguments": {"channel": 45}
+    "arguments": {"channel": 47}
 }
 device_db["zotino1"] = {
     "type": "local",
@@ -312,13 +326,13 @@ device_db.update(
         "type": "local",
         "module": "artiq.coredevice.ttl",
         "class": "TTLOut",
-        "arguments": {"channel": 46}
+        "arguments": {"channel": 48}
     },
     led1={
         "type": "local",
         "module": "artiq.coredevice.ttl",
         "class": "TTLOut",
-        "arguments": {"channel": 47}
+        "arguments": {"channel": 49}
     },
 )
 
@@ -328,4 +342,10 @@ device_db.update(
 
     loop_out="ttl4",
     loop_in="ttl0",
+
+    # Urukul CPLD with sync and io_update, IFC MODE 0b1000
+    urukul_cpld="urukul0_cpld",
+    # Urukul AD9910 with switch TTL, internal 125 MHz MMCX connection
+    # FIXME: MMCX not connected
+    # urukul_ad9910="urukul0_ch0",
 )
diff --git a/artiq/gateware/eem.py b/artiq/gateware/eem.py
index 34680a714..96050b748 100644
--- a/artiq/gateware/eem.py
+++ b/artiq/gateware/eem.py
@@ -75,7 +75,7 @@ class Urukul(_EEM):
             ),
         ]
         ttls = [(6, eem, "io_update"),
-                (7, eem, "dds_reset")]
+                (7, eem, "dds_reset_sync_in")]
         if eem_aux is not None:
             ttls += [(0, eem_aux, "sync_clk"),
                      (1, eem_aux, "sync_in"),
@@ -113,7 +113,7 @@ class Urukul(_EEM):
             ),
         ]
         ttls = [(6, eem0, "io_update"),
-                (7, eem0, "dds_reset"),
+                (7, eem0, "dds_reset_sync_in"),
                 (4, eem1, "sw0"),
                 (5, eem1, "sw1"),
                 (6, eem1, "sw2"),
@@ -148,7 +148,8 @@ class Urukul(_EEM):
         return ios
 
     @classmethod
-    def add_std(cls, target, eem, eem_aux, ttl_out_cls, iostandard="LVDS_25"):
+    def add_std(cls, target, eem, eem_aux, ttl_out_cls, sync_gen_cls=None,
+                iostandard="LVDS_25"):
         cls.add_extension(target, eem, eem_aux, iostandard=iostandard)
 
         phy = spi2.SPIMaster(target.platform.request("urukul{}_spi_p".format(eem)),
@@ -156,8 +157,13 @@ class Urukul(_EEM):
         target.submodules += phy
         target.rtio_channels.append(rtio.Channel.from_phy(phy, ififo_depth=4))
 
-        pads = target.platform.request("urukul{}_dds_reset".format(eem))
-        target.specials += DifferentialOutput(0, pads.p, pads.n)
+        pads = target.platform.request("urukul{}_dds_reset_sync_in".format(eem))
+        pad = Signal(reset=0)
+        target.specials += DifferentialOutput(pad, pads.p, pads.n)
+        if sync_gen_cls is not None:  # AD9910 variant and SYNC_IN from EEM
+            phy = sync_gen_cls(pad, ftw_width=4)
+            target.submodules += phy
+            target.rtio_channels.append(rtio.Channel.from_phy(phy))
 
         pads = target.platform.request("urukul{}_io_update".format(eem))
         phy = ttl_out_cls(pads.p, pads.n)
@@ -170,7 +176,6 @@ class Urukul(_EEM):
                 target.submodules += phy
                 target.rtio_channels.append(rtio.Channel.from_phy(phy))
 
-
 class Sampler(_EEM):
     @staticmethod
     def io(eem, eem_aux, iostandard="LVDS_25"):
@@ -522,7 +527,7 @@ class SUServo(_EEM):
         target.submodules += phy
         target.rtio_channels.append(rtio.Channel.from_phy(phy, ififo_depth=4))
 
