Message boards :
Number crunching :
Future ATI Card Support?
Message board moderation
Previous · 1 · 2 · 3 · 4 · Next
Author | Message |
---|---|
Send message Joined: 16 Feb 09 Posts: 109 Credit: 11,089,510 RAC: 0 |
Has the project team departed from their original idea to do longer, more science intensive work with the GPU application? |
Send message Joined: 8 Jan 09 Posts: 58 Credit: 53,161,741 RAC: 0 |
1000x bigger ? well calc'ed in SP with a ATI Radeon 4830 would take about 85min; on a 4870x2 about 31min - but on nVIDIA GPUs about 2h or more |
Send message Joined: 24 Dec 07 Posts: 1947 Credit: 240,884,648 RAC: 0 |
1000x bigger ? well calc'ed in SP with a ATI Radeon 4830 would take about 85min; on a 4870x2 about 31min - but on nVIDIA GPUs about 2h or more Perfect then... |
Send message Joined: 8 Jan 09 Posts: 58 Credit: 53,161,741 RAC: 0 |
but this kind of WU would be a 4000 to 8000 Credit WU (based on current Credits/WU : 2 ) |
Send message Joined: 21 Aug 08 Posts: 625 Credit: 558,425 RAC: 0 |
Mostly true, yes, but you live in a part of the world that is more socialist than capitalist. Many of the people I'm thinking of will go with the fundamental principles of a Capitalist economy, which is they'll go to wherever they can maximize their profit. Bottom line is that it would be best if GPUs can't play in the CPU sandbox and CPUs can't play in the GPU sandbox... This is exactly what Travis said in the March 25th news post: "After testing to make sure that it's working correctly we'll swap to only awarding credit to the GPU applications for GPU milkyway (that way our servers wont be as bogged down, which will mean better work availability)." |
Send message Joined: 9 Feb 09 Posts: 166 Credit: 27,520,813 RAC: 0 |
Did you copy the aticalx.dll files and rename them to amdcalx.dll? Yes i have found its the can't write error so not sure if i can solve this since the fix seem only to work with 8.12 driver. 1000 x bigger I would not have a problem with 100x bigger units because my cpu would crunch them nicely ;) But i guess the slower machines out there would have a stroke. Although i wonder if there are many slower machines here on MW other then the big network guys using their bosses systems ;) ( I was warned to get fired if i did ... ) Its new, its relative fast... my new bicycle |
Send message Joined: 24 Dec 07 Posts: 1947 Credit: 240,884,648 RAC: 0 |
1000 x bigger I would not have a problem with 100x bigger units because my cpu would crunch them nicely ;) 1000x bigger for the GPU only wu's... |
Send message Joined: 12 Mar 09 Posts: 61 Credit: 9,214,340 RAC: 0 |
Mostly true, yes, but you live in a part of the world that is more socialist than capitalist. Many of the people I'm thinking of will go with the fundamental principles of a Capitalist economy, which is they'll go to wherever they can maximize their profit.I have two cards that can do more than 140,000credits per day and i get about 90,000 actually if these cards could get 120,000 credits with MWgpu what would you do? Really stay with MW nongpu for capitalistic reasons??? Another point is that i have to babysit MW a lot and i would love to see WUs that run a few hours for the price of little more real credit against a lot more possible but not available credit - a good capitalistic deal i think - maximum profit for minimum effort |
Send message Joined: 21 Aug 08 Posts: 625 Credit: 558,425 RAC: 0 |
My computer, an AMD Athlon64 3700+ that is overclocked to basically the equivalent of a FX-57, which is the fastest non-HyperThreaded single-core system, takes around 30 minutes for the larger tasks we have now, 15-20 for the smaller. 30 * 100 = 3000 minutes / 60 = 50 hours. Deadline is at 72 hours. I would have to run the computer with BOINC running for about 70% of the deadline for the task to complete within deadline. The project has already said that 3 days is a bit long to wait for a task given the way the search is built upon the results of prior tasks, but they compromised and bumped up from 2 days to 3 days. Given all of this, increasing the complexity of the current tasks to levels of 100x the current level would be bad for the project from both a scientific and a Public Relations standpoint, as anything older than approximately a 3GHz P4 or an Athlon64 3400+ would not be able to make deadline on a consistent basis if they weren't completely dedicated systems to doing BOINC only and were not powered off. At present, based on available stats from the stat sites, that would eliminate at least 5,000 hosts, if not closer to 7,000 or 8,000 hosts...or about 15-20% of the total number of hosts attached to the project. What was done was the best decision based on the available data...assuming it works... If it doesn't work, then I'm sure we'll hear from the ATI camp loudly... |
Send message Joined: 21 Aug 08 Posts: 625 Credit: 558,425 RAC: 0 |
Mostly true, yes, but you live in a part of the world that is more socialist than capitalist. Many of the people I'm thinking of will go with the fundamental principles of a Capitalist economy, which is they'll go to wherever they can maximize their profit.I have two cards that can do more than 140,000credits per day and i get about 90,000 actually If you were able to get 120,000 / day with MW_GPU, but were able to get 125,000/day with MW "Classic" (CPU) utilizing the current application, would you switch? Yes, it's a hypothetical question for a situation that currently does not seem to exist for you, but if the load was lightened up enough by people with CUDA cards going over to the GPU project that it could happen, would you switch? Do you think everyone would switch? Do you think that 50% would switch? 33%? 25? 10? In theory, this situation should not happen based on the concept of being given the same amount of credit as here. Personally, based on the amount of continual complaining that's gone on here, I wouldn't mind it in the slightest if they increased the amount of credit given to GPUs running the GPU project to totally incentivize a move to the other project so that everyone can enjoy some peace and quiet. |
