Best Nvidia CUDA Card for HQPlayer

I'm about to buy a graphic card, so please make your suggestions about an affordable Nvidia graphic card with CUDA

Thanks

I am wondering too

Thanks.

I have one new card that I think is sort of "affordable", but haven't got time yet to install it yet.

I'll let you know how it works once I have it in system and running. Hopefully over the weekend.

I have one new card that I think is sort of "affordable", but haven't got time yet to install it yet.

 

I'll let you know how it works once I have it in system and running. Hopefully over the weekend.

Thanks Miska. The computer I'm using for HQPlayer is also one I use for editing very large photo files because it also required a fast processor, lots of RAM and a high speed graphics card with lots of its own RAM. Unfortunately, I didn't go with NVIDIA when I originally bought it. Do you have any view as to whether buying an NVIDIA Cuda card with high graphics capabilities (which make it more expensive) would have any negative effects on HQPlayer output if the card is only being used for audio or for sound, but not both at the same time?

I made a small write-up here about the GPU installation and initial testing results:

Upgrade GPU for more CUDA processing power - Blogs - Computer Audiophile

Any intuition on how an Asus 750 would perform?

Any intuition on how an Asus 750 would perform?

I would recommend going at minimum 900-series. 700-series is old already and doesn't necessarily help much.

Nvidia just launched the new 1000-series:

It is supposed to have about 16x faster double-precision floating-point performance compared to previous Maxwell-generation (900-series).

Looking at upgrading to a GTX 980.Right now have a Quadro K2000.Has very little cud a cores.When cuda offload is checked can't play any poly sinc filters.When box is checked only work with closed form.Upsampling to dsd256.Any ideas ?Dont want to spend money on a new card if it won't work.

Anyone ?

Looking at upgrading to a GTX 980.Right now have a Quadro K2000.Has very little cud a cores.When cuda offload is checked can't play any poly sinc filters.When box is checked only work with closed form.Upsampling to dsd256.Any ideas ?Dont want to spend money on a new card if it won't work.

What kind of machine the card is on? Since only part of the work can be offloaded, it could as well be limitation of the host computer, not necessarily the GPU.

At this point I would recommend waiting for the GTX 1080, unless you can get GTX 980 for really cheap. The performance difference for HQPlayer type of loads is supposed to be pretty huge. So something like GTX 1060 is supposed to be equal or faster than GTX 980 at significantly lower price.

Lenovo D30 think station.32 gab ram,windows 10,E5-2620 v2 6 core 12 thread processor.Video card right now is a Quadro K2000 which only has 364 cuda cores.Shows only 3 on compatibility chart

In the US, Micro Center has the 970 and 980 on sale at over 30% off list. I think I saw the 980 was $US500.

I suspect they are making ready for the 1000 series.

In the US, Micro Center has the 970 and 980 on sale at over 30% off list. I think I saw the 980 was $US500.

 

I suspect they are making ready for the 1000 series.

I'm sure that's the story. Ditto for some of the gaming laptops with the 960, 970 and 980 GPUs. Also on sale on the web.

As for the 1080, that looks like quite a powerful GPU - even before overclocking !

Nvidia GeForce GTX 1080 review: The most badass graphics card ever created | PCWorld

As for the 1080, that looks like quite a powerful GPU - even before overclocking !

At least here (I'd expect worldwide) sales start May 27th. List price for the Founder's Edition is 799€. The normal edition coming later is supposed to priced around 599€.

Quite a lot of processing power per EUR.

Should be more than enough processing power for upsampling to DSD 256 Multichannel and DSD 512 Stereo...

Hi Miska

I saw your blog about your new multi-channel PC monster with GTX 1080. What kind of CUDA-Z score does it give you? I read here that double precision FP performance is something NVIDIA have neglected in some of their GPU's e.g. the GTX Titan X is only about 40% of the older GTX Titan. So possibly there some bargains to be had using older GPU's for HQPlayer?

Happy New Year!

I saw your blog about your new multi-channel PC monster with GTX 1080. What kind of CUDA-Z score does it give you? I read here that double precision FP performance is something NVIDIA have neglected in some of their GPU's e.g. the GTX Titan X is only about 40% of the older GTX Titan. So possibly there some bargains to be had using older GPU's for HQPlayer?

Here are the scores for my system:

It is almost double the performance of GTX980 here:

Upgrade GPU for more CUDA processing power - Blogs - Computer Audiophile

So it is not bad in that sense.

However, Pascal is supposed to be almost 16x faster on DP FP than Maxwell... I don't know why the figures are not better.

