Message boards :
Number crunching :
CPU/GPU Comparison (do we need CPU apps when GPU app is available)
Message board moderation
Previous · 1 · 2
Author | Message |
---|---|
Send message Joined: 22 Jan 11 Posts: 375 Credit: 64,688,465 RAC: 1,873 |
No big deal :), their are a couple of guys on my team who were beta testers of SETI classic, IIRC their reg date was April '99! Team AnandTech - SETI@H, DPAD, F@H, MW@H, A@H, LHC, POGS, R@H, Einstein@H, DHEP, WCG Main rig - Ryzen 5 3600, MSI B450 G.Pro C. AC, RTX 3060Ti 8GB, 32GB DDR4 3200, Win 10 64bit 2nd rig - i7 4930k @4.1 GHz, HD 7870 XT 3GB(DS), 16GB DDR3 1866, Win7 |
Send message Joined: 30 Apr 14 Posts: 67 Credit: 160,674,488 RAC: 0 |
I have a little over 200days dedicated to Seti@home classic. When Seti switched to BOINC platform I stopped computing for several years (about 6), mostly because my laptop at the time was insufficient for such a task. |
Send message Joined: 8 May 09 Posts: 3339 Credit: 524,010,781 RAC: 0 |
I have a little over 200days dedicated to Seti@home classic. I am glad you returned to crunching!! |
Send message Joined: 25 Mar 09 Posts: 65 Credit: 53,099,671 RAC: 0 |
This post can be deleted as soon as read. I have through the years accumulated about 300 bucks worth of "beer cans" most of which I did drank myself... I have an older AMD Radeon HD 4800 series. Used to do awesome, now not so much anymore. 125K/day now maybe 30K. What can I buy as far as card that can get me back to it or above? you can reply to this thread or if deleted just post me a rough answer please. Martin Chartrand |
Send message Joined: 8 May 09 Posts: 3339 Credit: 524,010,781 RAC: 0 |
This post can be deleted as soon as read. I have been buying Nvidia 760 gpu's lately but ONLY those with dual or even triple fans on them. All the newer gpu's use the pci-e slot and are the version 3 type, but they also work just fine in the old standard version 2 slots. The 760 is a dual precision type card so will work here or almost any other project. I have been spending about 200 US dollars per card on them thru the Amazon warehouse: http://www.amazon.com/gp/node/index.html?ie=UTF8&merchant=&node=1267877011 |
Send message Joined: 30 Apr 14 Posts: 67 Credit: 160,674,488 RAC: 0 |
Radeon r9 280x do have best double precision available per watt. For milkyway it is the best. |
Send message Joined: 8 May 09 Posts: 3339 Credit: 524,010,781 RAC: 0 |
Radeon r9 280x do have best double precision available per watt. For milkyway it is the best. But the R series of AMD gpu's does have some problems at some projects. YES it is faster then the 760, by a long ways, but if it won't work then it isn't helpful. A list of projects the OP is interested in might help. For comparison I made a chart to help me in my choices when buying new gpu's. The AMD and Nvidia numbers are NOT comparable between them, only among themselves: card type-MW capable shaders Nvidia 8600 GT- no 128 Nvidia 9800 GT/512meg- no 112 Nvidia 9600 GSO/ 512meg- no 64 Nvidia GTX 560Ti-yes 384 Nvidia 650Ti-yes 768 Nvidia 660-yes 1344 Nvidia 670- yes 1344 Nvidia 680-yes 1536 Nvidia 690-yes 3072 Nvidia 750-yes 512 Nvidia 750Ti-yes 640 Nvidia 760-yes 1152 Nvidia 770-yes 1536 Nvidia 780-yes 2880 Nvidia 790-yes 3072 ATI 4650 AGP-no 320 ATI 4670-no 320 ATI 5550-no 320 AMD 5770-no 800 AMD 5870-yes 1600 AMD 6770-no 800 AMD 6850-no 960 AMD 6870-no 1120 AMD 6950-yes 1408 AMD 6970-yes 1536 AMD 7750-yes 512 AMD 7770-yes 640 AMD 7790-yes 896 AMD 7850-yes 1024 AMD 7870-yes 1280 AMD 7950-yes 1792 AMD 7970-yes 2048 AMD 7990-yes 4096 AMD R7 260x 896 AMD R9 270x 1280 AMD R9 280x 2048 AMD R9 290x 2816 The first column is the type of card, the yes/no means dual precision, the last column is the shaders. Nvidia does NOT call them that and as I said comparing AMD shaders numbers to Nvidia shaders numbers is NOT a valid comparison!! It simply shows for instance that an AMD 7750 has fewer shaders then an AMD R9 290x, and what the difference is. |
