..... In a way I regret my slow migration to the world of ATI because I really do like GPU Grid for example and there is no doubt that the CUDA applications are more common still ......

Been down that road as well, and in some ways still have a nagging itch about moving, I also have a real urge to help at GPUGrid due to one particular application there, waited and waited for Fermi, but it became clear it was going to be a dog so I moved, I just could not justify the cost/performance equation of Fermi. I had been using NVidia for 15/20 years - ever since they first hit the consumer market, so did not take the decision to move Vendor lightly.

I change cards every 3/4 years, and about 4 weeks ago finally took the plunge and went ATI. No real regrets, the numbers and economics speak for themselves, and were re-afirmed after the Fermi disaster. The only reason to go Fermi now is because of a particular need anyone has for CUDA - otherwise ATI have it in the bag now.

And they will do for at least 5 years. The days of "leapfrogging" the graphics lead are over for a while. NVidia are too far behind. A two year old 5870 takes out a Fermi 480, let alone the 5xxx mid life refresh due in Q3, let alone the ATI 6xxx cards with a completely new architecture due for release Spring 2011. NVidia have nothing left to respond with, apart from straw clutching and PR stunts for product releases.

Regards
Zy

..... In a way I regret my slow migration to the world of ATI because I really do like GPU Grid for example and there is no doubt that the CUDA applications are more common still ......

Been down that road as well, and in some ways still have a nagging itch about moving, I also have a real urge to help at GPUGrid due to one particular application there, ...

I too have a soft spot in my heart for GPU Grid as they were my first GPU love ... but ... the project fell into the 6 months in and we don't have to talk to the participants much any more kind of thing and well now we have applications that don't seem to run as well, the CPU usage is drifting up (again) and I don't see them actively working to make the client work better to meet their needs.

By that i mean, we should be able to have BOINC take into consideration prioritization so that if we have a GPU Grid task on hand we slot it next so that we can return it ASAP. Does not have to violate RS over time (though RS has fallen so far down the chain of things considered that I feel that it is not respected at all anymore ... can't prove it because of the lack of good logging and projects that run out of work so often ... but ...)

I was hoping DNETC would prove out, but, to this point they seem to be a disappointment also with an application that locks-up and their bland denials of the possibility ...

For me the real long term driver is total power consumption for work done ... I don't have the specific numbers because I have been running pairs of GPUs so long it is hard to say what the real earnings of specific cards are ... but, I just pulled one GTX260 out of a system which had a pair of them and it looks like my daily numbers are going to go from 70K + a day to 200K a day on the same system.

My 5870 + GTX280 system does nearly 300K a day ... I can only imagine what I would get from pairs of 5870 cards ... well, next card will let me know though I just bought a new KVM so that may have to be pushed off a month before I will know ... heck, replacing the GTX295 would double production (or more) and cut power consumption in half... well, 3 more upgrades to go ...

As to the leapfrogging, I am not so sure that it might not be possible for Nvidia to "get it" and come up with an innovation that will stun ATI ... because sure as taxes ATI is likely to get complacent, though they may own the next cycle or two as you have noted.

More food for thought...

Using BOINC Stats numbers for the last 30 days I had 24 days before I added in (again) a GTX280 card, 4 days with that card add by itself and then 2 days with the HD5870 added. The numbers are slightly biased by the fact I also added a change in OS from WIn XP-32 to Win7-64 and biased in the other direction with loss of computing because of mistakes in OS configuration, and other daily tevails documented elsewhere ...

Lastly, I am going to ignore the production of CPU work in that it is really a small percentage of the daily totals ... should demand warrant I can go back and try to get more precise but I don't really think that it matters that much ...

Normal project mix is MW, Collatz, GPU Grid (though not on all systems as the GTX295 seems unstable for those tasks, and I have run DNETC off and on in the last week or so ... all projects given equal weight of 100 RS, all projects only run on GPU, oh and I have run 4-5 SaH tasks on one system).

24 day configuration of cards was:
2 ea HD4870 (W04) i7 920
2 ea GTX260 (W03) i7 920
1 ea HD5870 (W02) i7 940
1 ea GTX295 (W01) Q9300

4 day configuration of cards was:
2 ea HD4870 (W04) i7 920
2 ea GTX260 (W03) i7 920
1 ea HD5870 (W02) i7 940
1 ea GTX280 (W02) i7 940
1 ea GTX295 (W01) Q9300

2 day configuration of cards was:
2 ea HD4870 (W04) i7 920
1 ea GTX260 (W03) i7 920
1 ea HD5870 (W03) i7 920
1 ea HD5870 (W02) i7 940
1 ea GTX280 (W02) i7 940
1 ea GTX295 (W01) Q9300

However, I feel the numbers are illustrative ...