Send message Joined: 12 Mar 09 Posts: 61 Credit: 9,214,340 RAC: 0 |
If the situation is lightened where is your problem then? |
Send message Joined: 21 Aug 08 Posts: 625 Credit: 558,425 RAC: 0 |
If the situation is lightened where is your problem then? I apparently have to enable "maximum verbosity" mode... :sigh: As it stands, you folks with the GPU cards are capping the server. You've even convinced someone who doesn't have a GPU card that the project is putting artificial limits in on how many tasks are created. If the people with CUDA cards who don't have ATI cards leave, it will lighten the load, but nobody should be expecting the problems here to completely go away by just CUDA being supported. What will happen is it will become *SLIGHTLY* better. This will appease some people temporarily, but not for very long. If then ATI support is added to the GPU project, and enough people head over there, but a few "industrious" people figure out that they can still participate here and, as fate would have it, they could get more credits here, those "industrious" people would surely be right back here hammering away at the server. Eventually more and more of the competitive types would figure this out and we'd have issues here again. Just be happy that you're getting your own sandbox, ok? Like I said, I don't even care if they want to increase the amount of credit they give you folks if it means you move over there... |
Send message Joined: 2 Aug 08 Posts: 24 Credit: 374,440,641 RAC: 0 |
Ok guys. The Code for the CUDA app is out. Who is able to help Travis. Look into the Code Applcation Thread |
Send message Joined: 24 Dec 07 Posts: 1947 Credit: 240,884,648 RAC: 0 |
Mostly true, yes, but you live in a part of the world that is more socialist than capitalist. Many of the people I'm thinking of will go with the fundamental principles of a Capitalist economy, which is they'll go to wherever they can maximize their profit.I have two cards that can do more than 140,000credits per day and i get about 90,000 actually Yes, but if in theory you could get 125,000 by running the cpu wu's on your gpu's but due to wu limitations at the cpu project you could only get 70,000 due to a wu shortage, but could get 120,000 by running the gpu wu's and have no wu shortages then I know what I would run! |
Send message Joined: 26 Jan 09 Posts: 589 Credit: 497,834,261 RAC: 0 |
|
Send message Joined: 21 Aug 08 Posts: 625 Credit: 558,425 RAC: 0 |
I know what I would run! Travis has stated that they intend on not having the GPU apps get credit here once they are up and running over there... |
Send message Joined: 12 Apr 08 Posts: 621 Credit: 161,934,067 RAC: 0 |
I know what I would run! And it could be as simple as having the validator mark as invalid any tasks that are completed too quickly. Abnormally low run time = GPU processing. |
Send message Joined: 4 Oct 08 Posts: 1734 Credit: 64,228,409 RAC: 0 |
And it could be as simple as having the validator mark as invalid any tasks that are completed too quickly. Abnormally low run time = GPU processing. I think that may be too simple when the AMD 955 and Intel i7s and Penryn class CPUs are concerned. Overclocked versions usually finish a WU of the ps_s23_17_ type, giving credit of 28.63 CS, in between 500 and 620 seconds. Would this speed be classified as a GPU? Also, if the output is used as another measure, a quad Penryn class PC will achieve RACs between 15K and 16K. Too close to the HD3850 GPU outputs and speeds for me to be comfortable and for the classification I presume the GPU client will give it away? Go away, I was asleep |
Send message Joined: 12 Apr 08 Posts: 621 Credit: 161,934,067 RAC: 0 |
And it could be as simple as having the validator mark as invalid any tasks that are completed too quickly. Abnormally low run time = GPU processing. Since I am, and never have been, a fan of over-clocking when the point is scientific accuracy, would not bother me to have them classed so ... HOwever, The GPU processing times were down in the double and single digits of seconds. So, even a discriminator of 100 seconds would still be plenty of leeway... In the same line, I am not a fan of the scriptors who in the pursuit of their own ego run scripts to pound on the server, when the main problem is server load ... in that light I am sure that there are going to be those that work diligently to fuel their own egos at the expense of the project's welfare and health. So, I am sure that someone is going to spend enormous amounts of effort to make a GPU application and framework so that they can pull more work from here instead of working with the needs of the project. |
Send message Joined: 12 Mar 09 Posts: 61 Credit: 9,214,340 RAC: 0 |
Travis has stated that they intend on not having the GPU apps get credit here once they are up and running over there... I know how my gpus look like a slightly overclocked dual xeon W5580 >:D |
©2024 Astroinformatics Group