How well could X360 handle F@H?

[Karma Police]

I'm a bit of a folding fan, but don't have a PS3. Assuming MS implemented it just like Sony did, would it be comparable to an R580? What problems would MS run into while trying to get the X360 or even just the Xenos to work?

EDIT: That said, since the Wii also has an ATi GPU in it (Hollywood), how hard would it be for Nintendo to do the same thing (no matter how unlikely it may seem).

 

[Rainbow Man]

I don't know anything about how great or not it might be. I hope it could be really great of course.

I just wonder if it can be great at folding AND able to draw a nice on-screen graphical display at the same time..assuming it was to do GPU folding that is.


Also. I wsander if MS would want to release a folding client if for some reason they would not be able to match PS3's performance here. Sad thought really but PR reasons could come to outweigh scientific and human benefits.

We can only hope MS sees the light in providing OTHER non-gaming applications for their great box than just watching videos and listening to musci.
Peace.

 

[Karma Police]

Also. I wsander if MS would want to release a folding client if for some reason they would not be able to match PS3's performance here. Sad thought really but PR reasons could come to outweigh scientific and human benefits.

That's just it, though. According to quiet a few people, ATi's GPUs are several times more powerful than the Cell at protein folding:

http://www.dailytech.com/article.aspx?newsid=6609

When it comes to pure performance though, the PS3’s Cell Broadband Processor is still no match for ATI GPUs for protein folding. The GPUs on Folding@home sit at 41 current TFLOPS, which come from only 700 processors. If there were as many GPUs folding as there are PS3s on the network, it can be extrapolated that GPUs could reach 876 TFLOPS.

 

[Dave Baumann]

I just wonder if it can be great at folding AND able to draw a nice on-screen graphical display at the same time..assuming it was to do GPU folding that is.

The "Graphical GPU Client" already does this on the PC.

Also. I wsander if MS would want to release a folding client if for some reason they would not be able to match PS3's performance here. Sad thought really but PR reasons could come to outweigh scientific and human benefits.

Its likely that the throughput would be greater if the GPU were used.

 

[quest55720]

Dave do you think xenos could be used for folding like the r500 series? Would it be because of memexport?

 

[Karma Police]

Dave, thanks for your help. You did the write up on the Xenos. Are there any technical problems setting something like this up? Do you think a person could code a Live Arcade "game" for F@H under the new 150 MB limit by themselves without MS's help?

 

[Rainbow Man]

The "Graphical GPU Client" already does this on the PC.

Oh! Okay I didn't know that (I have nvidia graphics card..grr can't fold on that..)

Its likely that the throughput would be greater if the GPU were used.

Well what's stopping them then! Get on with it MS!

Think how great they could look. Even greater computing performance AND cheaper to boot!
The only thing that would keep me form folding on both PS3 and 360 would be those darned fans screeching constantly in my ear.
Peace.


PS: Has anyone noticed weird graphical artifacts in the onscreen graphics of PS3 folding client?

Press select to get the big fusll-scren molecule display and pause the graphics update. XZoom in as close as you can and look at the edge between atoms.

Do you see little black pixels around the fringe? Where the heck do those come from?

And how does the client draw thos eballs anyway? Are they flat (textures?) or really dense polygon meshes? I can't see any blockyness even when zoomed in all yhe way.

 

[Dave Baumann]

EDIT: That said, since the Wii also has an ATi GPU in it (Hollywood), how hard would it be for Nintendo to do the same thing (no matter how unlikely it may seem).

F@H GPU requires fairly sophisticated shader capabilities, so I don't think you should look to the Wii for this type of thing.

Dave do you think xenos could be used for folding like the r500 series? Would it be because of memexport?

Current GPU Folding still uses DX as the interface to the GPU (although coded to via Brook) so Memexport isn't a requirement as its not part of the DX specification.

Are there any technical problems setting something like this up? Do you think a person could code a Live Arcade "game" for F@H under the new 150 MB limit by themselves without MS's help?

I don't know the nuances of the application itself, but ostensibly Xenos is a superset of a DX9 SM3.0 processor, so, hardware wise, it should have the capabilities to replicate the processing the PC version of the GPU client does - the EDRAM might present some challenge in porting, but it could also provide a benefit with all that bandwidth.

The problem is that there are very few people that a.) understand what folding is going and b.) can map that on to GPU processing. Mike Houston (who is familiar at the B3D forums outside of the Console forum) is just one of the few that can do both of these things.

 

[patsu]

That's just it, though. According to quiet a few people, ATi's GPUs are several times more powerful than the Cell at protein folding:

http://www.dailytech.com/article.aspx?newsid=6609

http://forum.beyond3d.com/showpost.php?p=954581&postcount=195

 

[Karma Police]

http://forum.beyond3d.com/showpost.php?p=954581&postcount=195

That would just help even up the Cell's and the GPU's then, since the GPU's seem to be putting out more TFLOPS than the Cell.

So GPU have twice the TFLOPS but have to run them twice for accuracy compared to the CELL, right?

Nothing wrong with that & certainly doesn't mean using GPU's for folding is a bad thing.

 

[patsu]

That would just help even up the Cell's and the GPU's then, since the current GPU's seem to be putting out more the TFLOPS as the Cell.

