nVidia F@H ??

[Denithor]

Just curious...

8800GTS/GTX/Ultra have been the most powerful cards on the planet for a year now...

8800GT is now launched with near GTX power for $250ish...

How come they haven't bothered to design the folding program to run on these cards?!?

 

[waffleironhead]

if i remember correctly, the main dev for gpufah has been/is in the process of his phd project. so phd>new gpu client.

 

[Assimilator1]

That's a shame ,they only have 1 F@H GPU dev??

 

[Insidious]

The F@H project leader has always demonstrated an apparent belief that there is really only one CPU manufacturer, (Intel) and only one worthy GPU as well, (ATI)

They get rather indignant when asked to consider writing optimized code for 'lesser' architectures.

I've had several good laughs at their forum reading the rationale (pronounced: 'dogma') for these short sighted project limitations.


-Sid

 

[waffleironhead]

I went over to the fah forums and read thru the discussion of the nvidia and it seems it all boils down to this(as they claim anyway)


"The question has been asked, and answered many times in this thread, and elsewhere.
The problem is that the drivers supplied by nVidia don't work with the GPU client. FAH is waiting on working drivers from nVidia. As you may know, nVidia has been having problems developing drivers due to the MS Vista release.
This is not an issue which can be solved by a FAH client.

Historically, the FAH GPU client development was started on an nVidia platform, but persistent problems with data accuracy/integrity seemed to be inherent with nVidia GPUs, so the GPU client was fully developed using the ATI X19XX platform instead.

NVidia has touted the potential of their latest processors for GPGPU implementations.
There is speculation that nVidia has realized that scientific inaccuracies remain in their current chips, and that the FAH GPU client would reveal those flaws (bad news for nVidia), so working drivers have not been forthcoming.
If the design flaw persists in the 88xx chips, then we won't see a working nVidia-FAH GPU until the design bug is fixed in their next generation chips.

NVidia is welcome to disprove the speculations... are they listening?"

 

[Insidious]

F@H's company line on this was originally that the client developed for the ATI GPU ran so slowly on Nvidia that it was not practical and F@H was not interested in developing two GPU clients.

But this is totally normal for Pande and crew... each time it is pointed out that they speweth fecal matter... they just do a re-write. (same ending of course)

:beer:

-Sid

another windmill bested on the field of honor

 

[Assimilator1]

Sid
Yes ,they were saying that prior to the release of the GF 8800s ,the GF 7950s are indeed less powerful than the X19xx ,though you'd think they'd be powerful enough *shrug*.The 8800s are of course a different story .

waffleironhead
Interesting ,I wonder if Nvidia will ever replie?

 

[Denithor]

Originally posted by: Assimilator1
Sid
Yes ,they were saying that prior to the release of the GF 8800s ,the GF 7950s are indeed less powerful than the X19xx ,though you'd think they'd be powerful enough *shrug*.The 8800s are of course a different story .

waffleironhead
Interesting ,I wonder if Nvidia will ever replie?

So...the 8800 series has been out for over a year now, ruling undisputed as the most powerful cards available...and FAH still won't take the steps necessary to put all that processing power to use?

Wow. So much for working together for a good cause.

Too bad my x1900gt went out on me recently, my scores have really suffered since and I plan to pick up an 8800gt once the gouging settles down. I guess I'll just keep folding with my e6400 and Opty 165.

 

[Assimilator1]

Maybe you should read waffleironhead's post ,apparently their waiting on Nvidia to fix a driver bug.

 

[Denithor]

Sounds more like they're too lazy to make it work.

Really too bad, there's an awful lot of gpu cycles going to waste out there.

 

[GLeeM]

I might be mistaken but I thought ATI helped alot to get the GPU client working.

 

[waffleironhead]

Originally posted by: GLeeM
I might be mistaken but I thought ATI helped alot to get the GPU client working.

you are correct ati had a lot of work on this, it was in their best interests to get a working fah driver. Kinda makes you wonder why the fah ppd are so low if we are indeed generating so many flops of computation. If I were ati I would lean on fah to increase the ppd (might increase sales).

 

[Insidious]

There has always been speculation with regard to 'help' from Intel around the same time AMD processors were 'found to be lacking' (QMD) and then later on another round of ATI 'help' and suddenly NVidia was also 'found to be lacking'. I always thought it was pretty obvious, but then.... I'm a cynic by nature.

-Sid

 

[GLeeM]

Originally posted by: waffleironhead

Originally posted by: GLeeM
I might be mistaken but I thought ATI helped alot to get the GPU client working.

you are correct ati had a lot of work on this, it was in their best interests to get a working fah driver. Kinda makes you wonder why the fah ppd are so low if we are indeed generating so many flops of computation. If I were ati I would lean on fah to increase the ppd (might increase sales).

