Yes, the simplest design that does the job is the best.
Unfortunately, the issue with building hardware for physics experiments is that
there is never a clear-cut and exact specification for what's needed for a
number of reasons:
- it wouldn't be research if we understood everything completely
- often, we can tolerate a decrease in HW performance by adding additional
complexity elsewhere in our experiment
- if we're going to spend large sums of money on HW, we want to make sure that
we have some performance overhead in case our requirements change in the future
-- at something like 10k a Sayma, we don't want to find that it doesn't meet
our long-term needs because we skimped on the specification
- we want to try to leave the door open for new use cases/experiments we
haven't yet thought of
So, the only way to arrive at a good design is to understand the trade-offs
between performance and its impact on users and complexity, so we can try to
make an informed decision about where to aim for. If you want to argue for a
decrease in design complexity, that's fine, but we need details. Various
changes are being discussed. How much difference would each of them actually
make to you? If you just keep repeating that "simpler designs are better"
without giving any more details, I don't see how we're supposed to make any
progress on this.
T
Dr Thomas Harty
Junior Research Fellow, St John's College
University of Oxford, Department of Physics
The Clarendon Laboratory
Parks Road, Oxford, OX1 3PU
Mob: +44 (0)7986 375 052
Lab: +44 (0)1865 272 572
From: Thomas Harty
Sent: 09 August 2018 08:12
To: Sébastien Bourdeauducq; artiq@lists.m-labs.hk; Robert Jördens
Subject: Re: SAWG
Sébastien,
Understood. But, that's quite a vague answer. We don't want to throw away
performance/flexibility unnecessarily. So, the questions are: how much do we
need to simplify the SAWG to make it okay to debug and maintain at the 1GSPS
data rate? and, what's the way of doing that, which has the least impact on
users? If we need a short exploratory project to answer those questions then so
be it.
Tom
Dr Thomas Harty
Junior Research Fellow, St John's College
University of Oxford, Department of Physics
The Clarendon Laboratory
Parks Road, Oxford, OX1 3PU
Mob: +44 (0)7986 375 052
Lab: +44 (0)1865 272 572
From: Sébastien Bourdeauducq
Sent: 09 August 2018 02:53:43
To: Thomas Harty; artiq@lists.m-labs.hk; Robert Jördens
Subject: SAWG
On Thursday, August 09, 2018 03:15 AM, Thomas Harty via ARTIQ wrote:
> it's a big FPGA and IIRC we're not
> really pushing the resources limits yet (but maybe I'm wrong about
> that?), so it's not clear that's actually a problem. I get that
> multi-hour compile times are death, but at least we haven't seen any
> Sayma bugs that depend on whether the Sawg is enabled or not for quite a
> while (since we fixed the power supplies). So, maybe it's not actually
> such an issue if the majority of Sawg testing can be done in simulation,
> and any non-SAWG issues on Sayma can be debugging in non-SAWG builds?
Let's not get too optimistic. There can be difficult problems with
timing closure and routing congestion, which are already a bit sensitive
with the current design. Those cannot be solved with a partial design
only. Also, even if there are no more Sayma-bugs, simulation/hardware
mismatches might still happen (due to Vivado miscompilations, Verilog
simulation issues, or Migen bugs). Having a smaller, faster to compile
design is valuable.
Sébastien
___
ARTIQ mailing list
https://ssl.serverraum.org/lists/listinfo/artiq