Best Nvidia CUDA Card for HQPlayer

[Snoozer]

Thank you Miska !

 

And what about an Nvidia RTX2070, will it be enough or should I go with an RTX2080 ? I in the market to buy the GPU and the 2070 is certainly cheaper option.

 

Duly noted, I will try to get an Ubuntu distro for cuda. No prob there.

 

cheers

[Miska]

19 hours ago, Kal Rubinson said:

The OKTO DAC8 will do DSD128.

 

With DoP max DSD64 and with native/raw DSD128. But likely no DSD256 (out of UAC spec).

 

[Snoozer]

No comments on Nvidia RTX2070 vs. RTX2080 ? .... I am holding the decision on which one to buy because I value a lot your input

 

thanx

[Miska]

15 hours ago, Snoozer said:

And what about an Nvidia RTX2070, will it be enough or should I go with an RTX2080 ? I in the market to buy the GPU and the 2070 is certainly cheaper option.

 

It is really hard to estimate, given that I have only 2080 and 2080Ti. And it also depends a lot on the convolution filter design (number of actually used taps).

 

[Snoozer]

thank you Miska !!

 

cheers

[Snoozer]

Just out of curiosity, would you please head me to an internet page (or even a book) that allow me to learn how to apply filtering directly in DSD ?. If there is no conversion in native DSD to apply the crossover filters, what are the mathematics behind DSD filtering ?.

 

Again, I have a lot of time to study and I really find joy in doing it. I apologise for digressing the theme of this thread to the readers.

 

thanx again Miska

[Kal Rubinson]

4 hours ago, Miska said:

With DoP max DSD64 and with native/raw DSD128. But likely no DSD256 (out of UAC spec).

Sure.  DSD256 not claimed.

[Bayan]

Does HQplayer support cuda with the NVIDIA TESLA V100 GPU accelerator?

[Miska]

6 hours ago, Bayan said:

Does HQplayer support cuda with the NVIDIA TESLA V100 GPU accelerator?

 

I don't have one to test with, but I don't see a reason why it wouldn't work. It should work fine.

 

[caesee]

On 6/16/2020 at 10:08 PM, Miska said:

 

I don't have one to test with, but I don't see a reason why it wouldn't work. It should work fine.

 

tesla GPU accelerator card may have different memory type. 8g model is cheap and 24 g is much more expensive. Is there any difference in hqplayer？

[AudioDoctor]

On 6/12/2020 at 5:07 AM, Snoozer said:

No comments on Nvidia RTX2070 vs. RTX2080 ? .... I am holding the decision on which one to buy because I value a lot your input

 

thanx

 

I have never used the 2070, but I have used the 2080 and if I were making the decision today I would go with the 2080.

[Miska]

5 hours ago, caesee said:

tesla GPU accelerator card may have different memory type. 8g model is cheap and 24 g is much more expensive. Is there any difference in hqplayer？

 

Tesla has HBM2 memory on 4096-bit wide bus, so it is different in that sense. Depending on what you want to do, 8 GB may tight.

 

I know the modern Teslas cost about $20k+ which is too much for me.

 

[caesee]

On 12/28/2021 at 3:32 AM, Miska said:

 

Tesla has HBM2 memory on 4096-bit wide bus, so it is different in that sense. Depending on what you want to do, 8 GB may tight.

 

I know the modern Teslas cost about $20k+ which is too much for me.

 

TESLA P100 or V100 is much more cheap than newer v100，nd same HBM2 memory...

[Miska]

1 hour ago, caesee said:

TESLA P100 or V100 is much more cheap than newer v100，nd same HBM2 memory...

 

Pascal is pretty old and doesn't quite have the feature set of the newer generations.

 

[oneguy]

Are any of the 95W TDP or less GPUs worth it to aid in helping offload CPU tasks for upsample?
 

Half-height preferred but full height acceptable. 

 

I’m trying to explore options to get to DSD512 with an xtr non-2s filter and 7ECV2 modulator while keeping CPU temps below 90C. That other row of heatsinks in the H5 case really need more of a purpose. 

[Miska]

2 hours ago, oneguy said:

Are any of the 95W TDP or less GPUs worth it to aid in helping offload CPU tasks for upsample?

 

RTX3060 draws 170W, but Nvidia doesn't specify TDP on their website. Mobile versions are probably lighter on TDP, but I have not seen anybody offering those on PCIe cards. I have a 14" laptop with AMD Ryzen 9 5900HS CPU and RTX3060 Mobile GPU. Although it's fans get loud on full load, it is pretty small package.