-        pads = target.platform.request("{}_dds_reset".format(eem_urukul0))
+        pads = target.platform.request("{}_dds_reset_sync_in".format(eem_urukul0))
         target.specials += DifferentialOutput(0, pads.p, pads.n)
 
         for i, signal in enumerate("sw0 sw1 sw2 sw3".split()):
diff --git a/artiq/gateware/targets/kasli.py b/artiq/gateware/targets/kasli.py
index fc7eaf92a..0475fd9a6 100755
--- a/artiq/gateware/targets/kasli.py
+++ b/artiq/gateware/targets/kasli.py
@@ -175,7 +175,8 @@ class Opticlock(_StandaloneBase):
             self.submodules += phy
             self.rtio_channels.append(rtio.Channel.from_phy(phy))
 
-        eem.Urukul.add_std(self, 6, None, ttl_serdes_7series.Output_8X)
+        eem.Urukul.add_std(self, 6, None, ttl_serdes_7series.Output_8X,
+                           ttl_simple.ClockGen)
         eem.Zotino.add_std(self, 7, ttl_serdes_7series.Output_8X)
 
         self.config["HAS_RTIO_LOG"] = None
@@ -574,6 +575,47 @@ class PTB(_StandaloneBase):
         self.add_rtio(self.rtio_channels)
 
 
+class PTB2(_StandaloneBase):
+    """PTB Kasli variant with Urukul1 SYNC and external reference clock"""
+    def __init__(self, hw_rev=None, **kwargs):
+        if hw_rev is None:
+            hw_rev = "v1.1"
+        _StandaloneBase.__init__(self, hw_rev=hw_rev, **kwargs)
+
+        self.config["SI5324_AS_SYNTHESIZER"] = None
+        self.config["SI5324_EXT_REF"] = None
+        self.config["RTIO_FREQUENCY"] = "125.0"
+        if hw_rev == "v1.0":
+            # EEM clock fan-out from Si5324, not MMCX
+            self.comb += self.platform.request("clk_sel").eq(1)
+
+        self.rtio_channels = []
+        eem.DIO.add_std(self, 0,
+            ttl_serdes_7series.InOut_8X, ttl_serdes_7series.Output_8X)
+        eem.DIO.add_std(self, 1,
+            ttl_serdes_7series.Output_8X, ttl_serdes_7series.Output_8X)
+        eem.DIO.add_std(self, 2,
+            ttl_serdes_7series.Output_8X, ttl_serdes_7series.Output_8X)
+        eem.Sampler.add_std(self, 3, None, ttl_serdes_7series.Output_8X)
+        eem.Urukul.add_std(self, 5, 4, ttl_serdes_7series.Output_8X)
+        eem.Urukul.add_std(self, 6, None, ttl_serdes_7series.Output_8X,
+                           ttl_simple.ClockGen)
+
+        for i in (1, 2):
+            sfp_ctl = self.platform.request("sfp_ctl", i)
+            phy = ttl_simple.Output(sfp_ctl.led)
+            self.submodules += phy
+            self.rtio_channels.append(rtio.Channel.from_phy(phy))
+
+        eem.Zotino.add_std(self, 7, ttl_serdes_7series.Output_8X)
+
+        self.config["HAS_RTIO_LOG"] = None
+        self.config["RTIO_LOG_CHANNEL"] = len(self.rtio_channels)
+        self.rtio_channels.append(rtio.LogChannel())
+
+        self.add_rtio(self.rtio_channels)
+
+
 class HUB(_StandaloneBase):
     """HUB Kasli variant
 