Send message Joined: 30 Apr 14 Posts: 67 Credit: 160,674,488 RAC: 0 |
http://en.wikipedia.org/wiki/List_of_AMD_graphics_processing_units Looking for "Double precision" gives GFLOPS value for double precision calculations. R280X have up to 1024 double precision GFLOPS. GeForce GTX 760 Ti have about 100 GFLOPS (10 times less)... http://en.wikipedia.org/wiki/List_of_Nvidia_graphics_processing_units GeForce GTX Titan is 1.5 times better than R280X in double precision, but at 3 times the cost. |
Send message Joined: 8 May 09 Posts: 3339 Credit: 524,010,781 RAC: 0 |
http://en.wikipedia.org/wiki/List_of_AMD_graphics_processing_units And a Titan costs $999.99, while a 280x costs $299.99, while a 760 costs only about 200 bucks, all at Newegg. If I had about a billion bucks I would build a Quantum machine and not worry about costs, but in the real world that isn't a possibility. |
Send message Joined: 30 Apr 14 Posts: 67 Credit: 160,674,488 RAC: 0 |
http://en.wikipedia.org/wiki/List_of_AMD_graphics_processing_units GeForce GTX 760 gives 94 GFLOPS. As already said, 10 times less for 99$ less. My cards were cheaper than $299.99 since I bought second hand cards. I payed 250$. If one needs a new card: Radeon R9 270 will be cheaper and will have better single precision and double precision GFLOPS than 760. Some projects also have problems with multiple NVidia cards. However even I thought about buying a single nVidia card, since some projects do not support OpenCL (yet!). I indeed would buy 760, 770 or used 780, but I would only use it in single precision computing. Right now I'm waiting for OpenCL application for Asteroids@Home :) |
Send message Joined: 22 Jan 11 Posts: 375 Credit: 64,688,465 RAC: 1,873 |
http://en.wikipedia.org/wiki/List_of_AMD_graphics_processing_units Under sadly do very poorly in MW, see my 1st benchmarking thread. Team AnandTech - SETI@H, DPAD, F@H, MW@H, A@H, LHC, POGS, R@H, Einstein@H, DHEP, WCG Main rig - Ryzen 5 3600, MSI B450 G.Pro C. AC, RTX 3060Ti 8GB, 32GB DDR4 3200, Win 10 64bit 2nd rig - i7 4930k @4.1 GHz, HD 7870 XT 3GB(DS), 16GB DDR3 1866, Win7 |
Send message Joined: 30 Apr 14 Posts: 67 Credit: 160,674,488 RAC: 0 |
Yeah. I wonder if GTX Titan can then do 5 tasks at the same time without crashing :) (it was stated that only 18% GPU utilization was seen). Still... I bought R280X strictly because it have best double precision GFLOPS/W. Seems I wasn't wrong :) |
Send message Joined: 22 Jan 11 Posts: 375 Credit: 64,688,465 RAC: 1,873 |
Don't know, but IIRC the guy with the titan ran 4 tasks at once. Btw, I see I missed out a word earlier (took me a while to work out what I said, lol, though you seem to get it), anyway, seeing as I can't edit here after 1hr, I meant to say. Under sadly, Titans do very poorly in MW, see my 1st benchmarking thread. Team AnandTech - SETI@H, DPAD, F@H, MW@H, A@H, LHC, POGS, R@H, Einstein@H, DHEP, WCG Main rig - Ryzen 5 3600, MSI B450 G.Pro C. AC, RTX 3060Ti 8GB, 32GB DDR4 3200, Win 10 64bit 2nd rig - i7 4930k @4.1 GHz, HD 7870 XT 3GB(DS), 16GB DDR3 1866, Win7 |
Send message Joined: 31 Mar 08 Posts: 4 Credit: 57,065,095 RAC: 0 |
Banning CPU (not PC with certain CPU), is not preventing the choice. From my personal viewpoint, it is. Your conception of freedom is probably different from mine... |
Send message Joined: 2 Jul 14 Posts: 15 Credit: 20,991,384 RAC: 0 |
Even with GPU apps, CPUs are the most common and hardworking. Aside from what mikey said about GPU vs CPU, older computers can't use onboard graphics, and newer computers still have slow onboard chips. Even though GPUs are fast, removing CPU apps would result in a lot less work coming in. That and the fact that the only people running the project would be 'rich' people with good cards. |