Yeah, I've thought the same thing.

Taking it one step further, I wonder if maybe the results are not all they thought they would be. Can't get info from PandeGroup. They don't push for more GPU clients. The regular posters (mods and admins) at the F@H community forum, (that should know stuff), don't encourage people to use the GPU client.

It's not like they have too many GPU folders and not enough WUs, one of the GPU servers has over 100,000 WU available!

Really, the only info is the original info that says that the GPU can get more science done than anything else.

 

[7im]

Originally posted by: Insidious
The F@H project leader has always demonstrated an apparent belief that there is really only one CPU manufacturer, (Intel) and only one worthy GPU as well, (ATI)

They get rather indignant when asked to consider writing optimized code for 'lesser' architectures.

I've had several good laughs at their forum reading the rationale (pronounced: 'dogma') for these short sighted project limitations.


-Sid

Sheesh, who's slinging the Dogma in this forum? isgust; I'm not here to pick a fight, but someone needs to set the record straight.

You should check your history Sid. The original GPU code was developed on NV hardware. However, by the time the GPU code was ready to be put in a F@h client, the ATI hardware had leap frogged the NV hardware in performance, so ATI is where the F@H client started. With the release of the latest NV cards, the F@h project would have two GPU clients IF NV could provide working code/drivers, as you well know from the GPU threads in the F@h forum. That's an NV issue, not a Stanford issue. So why would you say Vijay is single minded about hardware when he has clearly demonstrated they support multiple hardware and platforms... CPU clients, GPU clients, SMP clients, PS3 clients, Mac OSX clients...

Optimized code for lesser hardware? What lesser hardware? You calling AMD lesser hardware? You still crying about about the QMD work units that ended 2 years ago? Oh, wait, you are... I just hadn't gotten to your next post yet. Cripes, man, get with the present. You're just upset the SMP client performance on Barcelona doesn't smoke the C2Qs. :evil: And that's simply because of the smaller L2 cache compared to the Intel architecture. You can't blame Stanford for that one.

You can call F@h shortsighted if you want, but F@h is the first DC Project to run on an a GPU, and first to run on a PS3. If you call that shortsighted, you must be blind.:roll:

And yes, some people, like me, become rather indignant when people, like you, spread the kind of false manure you are slinger here.

7im
FCF Mod


P.S. To answer the orignal question, F@h is still working closely with NV to hammer out the problems, and get a GPU client built and released for NVidia owners. Vijay hinted an some new developments coming in time to make a good "present" whatever that means. Maybe Sid has a birthday coming up soon, and we'll have a nice big birthday cake, with a crow on top. At least that sounds encouraging to me.

@ GLeeM, I encourage people to run the GPU client all the time. I posted about it earlier today... http://www.maximumpc.com/forum...ic.php?p=727407#727407

 

[Insidious]

LOL

Well if Mohamad won't come to the mountain..... (Oh Brumhilda! :laugh: )

Welcome to Anandtech :beer:

-Sid

 

[GLeeM]

Originally posted by: 7im
I'm not here to pick a fight,

For those that don't know, he really thinks/means this, even though people here might say otherwise of posts like this

Originally posted by: 7im
the ATI hardware had leap frogged the NV hardware in performance,

I think this was in F@H performance, not necessarily other types of performance.

Originally posted by: 7im
Optimized code for lesser hardware? What lesser hardware? You calling AMD lesser hardware? You still crying about about the QMD work units that ended 2 years ago? Oh, wait, you are... I just hadn't gotten to your next post yet. Cripes, man, get with the present. You're just upset the SMP client performance on Barcelona doesn't smoke the C2Qs. :evil: And that's simply because of the smaller L2 cache compared to the Intel architecture. You can't blame Stanford for that one.

You can call F@h shortsighted if you want, but F@h is the first DC Project to run on an a GPU, and first to run on a PS3. If you call that shortsighted, you must be blind.:roll:

And yes, some people, like me, become rather indignant when people, like you, spread the kind of false manure you are slinger here.

This kind of vitriol is not appreciated here.
It is this kind of stuff at FCF that is responsible for chasing away thousands of people from F@H who might have otherwise helped find a cure.
I know you have heard that before but when you can't seem to grasp this, obvious to us, point I just have to keep trying, call me a "pessimist" but my hope springs eternal.
EDIT: Maybe I should take a different track and welcome you here whenever you need to get something out of your system with this kind of venomous language. It would be better for Pande Group because not as many potential folders would see it!

Originally posted by: 7im
@ GLeeM, I encourage people to run the GPU client all the time. I posted about it earlier today...

