PDA

View Full Version : Need suggestions on Nvidia GPU


DigitalDeviant
11th February 2009, 18:45
Can anyone recommend a cheap Nvidia GPU that can use CUDA, since it seems to be taking off, and do as much DXVA acceleration on VC-1 as possible? HDMI output would be a plus but it's not a deal breaker if it's not there.

neuron2
11th February 2009, 20:33
Define "cheap".

DigitalDeviant
11th February 2009, 20:58
As cheap as possible, preferable under $80USD. It's just for video playback not gaming so I'm hoping not to have to get a higher end card.

RunningSkittle
11th February 2009, 21:57
The 8600x series is pretty cheap these days, so is the 9600.

JackNF
11th February 2009, 22:17
Just look around on NewEgg or some other online store and find the cheapest 8 or 9 series Nvidia-based card you can find, as AFAIK even the lowliest of low end 8 series cards can handle HD playback via DXVA just fine.

For CUDA, here's what Nvidia has to say on which cards would work:
http://www.nvidia.com/object/cuda_learn_products.html

DigitalDeviant
12th February 2009, 16:27
Thanks for the info guys. One question, will going with DDR3/1GB give me and real performance/quality increase over DDR2/512MB?

Adub
13th February 2009, 04:22
You are comparing two things.

a) more memory vs less memory
b) two different types of memory, one aimed at higher clock frequencies and slower timings, the other aimed at lower frequencies but faster timings.

However, with respect to DXVA acceleration it shouldn't really matter. If you plan on staying "future safe" go with the bigger and badder 1gb, as it can help support bigger frame sizes. But still, any amount of memory that you look at now will be just fine.

saint-francis
13th February 2009, 06:21
So issues people speak of with needing memory on the graphics card for using CUDA are moot when we're in the 512 mb range? For hd (1080p) will 1 GB make a difference? What exactly uses the memory when using DGNV tools and how does it use it? Also how will it work in the foreseeable future (please chime in Neuron2)?

_DW_
24th February 2009, 18:54
The 8600x series is pretty cheap these days, so is the 9600.

I'd be careful when buy the 8x00 series because of heat. I bought a 8500 to replace the 6600 in my htpc a few months ago. The heat on the GPU run 73C when idle. Scratching around the net I found that to be the normal temp for this gpu.

Performance wise the card is perfect and I only paid 50 bucks for it from the egg. Besides watching video the card performs well for some light gaming. GTA:SA runs beautifully on it.

I also paid 50 bucks for a 9500 from newegg a few weeks ago. It's gpu is running right at about 45c.

swaaye
1st March 2009, 05:03
Don't buy a 8600 anything (fyi 9500GT is the exact same thing). They are about 25% as fast as an 8800GTX or 9800GTX. I'd go after a 9600GT as the cheapest decent NV option. It's in a whole different class than 8600/9500.

CUDA's future is not all that clear though. DirectX 11 will bring compute shaders to all DX11-capable cards (surely coming soon-ish) and there's also OpenCL. CUDA is NV-proprietary and crap like that never lasts.

squid_80
1st March 2009, 09:31
CUDA is NV-proprietary and crap like that never lasts.
I keep seeing misinformed statements like this posted on this forum. I find it hard to believe that any developer could look at both the CUDA and OpenCL APIs and not see the resemblance.

onesloth
1st March 2009, 15:01
Check the Nvidia GPU table here (http://en.wikipedia.org/wiki/PureVideo_HD#PureVideo_HD) to be sure you at least get a VP2 capable card.

laserfan
1st March 2009, 16:06
I can vouch for MSI's N9600GT OC 512. Got it on sale (after rebate) for less than your target (newegg iirc). Silent too.

Cyber-Mav
1st March 2009, 22:58
see if you can get a 9400gs or something that has VP3.

burfadel
1st March 2009, 23:09
Remember that many of the GPU's in the Geforce 9 series (and Geforce 8 mobile chips) had a manufacturing fault in the GPU itself, where failure can be expected after time. Certainly wouldn't want to buy old stock or a second hand version of one of these!

Its also worthwhile, as someone else has mentioned, working out whether the benefits you gain from using CUDA now will be worth the expense considering Directx 11 is coming out shortly, along with OpenCL, whereby the uptake of GPU processing under these technologies will be much higher.

Cyber-Mav
2nd March 2009, 01:36
yes iv heard about that failure, its to do with the die packaging of the gpu which nackers out after a lot of hot and cold cycles. but it mainly effected laptops since the gpu in those gets very hot.

just use aftermarket cooling to keep the desktop cards running cool. but from what iv experienced with ati, its better to get known faulty cards from nvidia than it is to get ati cards.

swaaye
20th March 2009, 19:19
I keep seeing misinformed statements like this posted on this forum. I find it hard to believe that any developer could look at both the CUDA and OpenCL APIs and not see the resemblance.

I've heard that a few times now. I hope OpenCL dies then too. ATI has said that for CUDA to be usable for them they'd basically have to redesign their GPU architecture into something similar to G80 (all of NV's modern GPUs are similar to G80). So lets hope that DX11 compute shaders take over GPGPU then. While that's great for MS, it's still better than some architecture-specific NVIDIA-controlled API.

swaaye
20th March 2009, 19:21
yes iv heard about that failure, its to do with the die packaging of the gpu which nackers out after a lot of hot and cold cycles. but it mainly effected laptops since the gpu in those gets very hot.

Yeah it has to do with ROHS-compliant solder apparently. The solder balls that are in between the GPU and the PCB can crack due the solder being susceptible to breakage after repeated thermal stress caused by powering a system on and off (or even just from GPU usage vs. idle).

Cyber-Mav
23rd March 2009, 01:28
I've heard that a few times now. I hope OpenCL dies then too. ATI has said that for CUDA to be usable for them they'd basically have to redesign their GPU architecture into something similar to G80 (all of NV's modern GPUs are similar to G80). So lets hope that DX11 compute shaders take over GPGPU then. While that's great for MS, it's still better than some architecture-specific NVIDIA-controlled API.

if thats true then how did NGOHQ.com manage to hack the ati drivers to support cuda and even got physx working?

its ati that doesnt want to support cuda. and cuda licencing is free to everyone.

littleD
23rd March 2009, 14:21
Well, dont confuse cuda with opencl. The latter is a part of opengl architecture and will have probably the same function as shader 5 for directx 11. The point is if you want use shader 5 you have to buy dx11 compliant card, while opencl should be working on older hardware.
Dont know about cuda but nvidia psychics works (in 3dmark) on ati card with hacked drivers. So its not about incompatibility, but competition between them two.
Anyway, we have cuda working already, and dont know when and how well ati will implement opencl into their drivers.