@@ -600,9 +642,12 @@ class HUB(_StandaloneBase):
         eem.DIO.add_std(self, 2,
             ttl_serdes_7series.Output_8X, ttl_serdes_7series.Output_8X)
         eem.Sampler.add_std(self, 3, None, ttl_serdes_7series.Output_8X)
-        eem.Urukul.add_std(self, 4, None, ttl_serdes_7series.Output_8X)
-        eem.Urukul.add_std(self, 5, None, ttl_serdes_7series.Output_8X)
-        eem.Urukul.add_std(self, 6, None, ttl_serdes_7series.Output_8X)
+        eem.Urukul.add_std(self, 4, None, ttl_serdes_7series.Output_8X,
+                           ttl_simple.ClockGen)
+        eem.Urukul.add_std(self, 5, None, ttl_serdes_7series.Output_8X,
+                           ttl_simple.ClockGen)
+        eem.Urukul.add_std(self, 6, None, ttl_serdes_7series.Output_8X,
+                           ttl_simple.ClockGen)
 
         for i in (1, 2):
             sfp_ctl = self.platform.request("sfp_ctl", i)
@@ -645,8 +690,10 @@ class LUH(_StandaloneBase):
         eem.DIO.add_std(self, 2,
             ttl_serdes_7series.Output_8X, ttl_serdes_7series.Output_8X)
         eem.Sampler.add_std(self, 3, None, ttl_serdes_7series.Output_8X)
-        eem.Urukul.add_std(self, 4, None, ttl_serdes_7series.Output_8X)
-        eem.Urukul.add_std(self, 5, None, ttl_serdes_7series.Output_8X)
+        eem.Urukul.add_std(self, 4, None, ttl_serdes_7series.Output_8X,
+                           ttl_simple.ClockGen)
+        eem.Urukul.add_std(self, 5, None, ttl_serdes_7series.Output_8X,
+                           ttl_simple.ClockGen)
         eem.Grabber.add_std(self, 6)
 
