Is it possible to retrieve information about installed GPUs via GUI_RPC?

<get_host_info> only returns CPU data. So does <get_state>.

Can't you read the sched_request*.xml file? The info about the videocard is in there as well.

In my case, for the Einstein sched_request*.xml file:

<coprocs>
<coproc_ati>
   <count>1</count>
   <name>ATI Radeon HD 3800 (RV670)</name>
   <req_secs>864.864000</req_secs>
   <req_instances>1.000000</req_instances>
   <estimated_delay>0.000000</estimated_delay>
   <target>3</target>
   <localRAM>512</localRAM>
   <uncachedRemoteRAM>256</uncachedRemoteRAM>
   <cachedRemoteRAM>64</cachedRemoteRAM>
   <engineClock>669</engineClock>
   <memoryClock>829</memoryClock>
   <wavefrontSize>64</wavefrontSize>
   <numberOfSIMD>4</numberOfSIMD>
   <doublePrecision>1</doublePrecision>
   <pitch_alignment>64</pitch_alignment>
   <surface_alignment>256</surface_alignment>
   <maxResource1DWidth>-1</maxResource1DWidth>
   <maxResource2DWidth>8192</maxResource2DWidth>
   <maxResource2DHeight>8192</maxResource2DHeight>
   <CALVersion>1.4.403</CALVersion>
    <atirt_detected/>
</coproc_ati>
    </coprocs>

---

Fred M and I are chatting with Jason about automating the flop-balancing for optimised SETI applications, while we're waiting for app_version specific TDCFs. (lunatics)

GUI_RPC would be a more 'professional' way to go than scraping sched_requests - though knowing Jason, he'll probably end up querying the hardware directly. Having it in the GUI - I think leaving it out is just an oversight - would also expose it to remote managers like BoincTasks.

But you only have to read it once. I don't expect my videocard/GPU to change every so many seconds, do you? So you'd only have to read it at app-start.

---

But you only have to read it once. I don't expect my videocard/GPU to change every so many seconds, do you? So you'd only have to read it at app-start.

I don't expect my CPU to change every few seconds, either.

It just seems inconsistent to restrict get_host_info to CPU only in this day and age. It shouldn't be any more difficult to spit out the <coprocs> vector in response to a GUI_RPC, than it is when preparing a sched_request.