new public NV projects 5508-5513, 5017-5022

[DanEnsign]

is for example "P5509_lam5w_330K_g91" a lot more worth for the project than p5504_fip35_ww_domain

It is now. Part of the reason is that we've collected so much data on the WW domains that we're probably 3 weeks behind in analysis -- and needless to say no one has ever studied this system so extensively (not even close), so we can 'quit while we're ahead,' so to speak. Same goes with the villin projects.

The new ones (lam5w, or lambda repressor 5-way mutant) are extremely cool because no one has ever folded a protein this big in simulations. We (PG and FAH donors) will though, just give it a few months.

Dan

[MoneyGuyBK]

Very Well Said:

Classic Bait and Switch.. Just kidding. Fold on.
So the run of bonus PpD with the nVidia client was good while it lasted.

For the record I was not complaining !!!

I have been around F@H long enough (since July '07) to know better.
From time to time the mix of WUs change... and I am just here to Fold what comes down the *chutes.

As was the case with the SMP Client:
This happened a while back when 2605 and 2653 were replaced by 2665.
Except back then the 2665 was worth more, even though it took longer to complete ... and still ended up with lower PPD.
Now, we have the 2662, 2668 and 2669 which seem to be more stable and robust, some credit going to the A2 core of course.

This just requires us to be more flexible and prepare for the *laundry
In the end... we all do this (or at least should) for Humanity & Science.
the new 430_Pointer WUs are doing about 3800_PPD on The Monster compared to 5300_PPD for the 480_Pointers.


Peace

[CoreX5]

Why if these projects are larger, are they worth less? Also why is the benchmark machine an ati 3850 when the same projects are not being run on nvidia as the ones being run on Ati clients? Just seeking some clarification as this could look like it was changed just to appease ati and their clients.

They are worth less because despite being bigger the benchmark machine processes them quicker than the old units (I think the simulations are shorter, just the proteins are bigger). The benchmark machine has always been a 3850 so nothing has changed.

And as an aside, the project staff have repeatedly said that nvidia's huge lead in the PPD stakes may change. It now has (I'm an nvidia folder too so I would be as miffed as anyone were it not for the fact that I don't give two hoots about points), so you can't say we weren't warned

So if the new projects are geared towards the ati that can do bigger proteins in a shorter simulation, perhaps the base assumptions are favoring the ati. Are projects being created that are better suited for an ati type machine? Is this an attempt to make ati folders happier after they have been complaining for months, by choosing projects that favor the ati. Is it perhaps time to take an Nvidia machine and an ati machine and average the results of the two for each new work unit to come up with the base point value?

[jevans64]

My GTX 280 is down from 7000 PPD to about 5800 PPD on a p5509. I've been expecting some kind of devalue for the nVidia GPUs to get them more in line with AMD. I think the nVidia WUs were over-valued to start with anyway.

[John Naylor]

The work units have not been devalued, nor is there a motive to bring down NV points to make ATI and NV produce roughly equal points... Mike [Houston, the guy who builds the ATI cores] has a lot of optimisations on the backburner for his core while they sort out some obscure issue. If they wanted to level the production field, then those optimisations would probably be in already (from what I've read the problem is extremely obscure and not science related). The GPUs were always going to be used for bigger units, their enormous speed makes them inherently suitable for processing big systems, which can bog down the uniprocessor clients into multi-day units (e.g. p1487, which has taken a minimum of 4 days and a maximum of 11 days to fold on my system, and anywhere inbetween [that's not including downtime]). Both NV and ATI cards will be used to fold big units, but this was known from the start. It just seems that the ATI cards excel with these units where NV excels with smaller units. There is no intentional redistribution of PPD with these units.

[toTOW]

There is no point about calling it favoritism : we have always said that the first batch of WU (with ~550 atoms) was doing very well on the NV hardware and that might not be the case with future WU (see the "Preliminary results" in the PPD thread title).

It's just a matter of architecture : NV chose to use a few amount of complex shaders which are fast, but few. ATI chose simpler shader, which are a bit less efficient, but a lot.

On NV architecture, more atoms (and thus complexity and parallelism) means that there are no free SP to use, as they were already used hence the PPD drop.