GPU ----> about twice the TFLOPS but have to run them twice for accuracy compared to the CELL, right?

Nothing wrong with that & certainly doesn't mean using GPU's for folding is a bad thing.

Dunno. Just helping to correct a (potential) misconception. Of course the more the merrier.

The PS3 F@H client also does not fully utilitize the Cell processor since I can do background download, play some enviromental sound (built into the client), browse the web and display the worldmap/molecule simultaneously without any noticeable drop in performance.

 

[aselto]

Dunno. Just helping to correct a (potential) misconception. Of course the more the merrier.

The PS3 F@H client also does not fully utilitize the Cell processor since I can do background download, play some enviromental sound (built into the client), browse the web and display the worldmap/molecule simultaneously without any noticeable drop in performance.

It's exactly like on PC. F@H has low priority, it frees resources when other applications demand it, but it doesn't mean the programme doesn't utilize resources fully.

 

[patsu]

It's exactly like on PC. F@H has low priority, it frees resources when other applications demand it, but it doesn't mean the programme doesn't utilize resources fully.

You mean it makes use of available resources fully. I am refering to the maximum resources. At no time does F@H use up all the available SPUs (1 is always left alone it seems).

 

[aselto]

You mean it makes use of available resources fully. I am refering to the maximum resources.

But that's the point of the program isn't it? That user can work on his PC(PS3) while F@H is doing calculations without disturbing user. Still, while you're in PS3's dashboard why would OS use more than one SPE (meaning 6/7 of Cell's resources are essentially available)

 

[patsu]

But that's the point of the program isn't it? That user can work on his PC(PS3) while F@H is doing calculations without disturbing user. Still, while you're in PS3's dashboard why would OS use more than one SPE (meaning 6/7 of Cell's resources are essentially available)

The F@H client on PS3 is a foreground application. You can get PS3 to start it automatically after at least 10 minutes of idle time. In theory, they should be able to max out the performance... but since 1 SPU is reserved by the OS at all times. It gets to play with only 5 SPUs.

What is the usage scenario and behaviour when we use GPU to run F@H on a PC ?

 

[Shifty Geezer]

The F@H client on PS3 is a foreground application. You can get PS3 to start it automatically after at least 10 minutes of idle time. In theory, they should be able to max out the performance... but since 1 SPU is reserved by the OS at all times. It gets to play with only 5 SPUs.

6 SPUs. More importantly that one SPU isn't going to make a huge difference versus GPUs, which is how it's been raised. 'Will GPUs be faster than Cell?' 'Well, PS3 isn't using all its resources.' A GPU with hundreds of MBs VRAM ona super fast bus, and lots and lots of execution units, isn't going to be slow...

As for ATi can Fold, nVidia can't, it's not like every ATi chip can fold! Only the latest (We now support serveral classes of GPU boards, including X1600, X1800, and X1900 class GPU's from ATI). Using the GPU client pretty much ties up the machine reading the FAQ (http://folding.stanford.edu/FAQ-ATI.html). You don't really want to multitask with it much, though playing audio files and low stress like that shouldn't affect it much (someone running the GPU client must be able to comment!).

 

[patsu]

6 SPUs.

Ouch... it seems that I can't count anymore.

More importantly that one SPU isn't going to make a huge difference versus GPUs, which is how it's been raised. 'Will GPUs be faster than Cell?' 'Well, PS3 isn't using all its resources.' A GPU with hundreds of MBs VRAM ona super fast bus, and lots and lots of execution units, isn't going to be slow...

Sure... Xenos's mem-export should be interesting too.

Some part of me still want that 1 SPU back for serious number crunching.

As for ATi can Fold, nVidia can't, it's not like every ATi chip can fold! Only the latest (We now support serveral classes of GPU boards, including X1600, X1800, and X1900 class GPU's from ATI). Using the GPU client pretty much ties up the machine reading the FAQ (http://folding.stanford.edu/FAQ-ATI.html). You don't really want to multitask with it much, though playing audio files and low stress like that shouldn't affect it much (someone running the GPU client must be able to comment!).

Running web browser can be intensive though. Refreshing the F@H stats page took 3-11% of my laptop CPU. I'm keen to observe PS3's performance profile under various usage scenarios.

 

[Dave Baumann]

As for ATi can Fold, nVidia can't, it's not like every ATi chip can fold! Only the latest (We now support serveral classes of GPU boards, including X1600, X1800, and X1900 class GPU's from ATI). Using the GPU client pretty much ties up the machine reading the FAQ (http://folding.stanford.edu/FAQ-ATI.html).

Well, all ATI SM3.0 board could fold, the only reason it doesn't span to X1300 is because it was deemed not to be faster enough. The client doesn't really tie up the machine - it puts it at 100% CPU utilisation because its polling the driver, but its not doing much in the way of heavy lifting. The GPU client also works with Vista's Aero interface, despite them both using the 3D pipeline.

http://forum.beyond3d.com/showpost.php?p=954581&postcount=195

Curiously, the CELL code is actually derived from the GPU code (one of the reasons they are limited to the same types of WU's).

 

[patsu]

Just curious, how many work units can a GPU client churn out per day ?

 

[22psi]

Hmmm not sure, maybe HDR would look a bit better