Core16/ATI P11293

Moderators: Site Moderators, PandeGroup

Re: Core16/ATI P11293

Postby p0wer » Sat Apr 02, 2011 8:25 am

Scandalman wrote:Weirdly going back into config doesn't list this client-type option in the 'extra slot options' section - which is where it appeared when it was added.

Did you change the XML file while FAHControl was already running? If so, then it means FAHControl reads config file only during startup.
User avatar
p0wer
 
Posts: 70
Joined: Fri Mar 18, 2011 7:46 am
Location: Poland

Re: Core16/ATI P11293

Postby p0wer » Sat Apr 02, 2011 8:33 am

Scandalman wrote:For instance, I am not sure if the <client-type v="advanced"/> option should be global like this...

It should definitely be local for each slot:

Code: Select all
<config>
- <!--  FahCore Control
  -->
  <cpu-affinity v="true" />
- <!--  Folding Slot Configuration
  -->
  <smp v="false" />
- <!--  Logging
  -->
  <verbosity v="3" />
- <!--  Remote Command Server
  -->
  <password v="" />
- <!--  Folding Slots
  -->
- <slot id="0" type="UNIPROCESSOR">
    <client-type v="advanced" />
  </slot>
- <slot id="1" type="UNIPROCESSOR">
    <max-packet-size v="big" />
  </slot>
- <slot id="2" type="SMP">
    <cpus v="2" />
    <max-packet-size v="small" />
  </slot>
 </config>


But you can of course set it global in Expert tab if you want to. This would force all the slots to be advanced.
User avatar
p0wer
 
Posts: 70
Joined: Fri Mar 18, 2011 7:46 am
Location: Poland

Re: Core16/ATI P11293

Postby Scandalman » Sat Apr 02, 2011 9:06 am

Well, that's what I thought would make sense. However I set the flag whilst editing the gpu slot in expert mode then stopped FAHClient and FAHControl after making the change; Then I inspected the config.xml file and the flag was set in what I consider to be the global section. I then restarted the FAHClient and then FAHControl.

Subsequently the flags do not appear at all in the bottom of the slot section 'Extra slot of options (expert only)' of configure for the GPU or smp slots, but it has picked up a core 16 WU for my 6950.

Best, Paul.
User avatar
Scandalman
 
Posts: 45
Joined: Fri Jan 14, 2011 11:38 am
Location: England

Re: Core16/ATI P11293

Postby p0wer » Sat Apr 02, 2011 10:01 am

Global flags won't appear in slots, they will be present in Expert mode on Expert config tab.
User avatar
p0wer
 
Posts: 70
Joined: Fri Mar 18, 2011 7:46 am
Location: Poland

Re: AMD GPU3 & Proj 11293

Postby JimboPalmer » Sat Apr 02, 2011 10:33 am

[Off topic] "We have no undetected errors that we are aware of."
Tsar of all the Rushers
I tried to remain childlike, all I achieved was childish.
A friend to those who want no friends
JimboPalmer
 
Posts: 915
Joined: Mon Feb 16, 2009 4:12 am
Location: Greenwood MS USA

Re: Core16/ATI P11293

Postby Scandalman » Sat Apr 02, 2011 2:03 pm

Thank you.

When adding the flag from the slots page it appears in the bottom of that page as you add it - and then isn't there after a restart. It's just a bit confusing - now you have pointed out the expert tab I can see it was assigned as a global flag and automagically disappears from the slot tab where it was added and re-appears in the expert tab. A bit confusing, but at least it works.

If I'd known this I would have added it in the expert tab and avoided my confusion.

Paul.
User avatar
Scandalman
 
Posts: 45
Joined: Fri Jan 14, 2011 11:38 am
Location: England

Re: Core16/ATI P11293

Postby p0wer » Sat Apr 02, 2011 2:32 pm

This sounds weird. So you added an option through FAHControl in slot configuration in Extra slot options, and it appears as global in Expert? Does not look right, can you check once again?
User avatar
p0wer
 
Posts: 70
Joined: Fri Mar 18, 2011 7:46 am
Location: Poland

Re: AMD GPU3 & Proj 11293

Postby gwildperson » Sat Apr 02, 2011 3:31 pm

[Off topic]If I know the "boss" (and if I remember that particular strip correctly) it didn't end there.
Boss: What will it cost in $ and in weeks to fix all the undiscovered bugs/undetected errors"
gwildperson
 
Posts: 726
Joined: Tue Dec 04, 2007 8:36 pm

Re: Core16/ATI P11293

Postby gwildperson » Sat Apr 02, 2011 6:37 pm

If you add it to the slot and you click "Save" the new config is printed to the log (perhaps depending on verbosity setting). Then it does appear if you go back to edit that slot again.