 

[oneguy]

1 hour ago, Miska said:

 

RTX3060 draws 170W, but Nvidia doesn't specify TDP on their website. Mobile versions are probably lighter on TDP, but I have not seen anybody offering those on PCIe cards. I have a 14" laptop with AMD Ryzen 9 5900HS CPU and RTX3060 Mobile GPU. Although it's fans get loud on full load, it is pretty small package.

 

Being able to plug the mobile version into PCIe slot would be cool if that were possible. 
 

So far the best I have found is the Nvidia GTX 1650 GDDR6

[Miska]

Just now, oneguy said:

Being able to plug the mobile version into PCIe slot would be cool if that were possible. 
 

So far the best I have found is the Nvidia GTX 1650 GDDR6

 

Possibly not very useful though.

[oneguy]

I’m thinking an Nvidia GeForce RTX 3060 12gb with a Raijintek Morpheus cooler sticking out the top of the case may be doable if I can identify a version of the cooler to mate with the 3060. 
https://www.raijintek.com/en/products_bycategory.php?CategoryID=1&SubCategoryID=22

 

It would require some custom casework to go down that avenue. 
 

Better shot at pulling a sufficient amount of processing load from the CPU to lower temps on DSD512?

[AudioDoctor]

3 hours ago, oneguy said:

I’m thinking an Nvidia GeForce RTX 3060 12gb with a Raijintek Morpheus cooler sticking out the top of the case may be doable if I can identify a version of the cooler to mate with the 3060. 
https://www.raijintek.com/en/products_bycategory.php?CategoryID=1&SubCategoryID=22

 

It would require some custom casework to go down that avenue. 
 

Better shot at pulling a sufficient amount of processing load from the CPU to lower temps on DSD512?

 

Have you seen the MonsterLabo The Beast passive cooling case?  I have one on order and it should arrive next month.

[oneguy]

4 hours ago, AudioDoctor said:

 

Have you seen the MonsterLabo The Beast passive cooling case?  I have one on order and it should arrive next month.

I have. I was thinking of ordering one but I would have blown a decent part of my budget on a case. My target for the the initial build was 3-4k and I came in at $3,650. In hind sight though I should have just passed on the HDPLEX 300w LPS and saved that $685. 
 

The idea on the The Beast may be worth revisiting though in the future. The 250w GPU cooler would allow a 3070 to be run.

[Rune]

14 hours ago, oneguy said:

I’m thinking an Nvidia GeForce RTX 3060 12gb with a Raijintek Morpheus cooler sticking out the top of the case may be doable if I can identify a version of the cooler to mate with the 3060. 
https://www.raijintek.com/en/products_bycategory.php?CategoryID=1&SubCategoryID=22

 

It would require some custom casework to go down that avenue. 
 

Better shot at pulling a sufficient amount of processing load from the CPU to lower temps on DSD512?

You could probably undervolt or choose a lower clock on the GPU to make it run cooler.

In my system the CPU runs a lot hotter than the GPU running stock settings. It is also possible that CUDA work loads is lighter on the GPU than Games so TDP rating is meaningless because you will not max it.

[oneguy]

2 hours ago, Rune said:

You could probably undervolt or choose a lower clock on the GPU to make it run cooler.

In my system the CPU runs a lot hotter than the GPU running stock settings. It is also possible that CUDA work loads is lighter on the GPU than Games so TDP rating is meaningless because you will not max it.

That’ll help. The Taijintek cooler doesn’t explicitly say it will mate to a 3060 which is another issue. It did see a guy on Reddit mate one with a 3080 so there is hope. But from what I understand there is so much variation in how companies implement the 30 series board that it may be a crapshoot when picking what to order. 

[Miska]

49 minutes ago, oneguy said:

That’ll help. The Taijintek cooler doesn’t explicitly say it will mate to a 3060 which is another issue. It did see a guy on Reddit mate one with a 3080 so there is hope. But from what I understand there is so much variation in how companies implement the 30 series board that it may be a crapshoot when picking what to order. 

 

Many use Nvidia reference design as basis and have mostly changes on cooler and such. But the board layout and some of the cooler mounting are often from the reference design.

 

Short card versions usually differ from the reference designs. So one can try to compare pictures of the reference board.

 

[Andrey Boldakov]

On 7/26/2017 at 11:40 PM, Miska said:

At least not explicitly. Depending on how Nvidia implemented some of the CUDA functionality, multiple GPUs could possibly work when you have also convolution enabled. But since I don't have any multi-GPU machines I have not checked whether this happens or not.

Miska, could you kindly answer if multi-GPU (dual GPU) works for offloading HQPlayer tasks?