On ATI hardware, there are still plenty SP available, so more complexity means better performance, as there were SP available.

We still see performance difference between two architectures because of SP efficiency, but both hardware tend to join their performances : NV slows a bit, and ATI gets a bit faster.

[CoreX5]

Given that they have different strengths would a combined benchmark machine not make sense than? When deciding on a point value average the results of an ati and nvidia machine? Would this not give a better starting value for all types of projects?

[toTOW]

Why are you complaining ? you are still getting better PPD than most of the ATI users with your NV GPU ...

CoreX5> this rises another problem : what should be the NV benchmark machine ... and what PPD should it get, taking into account the science output differences between ATI and NV

[slegrand]

No worries, there are plenty of NV optimizations in the pipe as well, but I need to validate them before turning them loose and right now, I'm swamped with some other F@H-related work. As things stand, the NV algorithm is more efficient but the ATI algorithm is simpler. Hence around lambda-sized molecules, ATI gets a perf spike while NV is still warming up and bogged down by the housekeeping incurred for its more efficient algorithm. For Fip35-sized molecules, NV is underutilized on GTX280, but ATI is doubly so on 4870 and that's why the perf differential there is just crazy.

In the long run in this GPU generation, I *expect* SP for SP that an NV GPU will be about 2.5x faster than an ATI SP. This is of course then modified by SP count and processor clock. If Mike finds a structural optimization to his code, then I will revisit that ratio, but that's my prediction for now. And the game resets with the next hw generation anyway so it's an ongoing arms race.

The work units have not been devalued, nor is there a motive to bring down NV points to make ATI and NV produce roughly equal points... Mike [Houston, the guy who builds the ATI cores] has a lot of optimisations on the backburner for his core while they sort out some obscure issue. If they wanted to level the production field, then those optimisations would probably be in already (from what I've read the problem is extremely obscure and not science related). The GPUs were always going to be used for bigger units, their enormous speed makes them inherently suitable for processing big systems, which can bog down the uniprocessor clients into multi-day units (e.g. p1487, which has taken a minimum of 4 days and a maximum of 11 days to fold on my system, and anywhere inbetween [that's not including downtime]). Both NV and ATI cards will be used to fold big units, but this was known from the start. It just seems that the ATI cards excel with these units where NV excels with smaller units. There is no intentional redistribution of PPD with these units.

[CoreX5]

I am not complaining, I am simply questioning. Is this not the basis of academia and the principle upon which scientific research should be conducted? I fail to understand why people get so defensive when they are questioned. I am just trying to determine if there is not a better way to go about assigning points which does not skew certain projects in favour of one type of gpu over a different type.

Also, you want people to base decisions and equipment they buy based on the science that these points are supposed to represent. I think an open discussion would be helpful to prevent sudden shifts in the landscape as new types of projects are started, compared to what people are used to.

[John Naylor]

The discussion is useful but the playing field will always shift. Back before F@H diversified and there was just one type of client - uniprocessor for windows, linux, and mac, the playing field would periodically shift. AMDs are currently and always have been good at the cores that do not use SIMD optimisations (the original TINKER core and the current AMBER core), whereas Intel have been faster at SSE and SSE2 units, a lead that was hugely increased by the appearance of Core (followed by Core 2 and presumably it will be sustained with the upcoming Core i7). The shifts are a natural part of the way F@H works, so the guys who have bought nVidia now may find that in a year's time they once again have the advantage, be it through a core upgrade or a return to smaller units. That is just how F@H works, unfortunately...

[Shiggity]

It was noted that these new work units are more intensive and I can really tell. My temps on my 8800GT core have gone up ~4-5C and my desktop is acting very sluggish now compared to before. Browsing the internet is a bit choppy so is opening folders really quickly. Anyone else noticing this?

[John Naylor]

How much VRAM does your 8800GT have?

[AJL]

Hmm, and I just bought a second 9600GT - well, it was cheap . If ATI cards get better and better, maybe I'll switch next year

[ChasR]

To a layman, it would appear these are the same or at least similar to ATI WUs p4742-p4747, valued at 584 points. Perhaps it is coincidence that the nVidia WUs are valued 26.4% less and nVidia production has declined by the same amount?