Choose which Nvidia GPU to use if multiple GPUs are present. #feature


Eric Inloes
 

So, It would be nice to choose which Nvidia GPU to use if multiple GPUs are present. While this is niche setup it would be a nice feature to have. 

I figure this wouldn't be very difficult to do hopefully. 


Simon Brown
 

Hi,

 

It wouldn’t – at the moment I select the best card, in fact I let the NVIDIA API do this for me.

 

Simon Brown, G4ELI

https://www.sdr-radio.com

 

From: main@SDR-Radio.groups.io <main@SDR-Radio.groups.io> On Behalf Of Eric Inloes
Sent: 09 July 2020 00:29
To: main@SDR-Radio.groups.io
Subject: [SDR-Radio] Choose which Nvidia GPU to use if multiple GPUs are present. #feature

 

So, It would be nice to choose which Nvidia GPU to use if multiple GPUs are present. While this is niche setup it would be a nice feature to have. 

I figure this wouldn't be very difficult to do hopefully. 


jdow
 

Perhaps if second best is good enough it should be used leaving the other one for some other high powered calculations. It probably also has a place in regression testing to have a "weak" card you can select for testing.

{^_^}

On 20200708 21:57:10, Simon Brown wrote:
Hi,
It wouldn’t – at the moment I select the best card, in fact I let the NVIDIA API do this for me.
Simon Brown, G4ELI
https://www.sdr-radio.com
*From:*main@SDR-Radio.groups.io <main@SDR-Radio.groups.io> *On Behalf Of *Eric Inloes
*Sent:* 09 July 2020 00:29
*To:* main@SDR-Radio.groups.io
*Subject:* [SDR-Radio] Choose which Nvidia GPU to use if multiple GPUs are present. #feature
So, It would be nice to choose which Nvidia GPU to use if multiple GPUs are present. While this is niche setup it would be a nice feature to have.
I figure this wouldn't be very difficult to do hopefully.


Simon Brown
 

Hi,

The API options are to select a specific card or the 'best' card. The widest FFT I've encountered is 100MHz, with x8 resolution it's been close to 1 billion bins / second. The GTX 1070 worked but it was getting close.

At some stage in 2020 I expect to interface with a VHF / UHF SDR which delivers 500MHz of bandwidth. Scary stuff, I doubt that my motherboard will handle the data, let alone FFT.

Simon Brown, G4ELI
https://www.sdr-radio.com

-----Original Message-----
From: main@SDR-Radio.groups.io <main@SDR-Radio.groups.io> On Behalf Of jdow
Sent: 09 July 2020 06:54
To: main@SDR-Radio.groups.io
Subject: Re: [SDR-Radio] Choose which Nvidia GPU to use if multiple GPUs are present. #feature

Perhaps if second best is good enough it should be used leaving the other one for some other high powered calculations. It probably also has a place in regression testing to have a "weak" card you can select for testing.

{^_^}


jdow
 

scary indeed. I'd be scared trying to think if a "better way" to visualize the data. 500 MHz on a UHD screen with ten full width spectrum spans is still more than 12.5 kHz per pixel.

I think I'd try for 8 screen widths and discard some cruft off the edges to use a 32768 FFT for full width visualization and then mixers and decimators to get regions that need higher resolution. Even with "many" such spectrum chunks it might use less processor and still give full information and flexibility.

That said, a display system that could display say a single pixel for each bin of a 536,870,912 bin FFT would be......... impressive. By then the resolution of the human eye limits the ability to grasp information.

{o.o}

On 20200709 00:29:56, Simon Brown wrote:
Hi,
The API options are to select a specific card or the 'best' card. The widest FFT I've encountered is 100MHz, with x8 resolution it's been close to 1 billion bins / second. The GTX 1070 worked but it was getting close.
At some stage in 2020 I expect to interface with a VHF / UHF SDR which delivers 500MHz of bandwidth. Scary stuff, I doubt that my motherboard will handle the data, let alone FFT.
Simon Brown, G4ELI
https://www.sdr-radio.com
-----Original Message-----
From: main@SDR-Radio.groups.io <main@SDR-Radio.groups.io> On Behalf Of jdow
Sent: 09 July 2020 06:54
To: main@SDR-Radio.groups.io
Subject: Re: [SDR-Radio] Choose which Nvidia GPU to use if multiple GPUs are present. #feature
Perhaps if second best is good enough it should be used leaving the other one for some other high powered calculations. It probably also has a place in regression testing to have a "weak" card you can select for testing.
{^_^}


Siegfried Jackstien
 

how many 4k monitors do you have to put in a row to have such high resolution??

" hey grab a coffee ... i take a "walk" from vhf to uhf ... and it will take some time ... should i take the hairy hound with me??" ...

500megs of bw?!? WOW

i can only imagine that monitoring a sat downlink (with lots of tv channels) would take such bw (in fact 500megs is not enough for the tv sat band) ... but for vhf uhf?? what needs such bw??

greetz sigi dg9bfc


Am 09.07.2020 um 08:19 schrieb jdow:

scary indeed. I'd be scared trying to think if a "better way" to visualize the data. 500 MHz on a UHD screen with ten full width spectrum spans is still more than 12.5 kHz per pixel.