In 24 days I was producing work and managed to generate 11,345,366 / 24 = 472,723 CS/day

In the 4 days of only adding in a GTX280 (with OS change as noted): 2,279,073 / 4 = 569,768 CS/day

Estimated improvement is 569,768 - 472,723 = 97,045 CS/Day

In the 2 days with the replacement of a GTX260 with an HD5870: 1,547,923 / 2 = 773,961 CS/day

Estimated improvement is 773,961 - 569,768 = 204,193

Based on machine specific numbers
The system with paired GTX260 cards did 65-75K per day
The system with paired HD4870 cards does 139-160K per day

I don't have great numbers for the 5870 + 280 system because it is less than a week old, but the "spot" number I have from a couple days ago is 284-294K per day (Using todays "weekly" of 3,040,255 / 7 = 434,322 CS/Day; so I am not sure what it the right number there)

If you estimate that the GTX260 does about half of the work of the total system at 72K per day or 35K, the 5870's estimated productivity is on the order of 230-240K per day...
{edit} this is borne out with todays "spot" number for that system of 256,262 from prior values of 60-75K{/edit}

This is just a rough thumb estimate ... but, I have to say it is gratifying to see my daily numbers jump from 300-550K a day to the 700-800K per day area ... only time will tell if this lasts ... if I can get the systems to run smoothly I should have fewer bad tasks and may see a slight improvement over time ... time will tell ...

I agree - there's just no contest all the way up and down the product range, ATI own it. NVidia had the opportunity with the failed GF300 development - but could not nail it - then had to use an existing GPgpu and mess with it to evolve to "Fermi" and we end up with a noisy egg-fryer.

You are right that its within NVidia's capability to pull a rabbit out the hat and jump back in - they are a good Company at heart - but the sheer realities of chip development mean thats not going to happen for 4 or 5 years. They dropped the ball big time, and first have to salvage what they can from "Fermi" before committing to "son of Fermi", sheer cost pressures dictate that, let alone fabrication which they still dont have a grip on.

ATI let NVidia into the market 15/20yrs ago because ATI got complaicent, NVidia have just returned the honours ..... (!)

I hope NVidia do pull out a rabbit in 4/5 years time, else the dynamics of competition die out, and we all lose big time with inflated price performance, and increased development cycles with no viable competition.

The big unknown is the move to combine cpu & gpu onto one chip (again - what goes around comes around!), and Intel are coming into play on the sidelines because of it. This migration back to one chip could well sink NVidia in the long term at consumer level at least, if not all levels, as it has no inherent cpu base.

My money is on the main long term term market being an ATI/AMD v Intel fight in 5 years going forward, with the two of them fighting over going back to one chip - a combined cpu/gpu. NVidia will still be there .... maybe .... but it would be the last throw at consumer level before they abandone consumer cards and go back to CUDA compute level only.

Regards
Zy

I hope NVidia do pull out a rabbit in 4/5 years time, else the dynamics of competition die out, and we all lose big time with inflated price performance, and increased development cycles with no viable competition.

The big unknown is the move to combine cpu & gpu onto one chip (again - what goes around comes around!), and Intel are coming into play on the sidelines because of it. This migration back to one chip could well sink NVidia in the long term at consumer level at least, if not all levels, as it has no inherent cpu base.

My money is on the main long term term market being an ATI/AMD v Intel fight in 5 years going forward, with the two of them fighting over going back to one chip - a combined cpu/gpu. NVidia will still be there .... maybe .... but it would be the last throw at consumer level before they abandone consumer cards and go back to CUDA compute level only.

I am not sure that on chip graphics are going to be a real big thing ... of course I have been wrong before... I suspect that there might be some on chip solutions but that the eternal market is not going to go away because the on-chip is likely to be a low end "common" solution that is just not going to satisfy the high end gamers ...

Especially because to upgrade the system the investment is now for a much larger and more expensive chip instead of a less costly drop in card for the same performance increase ... of course, for those of us that are interested in GPUs ... this would not be all bad... on chip GPU along with add on cards ... more performance ... :)

Still no news of the cards GTX470 and GTX480 will work soon.
I get no replies on any messages, why?

The new application MW3 works OK for these cards.

Possible to choose in preferences? All other jobs will cancel!

Any reaction, please??

Thx,

Ton (ftpd) - Netherlands

Still no news of the cards GTX470 and GTX480 will work soon.
I get no replies on any messages, why?

Um, because they are hard at work? :)

Seriously, the card just came out ... and all the projects are struggling with the changes needed. GPU Grid is likewise working on making their application work with the card. The sad truth is that living on the "bleeding edge" means suffering ... and in this case that suffering is that the applications are lagging ...

Since most users do not have this card, the updates to make it work are, as they should be, a lower priority than getting other things done ... like on MW getting the new application fielded ... which for you has the side benefit of allowing your card to work ...

Patience is required ...

I accept "bleeding", and thx for the quick answer, as it should!!

I also did some solutions!?

Ton (ftpd)

I accept "bleeding", and thx for the quick answer, as it should!!

I also did some solutions!?

Ton (ftpd)

The good news is that GPU Grid may be getting a handle on the new card though I do not know when they will be issuing work for the new application. There is a new post by GDF there that discusses the matter ...