         for i in (1, 2):
@@ -688,11 +735,13 @@ class Tester(_StandaloneBase):
         self.grabber_csr_group = []
         eem.DIO.add_std(self, 5,
             ttl_serdes_7series.InOut_8X, ttl_serdes_7series.Output_8X)
-        eem.Urukul.add_std(self, 0, 1, ttl_serdes_7series.Output_8X)
+        eem.Urukul.add_std(self, 0, 1, ttl_serdes_7series.Output_8X,
+                           ttl_simple.ClockGen)
         eem.Sampler.add_std(self, 3, 2, ttl_serdes_7series.Output_8X)
         eem.Zotino.add_std(self, 4, ttl_serdes_7series.Output_8X)
         eem.Grabber.add_std(self, 6)
-        eem.Urukul.add_std(self, 7, None, ttl_serdes_7series.Output_8X)
+        eem.Urukul.add_std(self, 7, None, ttl_serdes_7series.Output_8X,
+                           ttl_simple.ClockGen)
         eem.DIO.add_std(self, 8,
             ttl_serdes_7series.Output_8X, ttl_serdes_7series.Output_8X)
         eem.DIO.add_std(self, 9,
@@ -1079,8 +1128,8 @@ def main():
     soc_kasli_args(parser)
     parser.set_defaults(output_dir="artiq_kasli")
     variants = {cls.__name__.lower(): cls for cls in [
-        Opticlock, SUServo, SYSU, MITLL, MITLL2, USTC,
-        Tsinghua, Tsinghua2, WIPM, NUDT, PTB, HUB, LUH,
+        Opticlock, SUServo, PTB, PTB2, HUB, LUH,
+        SYSU, MITLL, MITLL2, USTC, Tsinghua, Tsinghua2, WIPM, NUDT,
         VLBAIMaster, VLBAISatellite, Tester, Master, Satellite]}
     parser.add_argument("-V", "--variant", default="opticlock",
                         help="variant: {} (default: %(default)s)".format(
diff --git a/artiq/test/coredevice/test_ad9910.py b/artiq/test/coredevice/test_ad9910.py
new file mode 100644
index 000000000..407755df6
--- /dev/null
+++ b/artiq/test/coredevice/test_ad9910.py
@@ -0,0 +1,180 @@
+from artiq.experiment import *
+from artiq.test.hardware_testbench import ExperimentCase
+from artiq.coredevice.ad9910 import _AD9910_REG_FTW
+from artiq.coredevice.urukul import (
+        urukul_sta_smp_err, CFG_CLK_SEL0, CFG_CLK_SEL1)
+
+
+class AD9910Exp(EnvExperiment):
+    def build(self, runner):
+        self.setattr_device("core")
+        self.dev = self.get_device("urukul_ad9910")
+        self.runner = runner
+
+    def run(self):
+        getattr(self, self.runner)()
+
+    @kernel
+    def instantiate(self):
+        pass
+
+    @kernel
+    def init(self):
+        self.core.break_realtime()
+        self.dev.cpld.init()
+        self.dev.init()
+
+    @kernel
+    def init_fail(self):
+        self.core.break_realtime()
+        self.dev.cpld.init()
+        cfg = self.dev.cpld.cfg_reg
+        cfg &= ~(1 << CFG_CLK_SEL1)
+        cfg |= 1 << CFG_CLK_SEL0
+        self.dev.cpld.cfg_write(cfg)
+        # clk_sel=1, external SMA, should fail PLL lock
+        self.dev.init()
+
+    @kernel
+    def set_get(self):
+        self.core.break_realtime()
+        self.dev.cpld.init()
+        self.dev.init()
+        self.dev.set_att(20*dB)
+        f = 81.2345*MHz
+        self.dev.set(frequency=f, phase=.33, amplitude=.89)
+        self.set_dataset("ftw_set", self.dev.frequency_to_ftw(f))
+        self.set_dataset("ftw_get", self.dev.read32(_AD9910_REG_FTW))
+
+    @kernel
+    def set_speed(self):
+        self.core.break_realtime()
+        self.dev.cpld.init()
+        self.dev.init()
+        f = 81.2345*MHz
+        n = 10
+        t0 = self.core.get_rtio_counter_mu()
+        for i in range(n):
+            self.dev.set(frequency=f, phase=.33, amplitude=.89)
+        self.set_dataset("dt", self.core.mu_to_seconds(
+            self.core.get_rtio_counter_mu() - t0)/n)
+
+    @kernel
+    def set_speed_mu(self):
+        self.core.break_realtime()
+        self.dev.cpld.init()
+        self.dev.init()
+        n = 10
+        t0 = self.core.get_rtio_counter_mu()
+        for i in range(n):
+            self.dev.set_mu(0x12345678, 0x1234, 0x4321)
+        self.set_dataset("dt", self.core.mu_to_seconds(
+            self.core.get_rtio_counter_mu() - t0)/n)
+
+    @kernel
+    def sync_window(self):