Adding a setting to a slot which happens to already be the default for that slot due to some global setting is redundant and does not get saved.

Global Setting <xxxx v="a"/>
Slot 2 setting <xxxx v="b"/> (will be saved)
Slot 3 setting <xxxx v="a"/> (will not be saved by FAHClient)
Slot 2 setting <xxxx v="c"/> (will be saved)
gwildperson
 
Posts: 726
Joined: Tue Dec 04, 2007 8:36 pm

Re: Core16/ATI P11293

Postby muziqaz » Sat Apr 02, 2011 6:45 pm

As I just found this thread, and did not have a chance to read it all, I just wanted to add to Dr. Pande's warning about video watching and core 16.
System freeze occurs when folding core 16 WUs with ATI 6 series cards. It happens when video game is launched, flash based content is accessed through any web browser and watching any kind of movies(avi, mkv, blue ray with GPU acceleration and without it). It happens with 11.2 and 11.4 drivers. Previous drivers might have the same effect, but I think it hasn't been tested.
5 series ATI cards do not seem to be effected by these issues. And GPU 2 WUs do not force those issues. And as Dr. Pande said, AMD and Pande Group are on this case.
User avatar
muziqaz
 
Posts: 596
Joined: Sun Dec 16, 2007 6:22 pm
Location: London

Re: Core16/ATI P11293

Postby p0wer » Sat Apr 02, 2011 7:49 pm

gwildperson, you can do everything and enough with FAHControl.
User avatar
p0wer
 
Posts: 70
Joined: Fri Mar 18, 2011 7:46 am
Location: Poland

Questions about core_16 / project 11293

Postby Devildoll » Sun Apr 03, 2011 2:56 pm

Hello fellas!

i just have a few questions about the new core_16 wu/wu's that came with v7 client.

how well does it utilize the whole range of graphics cards ?
i remember reading somewhere that the older clients / WU's were locked to 120 shaders in some cases, no idea if this was true or not , but it made me curious on how the new core / WU's work.

is the new stuff somehow locked to a certain number of shaders , or does it scale up infinitely automatically?
if locked , how many shaders are we talking about? and any idea if and how that would change in the future?


on core_11 i was getting around 4200 PPD with my 5870 , now with core_16 im getting just above 10 K at the same settings. ( 1 GHz core, modified environment variables and the latest dll's )
which is a hefty improvement , but if you compare it to for example a gtx 460 , 10 K points is still a bad daily catch.

considering the 460 loosing in stuff like gaming performance , and having significantly lower compute power , apparently 0.961 Tflops versus the 5870's 2,72 Tflops.
Got me wondering , what attributes is it that make the 460 superior at F@H?

or is the relatively poor performance just tied to the current WU's and the fact that the amd side of the new gpu client in such an early stage?
Image
Devildoll
 
Posts: 10
Joined: Sun Mar 06, 2011 7:28 pm

Re: Questions about core_16 / project 11293

Postby k1wi » Sun Apr 03, 2011 7:03 pm

It's been discussed a bit, bu the current core16 is working on smaller proteins. As larger proteins become available, better performnce and lower CPU use is anticipated to occur.
Image
k1wi
 
Posts: 1152
Joined: Tue Sep 22, 2009 10:48 pm

Re: Questions about core_16 / project 11293

Postby codysluder » Sun Apr 03, 2011 10:32 pm

Devildoll wrote:how well does it utilize the whole range of graphics cards ?

OpenCL is currently only supported on AMD HD5000/HD6000 series GPUs. Older AMD GPUs are still supported using Core 11 until at least September. I have not found any information about the future of OpenCL or other cores for Nvidia.
i remember reading somewhere that the older clients / WU's were locked to 120 shaders in some cases, no idea if this was true or not , but it made me curious on how the new core / WU's work.
It's not true.
is the new stuff somehow locked to a certain number of shaders , or does it scale up infinitely automatically?
if locked , how many shaders are we talking about? and any idea if and how that would change in the future?
Small proteins (ones with few atoms) don't do a good job of keeping lots of shaders busy. Proteins with lots of atoms keep more shaders busy more of the time. Nothing is locked, as I said before. It's like if you have a quad processor and all you're doing is browsing, the work you're doing doesn't keep one core busy and the other three of your cores are just waiting for work. Start up FAH-SMP and suddenly there is plenty of work for all your CPUs.

