Associate
- Joined
- 10 Mar 2006
- Posts
- 337
Just thought i'd post to let people know that OcUk are now advertising X1950 pro for AGP, could be a cheap upgrade/performance boost for all those people folding on AGP systems.
X1950 Pro AGP at OcUK!!!!!!!!!!!!!
- Thread starter br83taylor
- Start date
Anybody care to find out?VeNT said:
I imagine the X1950 Pro for either slot won't work properly as it only has 36 shaders instead of the 48 that FAH wants.
- Joined
- 10 Mar 2006
- Posts
- 337
well the PCI-E x1900gt only has 36 shaders and that works, so i can't see that as an issue.
all reviews i have seen that compare AGP vs PCI-E show next to no benifit of all the increased bandwidth in games so i can't see that as a problem either.
Soldato
Who's they?lay-z-boy said:
People have already tried out GPU folding on AGP 1600 series cards and it seemed to work - however it was very slow and it hasn't been proven that the returned results are actually useable (though I don't see why they wouldn't be personally). Frame times were around 25mins which with the current points system would give 190PPD ( at the time it was only giving about 125PPD so was quite pointless
)Of course we will have to wait for someone to test it out and tell us what sort of performance the X1950 Pro AGP will give - it may not be officially supported by the GPU beta test but if it's any good it will likely be phased in along with the X1800 PCI-E cards once the initial testing is done
edit: linky - note that running beta work on hardware it's not really meant for won't be helping the progress one bit
Last edited:
rich99million said:linky - note that running beta work on hardware it's not really meant for won't be helping the progress one bit
Reading through that thread; there is a guy who has had it running on a 1600 AGP so there should be no problems running it on a 1950 AGP - that's my take on things, anyway.
Stan
How 'bout we take up a collection and get one for my dual Xeon rig and we'll do our own tests and verification.Bigstan said:
br83taylor said:
I suspect the issue isn't so much the reduced bandwidth, as the nature of it. AGP is a half-duplex interface so data can only travel in one direction at the same time - a problem resolved in PCI-E. I imagine it's this capability which Folding takes advantage of, which would explain why it's so slow on AGP systems.
rich99million said:Who's they?
Erm, who else but stanford....
Dont try to tell me that x1600's work etc. im without any doubt that it will work on any x1xx0 series, just much, much slower than x1900.
It isnt good for ppd or the science so its pointless, if they do decide to do another core or something that is designed for slower gpu's then that would be a good idea, till then putting folding on anything lower than an x1900 is like giving a qmd to a celeron tualatin.
- Joined
- 10 Mar 2006
- Posts
- 337
i don't think the slow 25 mins per frame for an AGP 1600 card is due to the AGP interface. current times show that its much more dependant on the card type, eg.
X1900XT PCI-E = 8 mins
X1900GT PCI-E = 12 mins
X1800XT PCI-E = 18 mins
1600 series AGP = 25 mins
to me the time for the 1600 looks about right just based on the card type and i'd exect a 1600 PCI-E to be about the same.
Soldato
lay-z-boy said:
to me it sounds like 'dont use anythong other than an x1900 because its too slow'. Let's change that around a little.... 'dont use anything over than a conroe e6700 because everything else is too slow'
I didnt realise Folding was for the elite
silly people, those that get all their lesser systems to fold... my point is that does it really matter if its slower? its still folding, isnt it?
Last edited:
I think that's a tad harsh. The GPU client is being developed with the 1900 series cards in mind, not the earlier ones. From what we've seen so far, running it on a lesser card doesn't show a real return, on the work done per effort, since the GPU client still ties up a CPU core.
There may still be some benefits dependent on you actual CPU and GPU combination, i.e. if you have say a 3500 Athlon with a 1600 card, what might happen then?
Having said that, from what I can see at the moment the 1900 is the optimum card all round. the 1950XT PCI-e card seems to crunch at the same pace as the 1900XT cards, whcih obviously cost less, at least that's waht I see on my system.
There may still be some benefits dependent on you actual CPU and GPU combination, i.e. if you have say a 3500 Athlon with a 1600 card, what might happen then?
Having said that, from what I can see at the moment the 1900 is the optimum card all round. the 1950XT PCI-e card seems to crunch at the same pace as the 1900XT cards, whcih obviously cost less, at least that's waht I see on my system.
james.miller said:to me it sounds like 'dont use anythong other than an x1900 because its too slow'. Let's change that around a little.... 'dont use anything over than a conroe e6700 because everything else is too slow'
I didnt realise Folding was for the elite
Was that ment to be a dig or not?
If it was, my point is that it would be great to allow lesser gpu's to fold as well BUT it would probably be soo slow the energy used up and effort needed on both the client designers side and ours isnt worth it. (sort of like its better to have 2-3 conroe's than 150 pentium 3's) if you understand me.
No, stanford dont require the latest and greatest but its damm good to have it as its much faster than anything else. (gpu wise)
Converting the phrase i said into the conroe one you have is not the same.
The x1900 is a speed freak compaired to the other x1k's, the x1800, which is still very fast is much slower than the x1900.