+        self.core.break_realtime()
+        self.dev.cpld.init()
+        self.dev.init()
+        dly, win = self.dev.tune_sync_delay(self.dev.sync_delay_seed)
+        err = [0] * 32
+        self.sync_scan(err, win=win + 1)  # tighten window by 2*75ps
+        self.set_dataset("dly", dly)
+        self.set_dataset("win", win)
+        self.set_dataset("err", err)
+
+    @kernel
+    def sync_scan(self, err, win):
+        for in_delay in range(len(err)):
+            self.dev.set_sync(in_delay=in_delay, window=win)
+            self.dev.clear_smp_err()
+            delay(10*us)  # integrate SMP_ERR statistics
+            e = urukul_sta_smp_err(self.dev.cpld.sta_read())
+            err[in_delay] = (e >> (self.dev.chip_select - 4)) & 1
+            delay(50*us)  # slack
+
+    @kernel
+    def io_update_delay(self):
+        self.core.break_realtime()
+        self.dev.cpld.init()
+        self.dev.init()
+        bins = [0]*8
+        self.scan_io_delay(bins)
+        self.set_dataset("bins", bins)
+        self.set_dataset("dly", self.dev.io_update_delay)
+
+    @kernel
+    def scan_io_delay(self, bins):
+        delay(100*us)
+        n = 100
+        for i in range(n):
+            for phase in range(len(bins)):
+                bins[phase] += self.dev.measure_io_update_alignment(phase)
+        delay(10*ms)
+
+    @kernel
+    def sw_readback(self):
+        self.core.break_realtime()
+        self.dev.cpld.init()
+        self.dev.init()
+        self.dev.cfg_sw(0)
+        self.dev.sw.on()
+        sw_on = (self.dev.cpld.sta_read() >> (self.dev.chip_select - 4)) & 1
+        delay(10*us)
+        self.dev.sw.off()
+        sw_off = (self.dev.cpld.sta_read() >> (self.dev.chip_select - 4)) & 1
+        self.set_dataset("sw", (sw_on, sw_off))
+
+
+class AD9910Test(ExperimentCase):
+    def test_instantiate(self):
+        self.execute(AD9910Exp, "instantiate")
+
+    def test_init(self):
+        self.execute(AD9910Exp, "init")
+
+    def test_init_fail(self):
+        with self.assertRaises(ValueError):
+            self.execute(AD9910Exp, "init_fail")
+
+    def test_set_get(self):
+        self.execute(AD9910Exp, "set_get")
+        ftw_get = self.dataset_mgr.get("ftw_get")
+        ftw_set = self.dataset_mgr.get("ftw_set")
+        self.assertEqual(ftw_get, ftw_set)
+
+    def test_set_speed(self):
+        self.execute(AD9910Exp, "set_speed")
+        dt = self.dataset_mgr.get("dt")
+        print(dt)
+        self.assertLess(dt, 70*us)
+
+    def test_set_speed_mu(self):
+        self.execute(AD9910Exp, "set_speed_mu")
+        dt = self.dataset_mgr.get("dt")
+        print(dt)
+        self.assertLess(dt, 10*us)
+
+    def test_sync_window(self):
+        self.execute(AD9910Exp, "sync_window")
+        err = self.dataset_mgr.get("err")
+        dly = self.dataset_mgr.get("dly")
+        win = self.dataset_mgr.get("win")
+        print(dly, win, err)
+        # make sure one tap margin on either side of optimal delay
+        for i in -1, 0, 1:
+            self.assertEqual(err[i + dly], 0)
+
+    def test_io_update_delay(self):
+        self.execute(AD9910Exp, "io_update_delay")
+        dly = self.dataset_mgr.get("dly")
+        bins = self.dataset_mgr.get("bins")
+        print(dly, bins)
+        n = max(bins)
+        # test for 4-periodicity (SYNC_CLK) and maximal contrast
+        for i in range(len(bins)):
+            self.assertEqual(abs(bins[i] - bins[(i + 4) % 8]), n)
+
+    def test_sw_readback(self):
+        self.execute(AD9910Exp, "sw_readback")
+        self.assertEqual(self.dataset_mgr.get("sw"), (1, 0))
diff --git a/artiq/test/coredevice/test_urukul.py b/artiq/test/coredevice/test_urukul.py
new file mode 100644
index 000000000..2141303c4
--- /dev/null
+++ b/artiq/test/coredevice/test_urukul.py
@@ -0,0 +1,149 @@
+from artiq.experiment import *
+from artiq.test.hardware_testbench import ExperimentCase
+from artiq.coredevice import urukul
+
+
+class UrukulExp(EnvExperiment):
+    def build(self, runner):