on core_11 i was getting around 4200 PPD with my 5870 , now with core_16 im getting just above 10 K at the same settings. ( 1 GHz core, modified environment variables and the latest dll's )
which is a hefty improvement , but if you compare it to for example a gtx 460 , 10 K points is still a bad daily catch.

considering the 460 loosing in stuff like gaming performance , and having significantly lower compute power , apparently 0.961 Tflops versus the 5870's 2,72 Tflops.
Got me wondering , what attributes is it that make the 460 superior at F@H?

or is the relatively poor performance just tied to the current WU's and the fact that the amd side of the new gpu client in such an early stage?

While we're beta testing core16, everybody is evaluating stability and reporting crashes if they find one. Once the bugs have been cleaned up, optimization will start and proteins will start getting bigger. You can't judge anything from the current PPD. If you'd rather run a 460, nobody will criticize you, but check back later.
codysluder
 
Posts: 2128
Joined: Sun Dec 02, 2007 12:43 pm

Re: Questions about core_16 / project 11293

Postby Devildoll » Mon Apr 04, 2011 8:31 am

codysluder wrote:
Devildoll wrote:how well does it utilize the whole range of graphics cards ?

OpenCL is currently only supported on AMD HD5000/HD6000 series GPUs. Older AMD GPUs are still supported using Core 11 until at least September. I have not found any information about the future of OpenCL or other cores for Nvidia.

i was wondering more about if there were any utilization discrepensies between different supported cards , mostly wondering about core 16

codysluder wrote:
Devildoll wrote:i remember reading somewhere that the older clients / WU's were locked to 120 shaders in some cases, no idea if this was true or not , but it made me curious on how the new core / WU's work.
It's not true.



okay , awesome , so it has more to do with the fact that these test units are small than anything else.

codysluder wrote:
Devildoll wrote:is the new stuff somehow locked to a certain number of shaders , or does it scale up infinitely automatically?
if locked , how many shaders are we talking about? and any idea if and how that would change in the future?
Small proteins (ones with few atoms) don't do a good job of keeping lots of shaders busy. Proteins with lots of atoms keep more shaders busy more of the time. Nothing is locked, as I said before. It's like if you have a quad processor and all you're doing is browsing, the work you're doing doesn't keep one core busy and the other three of your cores are just waiting for work. Start up FAH-SMP and suddenly there is plenty of work for all your CPUs.

yeah that sounds reasonable.

codysluder wrote:
Devildoll wrote:on core_11 i was getting around 4200 PPD with my 5870 , now with core_16 im getting just above 10 K at the same settings. ( 1 GHz core, modified environment variables and the latest dll's )
which is a hefty improvement , but if you compare it to for example a gtx 460 , 10 K points is still a bad daily catch.

considering the 460 loosing in stuff like gaming performance , and having significantly lower compute power , apparently 0.961 Tflops versus the 5870's 2,72 Tflops.
Got me wondering , what attributes is it that make the 460 superior at F@H?

or is the relatively poor performance just tied to the current WU's and the fact that the amd side of the new gpu client in such an early stage?

While we're beta testing core16, everybody is evaluating stability and reporting crashes if they find one. Once the bugs have been cleaned up, optimization will start and proteins will start getting bigger. You can't judge anything from the current PPD. If you'd rather run a 460, nobody will criticize you, but check back later.


im well aware that that the client is in an early stage, dont get me wrong , im not disapointed in any way.
contrary to that , i am actually pretty excited.

but still very curious as to what causes the the relatively bad ppd of amd cards at this stage.

wondering if it is purely the difference in maturity on the nvidia and amd WU's/cores , test units being small and whatnot , if this is the case , could one then imagine that that 5870 would get more points than a 460 in the same ratio as the difference in Tflops between them , as time goes along and the clients/wu get polished . ( probably cant be that clear cut though )

or if nvidia cards/openCL implementation has a predefined edge on amd cards when it comes to this kind of compute work, if so , it would be cool to know what attributes of the cards govern their ability to do this kind of work.


But then , points arn't the most important thing here , getting the WU's done is the main thing , but the points/competition aspect of F@H still interests me.
Devildoll
 
Posts: 10
Joined: Sun Mar 06, 2011 7:28 pm

PreviousNext

Return to V7.1.52 Windows/Linux

Who is online

Users browsing this forum: No registered users and 1 guest

cron