I think I'd try for 8 screen widths and discard some cruft off the edges to use a 32768 FFT for full width visualization and then mixers and decimators to get regions that need higher resolution. Even with "many" such spectrum chunks it might use less processor and still give full information and flexibility.

That said, a display system that could display say a single pixel for each bin of a 536,870,912 bin FFT would be......... impressive. By then the resolution of the human eye limits the ability to grasp information.

{o.o}


On 20200709 00:29:56, Simon Brown wrote:
Hi,

The API options are to select a specific card or the 'best' card. The widest FFT I've encountered is 100MHz, with x8 resolution it's been close to 1 billion bins / second. The GTX 1070 worked but it was getting close.

At some stage in 2020 I expect to interface with a VHF / UHF SDR which delivers 500MHz of bandwidth. Scary stuff, I doubt that my motherboard will handle the data, let alone FFT.

Simon Brown, G4ELI
https://www.sdr-radio.com

-----Original Message-----
From: main@SDR-Radio.groups.io <main@SDR-Radio.groups.io> On Behalf Of jdow
Sent: 09 July 2020 06:54
To: main@SDR-Radio.groups.io
Subject: Re: [SDR-Radio] Choose which Nvidia GPU to use if multiple GPUs are present. #feature

Perhaps if second best is good enough it should be used leaving the other one for some other high powered calculations. It probably also has a place in regression testing to have a "weak" card you can select for testing.

{^_^}





jdow
 

All I can say is "heh heh it's not big enough for the 'what' I committed." (Martini olive satcom? Well, not QUITE that small and VERY wide band.)

{^_-}

On 20200709 03:37:12, Siegfried Jackstien wrote:
how many 4k monitors do you have to put in a row to have such high resolution??
" hey grab a coffee ... i take a "walk" from vhf to uhf ... and it will take some time ... should i take the hairy hound with me??" ...
500megs of bw?!? WOW
i can only imagine that monitoring a sat downlink (with lots of tv channels) would take such bw (in fact 500megs is not enough for the tv sat band) ... but for vhf uhf?? what needs such bw??
greetz sigi dg9bfc
Am 09.07.2020 um 08:19 schrieb jdow:
scary indeed. I'd be scared trying to think if a "better way" to visualize the data. 500 MHz on a UHD screen with ten full width spectrum spans is still more than 12.5 kHz per pixel.

I think I'd try for 8 screen widths and discard some cruft off the edges to use a 32768 FFT for full width visualization and then mixers and decimators to get regions that need higher resolution. Even with "many" such spectrum chunks it might use less processor and still give full information and flexibility.

That said, a display system that could display say a single pixel for each bin of a 536,870,912 bin FFT would be......... impressive. By then the resolution of the human eye limits the ability to grasp information.

{o.o}


On 20200709 00:29:56, Simon Brown wrote:
Hi,

The API options are to select a specific card or the 'best' card. The widest FFT I've encountered is 100MHz, with x8 resolution it's been close to 1 billion bins / second. The GTX 1070 worked but it was getting close.

At some stage in 2020 I expect to interface with a VHF / UHF SDR which delivers 500MHz of bandwidth. Scary stuff, I doubt that my motherboard will handle the data, let alone FFT.

Simon Brown, G4ELI
https://www.sdr-radio.com

-----Original Message-----
From: main@SDR-Radio.groups.io <main@SDR-Radio.groups.io> On Behalf Of jdow
Sent: 09 July 2020 06:54
To: main@SDR-Radio.groups.io
Subject: Re: [SDR-Radio] Choose which Nvidia GPU to use if multiple GPUs are present. #feature

Perhaps if second best is good enough it should be used leaving the other one for some other high powered calculations. It probably also has a place in regression testing to have a "weak" card you can select for testing.

{^_^}






Simon Brown
 

Hi,

2 x 8k monitors,

Simon Brown, G4ELI
https://www.sdr-radio.com

-----Original Message-----
From: main@SDR-Radio.groups.io <main@SDR-Radio.groups.io> On Behalf Of jdow
Sent: 09 July 2020 09:20
To: main@SDR-Radio.groups.io
Subject: Re: [SDR-Radio] Choose which Nvidia GPU to use if multiple GPUs are present. #feature

scary indeed. I'd be scared trying to think if a "better way" to visualize the data. 500 MHz on a UHD screen with ten full width spectrum spans is still more than 12.5 kHz per pixel.

I think I'd try for 8 screen widths and discard some cruft off the edges to use a 32768 FFT for full width visualization and then mixers and decimators to get regions that need higher resolution. Even with "many" such spectrum chunks it might use less processor and still give full information and flexibility.

That said, a display system that could display say a single pixel for each bin of a 536,870,912 bin FFT would be......... impressive. By then the resolution of the human eye limits the ability to grasp information.

{o.o}


jdow
 

Too small for more than NFM. I like the idea of one big spectrum with sub-spectrums calculated on smaller chunks of data to cover areas of interest. A panoramic panel of 4 8k screens is sort of small for NFM monitoring - about 31-32 kHz per pixel.