+        self.setattr_device("core")
+        self.dev = self.get_device("urukul_cpld")
+        self.runner = runner
+
+    def run(self):
+        getattr(self, self.runner)()
+
+    @kernel
+    def instantiate(self):
+        pass
+
+    @kernel
+    def init(self):
+        self.core.break_realtime()
+        self.dev.init()
+
+    @kernel
+    def cfg_write(self):
+        self.core.break_realtime()
+        self.dev.init()
+        self.dev.cfg_write(self.dev.cfg_reg)
+
+    @kernel
+    def sta_read(self):
+        self.core.break_realtime()
+        self.dev.init()
+        sta = self.dev.sta_read()
+        self.set_dataset("sta", sta)
+
+    @kernel
+    def switches(self):
+        self.core.break_realtime()
+        self.dev.init()
+        self.dev.io_rst()
+        self.dev.cfg_sw(0, 0)
+        self.dev.cfg_sw(0, 1)
+        self.dev.cfg_sw(3, 1)
+        self.dev.cfg_switches(0b1010)
+
+    @kernel
+    def switch_speed(self):
+        self.core.break_realtime()
+        self.dev.init()
+        n = 10
+        t0 = self.core.get_rtio_counter_mu()
+        for i in range(n):
+            self.dev.cfg_sw(3, i & 1)
+        self.set_dataset("dt", self.core.mu_to_seconds(
+            self.core.get_rtio_counter_mu() - t0)/n)
+
+    @kernel
+    def switches_readback(self):
+        self.core.reset()  # clear switch TTLs
+        self.dev.init()
+        sw_set = 0b1010
+        self.dev.cfg_switches(sw_set)
+        sta_get = self.dev.sta_read()
+        self.set_dataset("sw_set", sw_set)
+        self.set_dataset("sta_get", sta_get)
+
+    @kernel
+    def att(self):
+        self.core.break_realtime()
+        self.dev.init()
+        att_set = 0x12345678
+        self.dev.set_all_att_mu(att_set)
+        att_get = self.dev.get_att_mu()
+        self.set_dataset("att_set", att_set)
+        self.set_dataset("att_get", att_get)
+
+    @kernel
+    def att_speed(self):
+        self.core.break_realtime()
+        self.dev.init()
+        n = 10
+        t0 = self.core.get_rtio_counter_mu()
+        for i in range(n):
+            self.dev.set_att(3, 30*dB)
+        self.set_dataset("dt", self.core.mu_to_seconds(
+            self.core.get_rtio_counter_mu() - t0)/n)
+
+    @kernel
+    def io_update(self):
+        self.core.break_realtime()
+        self.dev.init()
+        self.dev.io_update.pulse_mu(8)
+
+    @kernel
+    def sync(self):
+        self.core.break_realtime()
+        self.dev.init()
+        self.dev.set_sync_div(2)
+
+
+class UrukulTest(ExperimentCase):
+    def test_instantiate(self):
+        self.execute(UrukulExp, "instantiate")
+
+    def test_init(self):
+        self.execute(UrukulExp, "init")
+
+    def test_cfg_write(self):
+        self.execute(UrukulExp, "cfg_write")
+
+    def test_sta_read(self):
+        self.execute(UrukulExp, "sta_read")
+        sta = self.dataset_mgr.get("sta")
+        print(hex(sta))
+        # self.assertEqual(urukul.urukul_sta_ifc_mode(sta), 0b0001)
+
+    def test_switches(self):
+        self.execute(UrukulExp, "switches")
+
+    def test_switch_speed(self):
+        self.execute(UrukulExp, "switch_speed")
+        dt = self.dataset_mgr.get("dt")
+        print(dt)
+        self.assertLess(dt, 5*us)
+
+    def test_switches_readback(self):
+        self.execute(UrukulExp, "switches_readback")
+        sw_get = urukul.urukul_sta_rf_sw(self.dataset_mgr.get("sta_get"))
+        sw_set = self.dataset_mgr.get("sw_set")
+        self.assertEqual(sw_get, sw_set)
+
+    def test_att(self):
+        self.execute(UrukulExp, "att")
+        att_set = self.dataset_mgr.get("att_set")
+        att_get = self.dataset_mgr.get("att_get")
+        self.assertEqual(att_set, att_get)
+
+    def test_att_speed(self):
+        self.execute(UrukulExp, "att_speed")
+        dt = self.dataset_mgr.get("dt")
+        print(dt)
+        self.assertLess(dt, 5*us)
+
+    def test_io_update(self):
+        self.execute(UrukulExp, "io_update")
+
+    def test_sync(self):
+        self.execute(UrukulExp, "sync")