Yeah, I saw that and had to laugh
But you have to admit that for the most part the encouragement is toward SMP for points :roll:

 

[Assimilator1]

The X1950s are faster than the 7950s in many games too ,maybe that's what he meant by leap frogging?

7im
Welcome to AT forums
Interesting post ,though maybe that last line of yours about manure wasn't too clever

 

[7im]

GLeeM, I apologize for the vitriol, but what you call fury, I call fervor. So then are you saying its okay to spread BS and Dogma in this forum, as long as its done calmly and politely? Or only if I agree with this forum's point of view for bashing the project/project lead? Or do I get a pass regardless of my views if I happen to fold for team # 198? Please explain the difference to me.

==============

Yes, I am only speaking to the performance of GPUs for F@H. The GPU code was developed on NV. The GPU client was released on ATI because the X1xxx series of cards folded significantly better than the previous NV generation. Then the NV 86xx and 88xx came along, and the ATI and NV were on about equal ground for F@h performance. However, NV wasn't as helpful as ATI in getting a working driver for the F@h GPU client, and that's where we are today. Game performance of a GPU has no bearing on the development or usage of a GPU for a F@h client.

And I find it ironic that some would claim the project gives preference to only Intel and ATI, which would also indicate that AMD and NV are getting shorted (They aren't!). Well, it's going to be harder to spread that BS since ATI is now AMD. With that line of thought, the project would have to end the GPU client altogether, and the opposite is going to happen in the next few months.

And yes, the current encouragement appears to be for SMP, as it is one of the most productive (scientifically) clients at this point in the project. However, as I hinted, crow will be served shortly for those who think GPUs aren't being encouraged as well. Don't dump your X1xxx cards just yet boys and girls. Neither should the NV crowd.

 

[natethegreat]

Just curious, how can you compare F@H performance on ATI and NVIDIA GPU's when there is no client for NVIDIA GPU's? How is Stanford running F@H on NVIDIA hardware?

 

[GLeeM]

Originally posted by: 7im
GLeeM, I apologize for the vitriol, but what you call fury, I call fervor. So then are you saying its okay to spread BS and Dogma in this forum, as long as its done calmly and politely? Or only if I agree with this forum's point of view for bashing the project/project lead? Or do I get a pass regardless of my views if I happen to fold for team # 198? Please explain the difference to me.

==============

Yes, I am only speaking to the performance of GPUs for F@H. The GPU code was developed on NV. The GPU client was released on ATI because the X1xxx series of cards folded significantly better than the previous NV generation. Then the NV 86xx and 88xx came along, and the ATI and NV were on about equal ground for F@h performance. However, NV wasn't as helpful as ATI in getting a working driver for the F@h GPU client, and that's where we are today. Game performance of a GPU has no bearing on the development or usage of a GPU for a F@h client.

And I find it ironic that some would claim the project gives preference to only Intel and ATI, which would also indicate that AMD and NV are getting shorted (They aren't!). Well, it's going to be harder to spread that BS since ATI is now AMD. With that line of thought, the project would have to end the GPU client altogether, and the opposite is going to happen in the next few months.

And yes, the current encouragement appears to be for SMP, as it is one of the most productive (scientifically) clients at this point in the project. However, as I hinted, crow will be served shortly for those who think GPUs aren't being encouraged as well. Don't dump your X1xxx cards just yet boys and girls. Neither should the NV crowd.

Much better! Thank you :thumbsup:

Yes, at least in the DC forum it must be done politely.

This forum has no point of view for bashing a project/project lead. But we are a TeAm. When a TeAmmate is injured we all feel his pain. This forum is "home", a place where we feel safe and wanted. So we deal gently with each other, (usually).

To show how serious this is: IF you could find anything like your rudeness in this forum, you would have to go back more than three years!!! (This from PMs, shocked and disgusted, wondering what they could/should do about your post)

So yeah, over the years you have made it difficult:
1. Getting TeAmmates to fold.
2. Keeping TeAmmates in F@H.

Doesn't really matter though, everyone here is free to crunch what they want. I crunch and have crunched many other projects. The main reason I stick mostly with F@H and help to get/keep more folders is a friends mother had Alzheimers.

So for my friend :brokenheart: please deal kindly with F@H volunteers!

 

[Assimilator1]

7im
To disagree with someone is fine ,to say someone is plain wrong or simply inaccurate is fine ,to say someones info is out of date is fine too, but to say someone is deliberatly lieing is not fine,it is rude afterall (unless you have hard proof).
Using tact & a little diplomacy is a good idea & may help you get your point across sometimes

I look forward to crunching more F@H on my 1950 Pro (after the SETI race)

 

[7im]

Originally posted by: natethegreat
Just curious, how can you compare F@H performance on ATI and NVIDIA GPU's when there is no client for NVIDIA GPU's? How is Stanford running F@H on NVIDIA hardware?

