Computing desk
< June 4	<< May \| June \| Jul >>	June 6 >

Welcome to the Wikipedia Computing Reference Desk Archives
The page you are currently viewing is an archive page. While you can leave answers for any questions shown below, please ask new questions on one of the current reference desk pages.

June 5[edit]

Memtest but for graphics cards?[edit]

Memtest86+ applies a test to RAM, checking the result. I want to do the same thing with my graphics card. FurMark was the first hit on Google but if I'm not mistaken, it's approach seems to be to stress the card in an attempt to cause a BSOD. Surely it will miss many miscalculations made by the card. Is there no more rigorous method to test graphics cards are performing properly? 78.148.110.113 (talk) 18:13, 5 June 2014 (UTC)[reply]

I would recommend using free benchmark tools and comparing the results to the results obtained by others. Guru3d.com has some links to a few.8.17.117.40 (talk) 19:41, 5 June 2014 (UTC)[reply]

I don't know of tools for verifying proper rendering, but I'm pretty sure there are LINPACK benchmarks for video cards. This will have them do a complicated linear algebra computation and verify the result. It's at least somewhere to start. A well-written one should make use of most of the memory and a good portion of the card's computing resources. Katie R (talk) 13:47, 6 June 2014 (UTC)[reply]

I'm somewhat confused what you're looking for. The various reasons of Memtest only test RAM. They're good at it (although from my experience overclocking the memory subsystem with with modern APU, do have their limitations). But they don't really test any result, at least not in a simple way (they primarily write stuff to the RAM with various patterns and read it back and make sure it's what is expected, you could call this 'the result' but it confusing and makes me understand if you really understand what memtest is doing and what it's good for).

Something like Prime95, or OCCT or Linpack or whatever does test the results of calculations made by the CPU. They are essential for any stress test of a computer, unless for some reason you have some reason to think only the RAM is problematic. (And note as I mentioned, from my experience memtest may fail to pick up problems which are likely with the memory subsystem which the CPU stress tests will pick up.)

OCCT has a GPU stress test with error checking function [1]. It's also supposed to have a memtest for GPU which doesn't seem to show up in my version. But definitely the normal GPU test and probably even the memtest is much more like the CPU OCCT and Linpack tests and other such atress tests rather than memtest86/+. The EVGA OC scanner also has an error check function I believe, but it only support eVGA cards [2]. IIRC there are also a bunch of other tools with GPU error checking functions, which I can't recall the name of.

CUDA GPU memtest [3] which despite the name also supports OpenCL but is apparently Linux only and memtestcl/memtestG80 [4] seem to be trying to replicate memtest86 for the GPU. These will probably be useful if you can get them to work but bear in mind what I said earlier, the are likely only part of stress testing the GPU. That said, I'm not totally sure whether they only test the memory either or do some general purpose stress testing. (Even if they do, I would still recommend another general purpose GPU stress testing tool like OCCT unless you are totally sure you only want to test the GPU memory).

Nil Einne (talk) 15:51, 6 June 2014 (UTC)[reply]