Heh, four panoramic 8k screens above a fifth showing detail spectrum regions. You can use small FFTs for the panoramic view and use more detailed views on converted and filtered chunks of spectrum for detail views. I'd sort of pick on something between 8kHz and 16kHz per pixel for subsidiary windows unless you are looking for "scratches", very narrow signals. For that I'd go to a tertiary display with the display oriented to emphasize scratches in a 1 Hz/bin spectrum over less than 100 kHz to maybe 1 MHz max. Bin size should be adjustable to the bandwidth of the scratch you are looking for.

It depends on the target application. You want to present the user with the data he needs already emphasized rather than all the information that could be extracted from a random 500 MHz chunk of spectrum.

Another trick I'd use with channel sized bins is to shift the conversion frequency through four quarter channel size chunks and display power in each of the sub bins. That way you can get intentionally mis-tuned signals.

(Can you tell something about my imagination and sordid past with unnameable organizations? ECCM was more fun, really. But, ECM is a valuable tool. What's REALLY fun is figuring out how to make your signal look like HIS jammer.)

{o.o}

On 20200709 06:36:29, Simon Brown wrote:
Hi,
2 x 8k monitors,
Simon Brown, G4ELI
https://www.sdr-radio.com
-----Original Message-----
From: main@SDR-Radio.groups.io <main@SDR-Radio.groups.io> On Behalf Of jdow
Sent: 09 July 2020 09:20
To: main@SDR-Radio.groups.io
Subject: Re: [SDR-Radio] Choose which Nvidia GPU to use if multiple GPUs are present. #feature
scary indeed. I'd be scared trying to think if a "better way" to visualize the data. 500 MHz on a UHD screen with ten full width spectrum spans is still more than 12.5 kHz per pixel.
I think I'd try for 8 screen widths and discard some cruft off the edges to use a 32768 FFT for full width visualization and then mixers and decimators to get regions that need higher resolution. Even with "many" such spectrum chunks it might use less processor and still give full information and flexibility.
That said, a display system that could display say a single pixel for each bin of a 536,870,912 bin FFT would be......... impressive. By then the resolution of the human eye limits the ability to grasp information.
{o.o}


Simon Brown
 

Exactly what's planned. For example with the ELAD S3 and to lesser extent SDRplay you could display quite a few bands: ham, broadcast etc.

My gut feeling is that in a few years we’ll have sub-$299 16-bit SDRs which cover the whole of HF in one go with good dynamic range so this will be a reality.

As for 4 x 8k - would be a hell of PC but is very doable.

Simon Brown, G4ELI
https://www.sdr-radio.com

-----Original Message-----
From: main@SDR-Radio.groups.io <main@SDR-Radio.groups.io> On Behalf Of jdow
Sent: 10 July 2020 03:09
To: main@SDR-Radio.groups.io
Subject: Re: [SDR-Radio] Choose which Nvidia GPU to use if multiple GPUs are present. #feature

Too small for more than NFM. I like the idea of one big spectrum with sub-spectrums calculated on smaller chunks of data to cover areas of interest. A panoramic panel of 4 8k screens is sort of small for NFM monitoring - about
31-32 kHz per pixel.

Heh, four panoramic 8k screens above a fifth showing detail spectrum regions.
...


jdow
 

I believe my video card could do that. And if I recall specs the video card in the video machine can do 6x8k. (AMD) With the ability to specifically pick video cards it could actually do 7x8k with the Intel video turned on. The Direct Sound applications kept getting GPUs tangled up when I turned on the Intel card.

{o.o}

On 20200709 22:46:26, Simon Brown wrote:
Exactly what's planned. For example with the ELAD S3 and to lesser extent SDRplay you could display quite a few bands: ham, broadcast etc.
My gut feeling is that in a few years we’ll have sub-$299 16-bit SDRs which cover the whole of HF in one go with good dynamic range so this will be a reality.
As for 4 x 8k - would be a hell of PC but is very doable.
Simon Brown, G4ELI
https://www.sdr-radio.com
-----Original Message-----
From: main@SDR-Radio.groups.io <main@SDR-Radio.groups.io> On Behalf Of jdow
Sent: 10 July 2020 03:09
To: main@SDR-Radio.groups.io
Subject: Re: [SDR-Radio] Choose which Nvidia GPU to use if multiple GPUs are present. #feature
Too small for more than NFM. I like the idea of one big spectrum with sub-spectrums calculated on smaller chunks of data to cover areas of interest. A panoramic panel of 4 8k screens is sort of small for NFM monitoring - about
31-32 kHz per pixel.
Heh, four panoramic 8k screens above a fifth showing detail spectrum regions.
...


Bart AA7VA
 

On Thu, Jul 9, 2020 at 07:09 PM, jdow wrote:
"It depends on the target application. You want to present the user with the data he needs already emphasized rather than all the information that could be extracted from a random 500 MHz chunk of spectrum."
Somewhat on the idea as used by Flex 6x00 radios?