Send message Joined: 2 Jul 14 Posts: 15 Credit: 20,991,384 RAC: 0 |
but you are talking about alienating your user base, ie your work force. And that is most likely not a viable option for a volunteer project. Exactly! BOINC Projects are all volunteer based. You can't pick and choose with something like this. |
Send message Joined: 16 Dec 13 Posts: 23 Credit: 48,851,144 RAC: 388 |
Older hardware will eventually get phased out as updating applications for them becomes too time and resource-consuming compared to their contributions to the project. There are only a handful of projects that are still around which can crunch on PowerPC processors, MilkyWay being one of them. I still have a handful of those machines (a bit of a hobby) and my "workbench" machine, a Dual G5, still completes work though mostly for Moo!, Enigma and SIMAP because MilkyWay tasks are usually too long to complete on time within my usage pattern. Similar processor restrictions exist for some other projects; BURP requires a 64-bit processor, others require specific instruction sets such as SSE3, etc. You might quote statistics about cost versus efficiency, but they are moot when I would be using the machine anyway; in my case I only have one host that is on 24/7, the rest when I'm in the office or actively using them. I absolutely agree with the "anything helps" philosophy, and it's my understanding that it's one of the reasons BOINC and grid computing in general was developed. I understand that supporting older hosts can become too difficult for the benefit, but I appreciate it when projects take the extra time and effort to do so. You may not see their contribution as significant, but it matters to people like me, and "contribution" can be pretty subjective depending on a project's function and goals. Click here to see My Detailed BOINC Stats |
Send message Joined: 30 Aug 07 Posts: 2046 Credit: 26,480 RAC: 0 |
So just for someone from the project to chime in. This project is actually interesting in that the results which are returned slower actually do provide some benefit to what the project is doing as a whole. Basically, we're running evolutionary algorithms to do parameter optimization. This means that we're trying to find a set of parameters which will make the astronomical models (either the n-body simulation or the steam fitting project) match the stars we see in the sky. Running an evolutionary algorithm is basically a balancing act between exploration (we want to try new sets of parameter in regions of all possible parameters that we haven't tried before) and exploitation (we want to refine parameter sets in regions where we know there are good solutions). What can happen, is if there is too much exploitation the search ends up converging to a region of pretty good solutions, but not the overall best solutions. The results that are returned slower actually kind of act as a balance against this, allowing better exploration of the search space. We actually analyzed this a bit in one of our papers (An Asynchronous Hybrid Genetic-Simplex Search for Modeling the Milky Way Galaxy using Volunteer Computing ) if you want to read it. At any rate, even though the CPUs are chugging a long a lot slower than the GPUs, their results actually are still useful. Of course, if the CPUs all went away, we could tweak the parameters to how our evolutionary algorithms work to allow for more exploration, but having them around does make our evolutionary algorithms more robust against premature convergence to local optima. |
©2024 Astroinformatics Group