I have donated the two fermi-cards!

Ton

Paul - saw your post - FYI I have 2 machines with 2 5870 cards - I am getting about 450-460k daily in MW and about 350-370k in Collatz (when MW is down).

Paul - saw your post - FYI I have 2 machines with 2 5870 cards - I am getting about 450-460k daily in MW and about 350-370k in Collatz (when MW is down).

Ah thanks ... :)

The only "solid" number I have is for the Dual 4870 system (~139K) and for the 5870 & GTX260 system (~290K) ... so that is good news ... I hope when I replace the GTX260 I will see about 150-200K boost in my totals...

Schedule says the card is here Friday ... so should be in and working something Friday and better data should come next week ...

I have noticed that the RAC in BOINC Stats is pointing up like a rocket from the last exchange of a 5870 for a GTX260 and I went from what looks like a 450K average to 600-650K a day average ... hard to know when the data bounces around so much ... I saved a copy of teh 60 day history in BOINC Stats and plan to save several more over the next couple of weeks to see when I go ...

I don't think I will get close to 1M a day until I get two more 5870s going ... and that is going to take all summer ...

But thanks for the data ... :)

{Edit-add}
Just looked at the number for the two systems and they are more "reasonable" as in more of what I would expect. The early numbers were showing that the system with the GTX280 was far and away out performing the one with the GTX260 (both have a HD5870) but now with one week's data the weekly number for the 280 equipped system is actually showing a 222K daily production which is close to the 200K+ showing on the 260 equipped system ...

Youi can see here the two systems, both are running WIn7-64, the 920 is 5870 & 260 and the 940 is the 5870 and 280 ...

The other i7 920 has the dual 4870 cards ...

I am getting about the same as Farscape, albeit on a 5970 - its real world comparible to 2 x 5870, just on one card as its the same 2 GPUs.

Regards
Zy

Paul - I think you will really enjoy the ATI cards - they operate smoothly and crank! The only time I sing the blues is when MW goes down and Collatz gets overloaded.....I have a few nice CPUs but not enough....

Paul - I think you will really enjoy the ATI cards - they operate smoothly and crank! The only time I sing the blues is when MW goes down and Collatz gets overloaded.....I have a few nice CPUs but not enough....

Oh, my yes, I do love them ... as I just noted elsewhere I am 5 and 2 Friday when the next 5870 shows up on my doorstep ...

With the latest card I will hope to be doing about 700K consistently and with luck maybe a little more ... I want to be careful not to poke the beast too much lest he retaliate and murphy strike ...

Paul - I think you will really enjoy the ATI cards - they operate smoothly and crank! The only time I sing the blues is when MW goes down and Collatz gets overloaded.....I have a few nice CPUs but not enough....

I am a recent convert (a month) after 15 yrs with NVidia having seen results like you and others get, never thought I would ever go ATI, but there ya go, times are a changing :)

Have a peak at DNETC, you can run it with the same setting as MW and "pay's" well, so would be a good b/up if you get caught in the future the way you describe re MW & Collatz

Regards
Zy

Paul - I think you will really enjoy the ATI cards - they operate smoothly and crank! The only time I sing the blues is when MW goes down and Collatz gets overloaded.....I have a few nice CPUs but not enough....

I am a recent convert (a month) after 15 yrs with NVidia having seen results like you and others get, never thought I would ever go ATI, but there ya go, times are a changing :)

Have a peak at DNETC, you can run it with the same setting as MW and "pay's" well, so would be a good b/up if you get caught in the future the way you describe re MW & Collatz

Regards
Zy

Zy,

I am 931,844 at DNETC ... I had a run of failures but now seem to be running them successfully (again, for now) ... so, yes I am aware ... :)

The one thing that bothers me is that they are like Aqua, they want to use all the resources and BOINC does not do the hand over well from running 8 applications to running one on all 8 cores ... it has a bad habit of running on idle for fairly long stretches ... the same with DNETC wanting to run on all the GPUs at the same time...

Personally, I prefer to run more diverse workloads because on HT that improves throughput ... running the same code on all 8 virtual processors means that you get less from HT than if your workload is more diverse ... that is one of the reasons I don't run Aqua that much and when I have run DNETC, it is on systems where I have mixed cards so I don't have it monopolize my GPUs ...

Besides, they have a nasty bug where the cards / tasks get snarfled...

.... I am 931,844 at DNETC ... I had a run of failures but now seem to be running them successfully (again, for now) ... so, yes I am aware ... :)
..........
Besides, they have a nasty bug where the cards / tasks get snarfled...

I know you are aware of it, the suggestion was for Farscape.

You are on my "guru to ask list", no way do I try the reverse rofl, you're way above my league :)

I hate that bug ..... but at least the error trap catches it now and mitigates the old problem of being blocked out for hours.

Regards
Zy

What has all this to do with gtx470 or gtx480?

Make another thread?

Ton (ftpd)

What has all this to do with gtx470 or gtx480?

Make another thread?

Ton (ftpd)