Changing GPU - But which will be better? |
![]() |
| log in |
Message boards : Number crunching : Changing GPU - But which will be better?
| Author | Message |
|---|---|
|
Hello, | |
| ID: 1324732 · | |
|
Not saying I am an expert, but it looks like the GPU list is outdated since it doesn't have any of the new Kepler GPU's. Generally, the more CUDA cores a GPU has, the more work it can do. So, the GTX580 will perform marginally better than the GTX 570. Also, memory bandwidth can make some small but measurable difference so a card with a 384 or 256 bit memory bus will outperform a GPU with only a 192 or 128 bit bus. Clocking a card higher in graphics or memory always means it will do more work in a given period. So choose the highest clocked card you can get and still live with the noise, heat and instability. The newer card families all use less power than previous so less heat and noise. Also, the newer cards have optimized CUDA drivers and applications so they process more work than previous cards using lesser drivers and applications. Use of an optimized client will often net a better performance enhancement using an older card technology than getting a newer GPU using the stock applications. For Nvidia cards, the newest Kepler family of cards is performing better than the GTX 570 or 580 when using the optimized clients. They also use less energy. The Kepler family of GPU's do have a big performance hit with other projects like MilkyWay@Home which needs double precision math as Kepler only does 1/24 FPU work versus the older Fermi family which does 1/8 FPU work. They do make up some performance lost in the FPU work in their sheer much higher clocking. So, pick a price point you can live with and get a new card for Seti@Home or look for an older GTX 570 or GTX580 if you need real good double precision math performance for other projects. | |
| ID: 1324738 · | |
|
There are a unwrited roules for GPU´s: | |
| ID: 1324743 · | |
|
rule 1 how much including pc build cost does it cost per wu. | |
| ID: 1324782 · | |
|
Thanks guys for your help! | |
| ID: 1324993 · | |
Hello, That GPU list is a raw speed comparison based on the lowest average time per WU. For recent GPUs it's better to run multiple tasks simultaneously, although that slows each task more tasks get done per unit time. The list in http://www.efmer.eu/forum_tt/index.php?topic=981.0 has better statistics for anyone planning to use an NVIDIA GPU efficiently. Joe | |
| ID: 1324994 · | |
|
I like to make some comments. | |
| ID: 1326387 · | |
I like to make some comments. Hasn't that something todo with the Reg_Size, being twice as big in the 600 series (KEPLER), as the FERMIs with 32K regsize.
Setiathome_CUDA: Found 1 CUDA device(s):
Device 1: GeForce GTX 480, 1535 MiB, regsPerBlock 32768
computeCap 2.0, multiProcs 15
clockRate = 1401000
AstroPulse Two GPU results, 1 rev.1363 (LUNATICs Optimized), 1 stock app. and CPU. I'm mistaken, wanted to compaire 400 (FERMI) and 600 (KEPLER) and CPU, but it's a GTX 550 Ti. And a result overflow :-/. ____________ Knight Who Says Ni N!, OUT numbered................. | |
| ID: 1326407 · | |
I like to make some comments. The cuda cores from the 6xx cards are a lot lot less powerfull? is it because of the "power efficient"?! Especially I'm looking for a GTX 650 Ti - but i couldn't find it at the list right here: http://www.efmer.eu/forum_tt/index.php?topic=981.0 Between which GPUs would it be placed at the list!? Thanks! ____________ | |
| ID: 1327324 · | |
|
I do have a 650 Ti installed in my machine alongside a 670. | |
| ID: 1327367 · | |
|
Has anyone any experience with an EVGA GTX560 Superclocked of one of two types, the 02G-P3-1469-KR or the 02G-P3-2069-KB? | |
| ID: 1327645 · | |
Message boards : Number crunching : Changing GPU - But which will be better?
| Copyright © 2013 University of California |