Conroe clock for clock, may be faster than anything out there currently but its not like it a pebble vs a huge rock, in performance terms, (against say, an x2)
Catering for slower gpu's as of now isnt on their top priority, i hope it does become one soon, i for one, will be willing to run it on anything possible.
Whats with this 'elite' business?, its not that, i would say its more of a 'min requirements for decent crunching speed'
Have not seen you post round here before, sorry if you do, its just seems to be the first time ive seen your name in this area.
Soldato
Stanford want their results as soon as possible so I don't see a problem with them starting testing with the fastest currently capable card.
I hardly think it is elitism to say "anything under an X1900 is too slow" especially since graphics cards can be power hungry. So not only are lower spec cards doing less work, they're also taking more power to do it.
I suspect Stanford may include the X1800 in the future but the way I see it, what's the point of including old cards. In only a short time there is going to be another generation of graphics cards showing up and in years time the X1900 will probably be considered "old".
SiriusB
I hardly think it is elitism to say "anything under an X1900 is too slow" especially since graphics cards can be power hungry. So not only are lower spec cards doing less work, they're also taking more power to do it.
I suspect Stanford may include the X1800 in the future but the way I see it, what's the point of including old cards. In only a short time there is going to be another generation of graphics cards showing up and in years time the X1900 will probably be considered "old".
SiriusB
Soldato
lay-z-boy said:Was that ment to be a dig or not?
If it was, my point is that it would be great to allow lesser gpu's to fold as well BUT it would probably be soo slow the energy used up and effort needed on both the client designers side and ours isnt worth it. (sort of like its better to have 2-3 conroe's than 150 pentium 3's) if you understand me.
No, stanford dont require the latest and greatest but its damm good to have it as its much faster than anything else. (gpu wise)
Converting the phrase i said into the conroe one you have is not the same.
The x1900 is a speed freak compaired to the other x1k's, the x1800, which is still very fast is much slower than the x1900.
Conroe clock for clock, may be faster than anything out there currently but its not like it a pebble vs a huge rock, in performance terms, (against say, an x2)
Catering for slower gpu's as of now isnt on their top priority, i hope it does become one soon, i for one, will be willing to run it on anything possible.
Whats with this 'elite' business?, its not that, i would say its more of a 'min requirements for decent crunching speed'
Have not seen you post round here before, sorry if you do, its just seems to be the first time ive seen your name in this area.
no, no, that wasnt a dig. it sounds far worse than intended, sorry lol
im quite interesed in what they can do with gpu folding, it's been on the cards for quite a while now. I have a question though - whats the relationship between gpu folder and cpu usage? at what point would it become more economical to fold purely on the cpu alone?exactly my point, stanford are drooling over the idea of the r600.
Is it going to unfair if it runs the fastest on that?
I think not tbh.
Even top end hardware gets outdated quick, it seems to be happening faster and faster each year round though....
Is it going to unfair if it runs the fastest on that?
I think not tbh.
Even top end hardware gets outdated quick, it seems to be happening faster and faster each year round though....
Actually, at 25 mins per frame a 1600 would give you a ppd of approximately 190 and at 18 mins, an X1800 would give you about 264.
With the majority of CPU WUs at the moment being the 149 and 44 pointers, most machines would be better off running the GPU client on a 1600 and certainly on an 1800.
My 4000+ overclocked to 3.2GHz only gets about 190 ppd on 149s and considerably less on 44s.
Even the two C2D rigs only get 240ish for 44s and 260ish for 149s.
The other rigs get a lot less.
Stan
/Eyes up the X1800 and X1600 with a glint in his eye
With the majority of CPU WUs at the moment being the 149 and 44 pointers, most machines would be better off running the GPU client on a 1600 and certainly on an 1800.
My 4000+ overclocked to 3.2GHz only gets about 190 ppd on 149s and considerably less on 44s.
Even the two C2D rigs only get 240ish for 44s and 260ish for 149s.
The other rigs get a lot less.
Stan
/Eyes up the X1800 and X1600 with a glint in his eye
Soldato
james.miller said:no, no, that wasnt a dig. it sounds far worse than intended, sorry lol
With the current Beta client each GPU client will take approximately the whole CPU in a single core machine, and a whole core in a dual/quad core machine. This is due to the CPU polling the GPU to see if it is done.
I haven't seen any figures that show what impact slower CPUs have over folding on a GPU - ie how slow does a CPU have to be to show a marked impact on the GPUs performance.
This may actually be a moot point anyway, as one of Stanfords goals is to try and reduce the amount of polling done by the CPU. Which means if they reduce it by a fair bit you would need an incredibly slow CPU to affect the GPU.
EDIT: Stan, no more crunching, I suspect you're single-handedly increasing global temperatures with your farm!
SiriusB
james.miller said:no, no, that wasnt a dig. it sounds far worse than intended, sorry lol
Well if my conroe E6600 @ a paltry 2.6GHz can do a 600 pointer gromac in around 30 hours, that give around 20 points per hour. The GPU WUs are worth around 330 points, that equates to only 22.5 points per hour. On that basis if points represents the true value of the science, running the GPU client is very marginal even in a high spec machine.