If you look at the figures here:

https://en.wikipedia.org/wiki/List_of_Nvidia_graphics_processing_units

Titan X, Quadro P6000 and Tesla P40 are about the same ~350 GFLOPS. All three use GP102 so it is to be expected too... But the difference between single and double precision FP is huge as is for all the other models. But then if you look at the Tesla P100 the ratio between single and double is just 2x like typically on normal CPUs because it is mostly limited by RAM bandwidth. And actually SP FP is a bit slower than P40, but now the figures are what should be expected, 4000+ GFLOPS DP FP! Also from RAM bandwidth point of view, P100 has less than double the bandwidth of Titan X.

So based on 2x ratio Titan X, Quadro P6000 and Tesla P40 should give closer to 5000 GFLOPS DP FP. And GTX 1080 closer to 4000 GFLOPS!

So a bit of mystery to me - did they cut out some crucial piece of double precision hardware from P100 going to all smaller models?

As a side note, would be interesting to try HQPlayer on one of these...

This is my result of GTX Titan Black, the same GPU die is used for Titan supercomputer. Double precision performance is 1700 Gflops. I think this is still the best card for home GPU computing

This is my result of GTX 1080. Double precision performance is about 300 Gflops.

On Nvidia control panel, Double precision function of the Titan GPU should be enabled (Default value is disabled. I don't know why). It seems there is no double precision menu appeared on GTX 1080.

Thanks for the CUDA-Z scores. Anyone else care to contribute?

@yamamoto2002 could I borrow that for a wee while

@Miska In the Anandtech review link above there is some background re FP64 design/marketing decisions made by Nvidia.

Looking at the broader picture, I’m left to wonder if this is the start of a permanent divorce between graphics/FP32 compute and FP64 compute in the NVIDIA ecosystem.

I was playing with a spreadsheet when Yamamoto posted, and the theoretical performance quoted on Wikipedia seems to be a fair guide but not 100% accurate predictor of Cuda-Z or Compute: Folding@Home: Explicit double precision FP Score. So it does seem you can use Wikipedia to find a high performance GPU for HQPlayer, but there are not so many older generation cards for sale at a nice price!

Dual-GPU cards such as GTX Titan Z, GTX 690, GTX 590 have two separated global memory for each GPU and it is a bit more difficult to program than conventional single GPU cards. It is recommended to check if the GPGPU application you runs on it supports such a exotic cards before you purchase

I was playing with a spreadsheet when Yamamoto posted, and the theoretical performance quoted on Wikipedia seems to be a fair guide but not 100% accurate predictor of Cuda-Z or Compute: Folding@Home: Explicit double precision FP Score. So it does seem you can use Wikipedia to find a high performance GPU for HQPlayer, but there are not so many older generation cards for sale at a nice price!

Can you explain what this all means. Which are the bargain cards below the Titan and 10x0 series?

Can you explain what this all means. Which are the bargain cards below the Titan and 10x0 series?

Sorry if it wasn't clear. The Cuda-z score is a way checking if a graphics card has enough oomph for HQPlayer. I was looking for a way around buying a completely new i7-6700K computer so I can upsample PCM to DSD512, or at least DSD256. From the charts you can see the older Titan GPU's are very powerful, but I've since discovered, more expensive than a new i7-6700K system, and not easily found 2nd hand.

Also it looks like we can rule out Dual-GPU cards such as GTX Titan Z, GTX 690, GTX 590 (thanks for the info yamamoto2002), and older GPU's not meeting HQPlayer's Cuda Compute level 3. So this leaves cards like GeForce GTX 780 Ti/GeForce GTX 980 Ti and new cards like GeForce GTX 1060/1070/1080 which are all quite expensive in the UK.So unless something nice shows up in the 2nd hand market it looks like an i7-6700K is the best option for me.

Sorry if it wasn't clear. The Cuda-z score is a way checking if a graphics card has enough oomph for HQPlayer. I was looking for a way around buying a completely new i7-6700K computer so I can upsample PCM to DSD512, or at least DSD256. From the charts you can see the older Titan GPU's are very powerful, but I've since discovered, more expensive than a new i7-6700K system, and not easily found 2nd hand.

Also it looks like we can rule out Dual-GPU cards such as GTX Titan Z, GTX 690, GTX 590 (thanks for the info yamamoto2002), and older GPU's not meeting HQPlayer's Cuda Compute level 3. So this leaves cards like GeForce GTX 780 Ti/GeForce GTX 980 Ti and new cards like GeForce GTX 1060/1070/1080 which are all quite expensive in the UK.So unless something nice shows up in the 2nd hand market it looks like an i7-6700K is the best option for me.

I see.

Does this mean that the i7-6700k machine would not have a beefy GPU? I am using an i7-4770k and wondering if I need a nice GPU to compliment it.

Best way to find out is try the free trial if you haven't done so. From what I've read here i7-4770k may not get you up to DSD512, so maybe a GPU would be a good idea. However DSD256 is rated very highly, so you maybe satisfied with that.

Maybe someone else has experience of i7-4770k or similar?