Radeon 5870 performance in distributed computing?

[Grall]

Have anyone seen any figures regarding this yet?

The 4800-series has been somewhat strangled in folding performance due to too-small work units (or so AMD claims anyway), and with the 5870 offering double the amount of stream processors I'm wondering if folding efficiency will be even worse on this chip...

 

[FrameBuffer]

Have anyone seen any figures regarding this yet?

The 4800-series has been somewhat strangled in folding performance due to too-small work units (or so AMD claims anyway), and with the 5870 offering double the amount of stream processors I'm wondering if folding efficiency will be even worse on this chip...

don't hold your breathe.. iirc, F@H comments were along the line that they saw little need in addressing 4XXX (and subsequently 5000) series instead were focusing on maintaining compatibility. IMO, ATI/F@H has abandoned any real desire to persue optimization for ATI, we havent seen any noticable perf increase since the introduction of the HD 3000. History would save that if we haven't seen it by now on the current programming model, then we should expect it ((unless of course a new model is adapted -- highly unlikely imo, or ATI puts some serious effort behind it)).

All IMHO

 

[Silent_Buddha]

I don't expect to see much until/unless FAH moves to OpenCL or DX Compute. As developement of Brook+ has been frozen.

Then again, I'd be surprised if FAH didn't move on to OpenCL.

Regards,
SB

 

[Grall]

There is a thread on the official folding board, discerning from the information there the 5870 is about 30% faster than a 4890, despite 100% more computational resources.

The reason for this is because the ATI client hasn't been optimized for any new Radeon card since the 3xxx series...

And a OpenCL client is apparantly in development - one guy started working on it like a week ago, so it'll be out in roughly a quarter eternity...

 

[Arnold Beckenbauer]

Have anyone seen any figures regarding this yet?

The 4800-series has been somewhat strangled in folding performance due to too-small work units (or so AMD claims anyway), and with the 5870 offering double the amount of stream processors I'm wondering if folding efficiency will be even worse on this chip...

Take a look on these both threads:
http://foldingforum.org/viewtopic.php?f=51&t=10442
http://foldingforum.org/viewtopic.php?f=51&t=11572

Some PPDs: http://www.rage3d.com/previews/video/ati_hd5870_performance_preview/index.php?p=13
It's really (s-)low, I'm getting PPD of 2.900 with these projects.

 

[neliz]

Take a look on these both threads:
It's really (s-)low, I'm getting PPD of 2.900 with these projects.

I thought I read in one of the reviews of the HD5870 doing 4500ppd at 70% gpu utilization.

 

[Arnold Beckenbauer]

I thought I read in one of the reviews of the HD5870 doing 4500ppd at 70% gpu utilization.

70% GPU utilization? That's a sign of two left hands.

No, honestly: To get the full utilization you have to play a bit with environmental variables.
And the next question: What project?

 

[Grall]

Last I checked with GPU-Z, my 4890s were at 90ish percent and ~96-98% respectively without changing any variables at all, folding at roughly 3800ish up to maybe 4600PPD each (guesstimated, depending on the project of course...)

Why would I need to set any variables?