Page 1 of 1

Simultaneous processing

Posted: Mon Oct 26, 2015 8:25 pm
by Ch3vr0n
According to my GPU specs it has 4 GPC's (Graphic Processing Clusters) and 8 Streaming Multiprocessors (SMX), and 4 memory controllers. Yet CXV only picks 2 by default. Why does it do that and can i increase it to 4 or 8 to maximize performance ?

Re: Simultaneous processing

Posted: Tue Oct 27, 2015 9:15 am
by cedric
Hello,

The number of 2 hardware encoding instances is only a default value. (some cheaper cards can only convert one 1080p conversion at once)

You can try to increase this setting as long as the encoding start...

Regards,

Re: Simultaneous processing

Posted: Tue Oct 27, 2015 2:29 pm
by Ch3vr0n
yeah but to which one. 4 or 8 (by card specs)

Re: Simultaneous processing

Posted: Wed Nov 04, 2015 10:59 am
by cedric
You can't evaluate the number of maximum simultaneous instances only by checking your graphic card specs.
This max number is dependent of output resolution (720p, 1080p...), graphic card memory USABLE when starting the conversion, and a lot of another things...

You must increase the number of instances until the conversions doesn't start.

Regards,

Re: Simultaneous processing

Posted: Sat Jan 23, 2016 3:34 am
by Brainiarc7
Ch3vr0n wrote:According to my GPU specs it has 4 GPC's (Graphic Processing Clusters) and 8 Streaming Multiprocessors (SMX), and 4 memory controllers. Yet CXV only picks 2 by default. Why does it do that and can i increase it to 4 or 8 to maximize performance ?
Hello there,

Allow me to correct the answer provided by Cedric, because it's wrong.

With hardware accelerated encoding on your Nvidia Kepler GPU, you're not using CUDA, but NVENC.
NVENC is a dedicated Silicon Intellectual property Core by Nvidia that offers both fixed and programmable function video encode, without using your GPU's cores.
It's an equivalent to Intel's QuickSync on Sandybridge and beyond and AMD's Video Coding Engine (VCE) on GCN-series GPUs.

For Nvidia, NVENC sessions are limited to 2 (simultaneous encodes) in consumer GPUs.
In Quadro GPUs, you can have as many simultaneous encodes as you wish as long as VRAM permits.

Similar limitations exist on equivalent SIP blocks, each with it's limiter.
For example:

1. With Intel, the maximum you can have is ~4 , and this number corresponds to the maximum hardware decode sesssions supported by the IGP via DXVA and Intel's MSDK MFT filters.
2. With AMD, this number is ~ 2 , and again, only in hybrid encode path (where both GPU shaders + VCE IL work together, ~4), otherwise, in Desktop Encode Mode (default), this is again limited to ~2.

Hope this clarifies the situation for you.