The original code base for the F@h GPU clients is Brook. As I said, the GPU client (Brook back end) was developed on the old 7xxx series NV cards, so Stanford knows very well how the code performs. And portions of Brook run on the current generation of NV cards, portions that perform about the same as on ATI cards. Based on that, Mike Houston (GPU Guru at Stanford) estimates the two would perform about the same in F@h, same as they do for most games. When Stanford is able to get working code and drivers for the NV cards, and the new ATI cards, then all portions of Brook will work correctly, and you'll see some new beta clients to play with, eventually.

That, and I read the GPGPU development forum... http://www.gpgpu.org/forums/

Or get your Geek On, and read the slide presentation about the X2900 given by Eric Demers (AMD Sr Architect - Graphics Group) at Stanford earlier this year... https://graphics.stanford.edu/...t=Eric_Demers_R6XX.pdf

 

[biodoc]

Originally posted by: 7im

"You should check your history Sid. The original GPU code was developed on NV hardware. However, by the time the GPU code was ready to be put in a F@h client, the ATI hardware had leap frogged the NV hardware in performance, so ATI is where the F@H client started. With the release of the latest NV cards, the F@h project would have two GPU clients IF NV could provide working code/drivers, as you well know from the GPU threads in the F@h forum. That's an NV issue, not a Stanford issue. So why would you say Vijay is single minded about hardware when he has clearly demonstrated they support multiple hardware and platforms... CPU clients, GPU clients, SMP clients, PS3 clients, Mac OSX clients...

Optimized code for lesser hardware? What lesser hardware? You calling AMD lesser hardware? You still crying about about the QMD work units that ended 2 years ago? Oh, wait, you are... I just hadn't gotten to your next post yet. Cripes, man, get with the present. You're just upset the SMP client performance on Barcelona doesn't smoke the C2Qs. :evil: And that's simply because of the smaller L2 cache compared to the Intel architecture. You can't blame Stanford for that one.

You can call F@h shortsighted if you want, but F@h is the first DC Project to run on an a GPU, and first to run on a PS3. If you call that shortsighted, you must be blind."

----------------------------------------------------------------------------------------------------

IMHO you have to consider where the $$ to support the Stanford lab are coming from too:

Apple (good OSX client)
ATI (GPU client)
Sony (PS3 client)
Intel (QMD bonus, C2Q performance/bonus)
NSF & NIH (our tax $)
Crunchers (hardware & electricity $$)

NV & AMD need to provide funding & internal help to Stanford to get the same attention.

I'm not critical of the Stanford lab for pandering to these companies that offer them financial support and I know they have to bring in $$ to survive in a highly competetive academic environment but there's no reason to put them on a pedestal either.

The crunchers with AMD & NV hardware are feeling a bit left out so the risk to the Stanford lab is they may take their cycles to another project.

In the end, it's the science that's important, right?
:beer:

 

[7im]

Originally posted by: biodoc...

In the end, it's the science that's important, right?
:beer:

Yes, that's why they do things like QMD work units, or release a Linux SMP client months before a Windows client is possible. It's not to screw AMD users, or Microsoft. It's becasue Stanford DOES follow the science, wherever it leads even if it doesn't please everyone, not because someone pays them to do it.

Grant money is for the science, not to build a specific client on a specific platform. NIH doesn't give money to "go build a SMP client that runs really fast on Intel hardware." NIH gives money to help with the science of the project, so we can eventually find treatments or cures.

However, if one company is willing to donate hardware, and provide development support, (non-monetary help) then those developments move along much quicker. Hence a working ATI client, but not NV (yet). And Sony put in a lot of help to get that client working. A couple of free GPUs or PS3s are nothing compared to the help a Sony Cell programmer can provide, or an ATI chip designer can provide.

I can understand why some people feel left out, but please remember why the science gets the priority (not user's feelings) and why we donate to this project. But people's memories are either very short, or very selective. For instance, I might have felt left out 2 years ago when AMD released the X2 dual core chips long before Intel did, and AMD users were cleaning up on points at twice the rate of Intel users. But then all I ever hear about are QMDs. Get some perspective there fanboys. :roll: The hardware/performance pendulum tends to swing both directions. Intel wins for a while, then AMD, now Intel again, then AMD will come back. Same thing happens with ATI and NV.

Complaints from the hardware owners of the "out of fashion at the moment" crowd just sounds like sour grapes or poor sportsmanship, not constructive criticism for improving the project. And we never hear from them when they are on top, but only when they're on the bottom. Until the sportsmanship improves and the blame game ends, I'll just have to chalk it up to human nature and deal with it.