NVIDIA Publishes Signed Ampere Firmware To Finally Allow Accelerated Open-Source Support

[zexelon]

It's great to see More Then Nothing from Nvidia.
Being proprietary is costing them a lot of business to AMD.
Intel needs to redo their hardware architecture they continue to present in
different forms with special limitations features and new security vulnerabilities.
I believe the best hardware manufacture is AMD, better then Intel and Nvidia combined
in hardware and software situation.

Alas, the reality is being proprietary has cost Nvidia nothing. They have the largest market share (by far), and the highest profit margins (its embarrassing!) in the GPU or GPUC markets**!

Discrete GPU market:

	Q2'20	Q2'21	Q3'21
AMD	20%	17%	17%
Nvidia	80%	83%	83%

Nvidia has absolutely no need to support FOSS! As you can see Nvidia's market share has actually grown while the Linux community has been standing in the corner wishing they could even be close to the stage but not understanding why no one wants to bring the stage to them.

Its even more tilted in the GPUC market where last stat I saw Nvidia holds something like 97% market share. Ironically that is also an almost exclusively Linux market... and no one there cares at all about FOSS religious nuttery, they just need something that works and crunches numbers... if AMD would even show up at the table with ROCm they would be very welcomed (FOSS or not). Actually on that note its been very refreshing to see AMD actually putting effort into ROCm, I hope it reaches something even close to CUDA some day!

**Discounting Intel's embedded GPUs, its a tough one to get clear stats on as Intel by far dominates in shipments but a large number of those onboard GPUs get augmented with an AMD or Nvidia card.

[ATFx]

Alas, the reality is being proprietary has cost Nvidia nothing. They have the largest market share (by far), and the highest profit margins (its embarrassing!) in the GPU or GPUC markets**!

Discrete GPU market:

	Q2'20	Q2'21	Q3'21
AMD	20%	17%	17%
Nvidia	80%	83%	83%

Probably most of that is Windows because
Nvidia's Proprietary Beats AMD's Proprietary.
I was more so relating to, in time, a different time.
I also didn't factor where else Nvidia had been driving in Open Sourcing Software, they have a good chance of sustaining.

[sinepgib]

Yes, they go by what successful are doing, they are only stepping up open source support because
of AMD annihilating them in hardware, providing Open Source drivers to other Businesses.
Who prefer or have to have control of their own hardware to a source extent for their own products/services.
Data Centers and Supercomputers of any kind prefer it most for scalability.
Think of all of the Customers Enterprise or not that did not have Access to a Viable Alternative
for that because Intel wasn't making Graphics Cards.
AMD had terrible Proprietary Drivers and Graphics Hardware that made a terrible confliction for a Good choice.
Intel realizes the opportunity AMD is taking Advantage of, Nvidia sees both and says o ****
I'm pretty screwed being the oddball in what I offer I can't get away with anymore I'm going to Lose all of mine
If I don't change in that regard.

zexelon already addressed the annihilation part.
Regarding data centers and supercomputers, I don't understand by the way you write whether you're saying they use nvidia or AMD/Intel, but the reality is they use mostly nvidia and most GPGPU programs use CUDA. Even the likes of PyTorch don't support OpenCL and just recently started supporting ROCm.
In those segments it's not that much about control as long as things are fast, they'll be running in shared clusters in most cases anyway.

[sinepgib]

Probably most of that is Windows because
Nvidia's Proprietary Beats AMD's Proprietary.
I was more so relating to, in time, a different time.

So? nvidia doesn't care what OS you use, it cares if you pay them or not. The fraction of a fringe market they lose is irrelevant to their shareholders.

[ATFx]

So? nvidia doesn't care what OS you use, it cares if you pay them or not. The fraction of a fringe market they lose is irrelevant to their shareholders.

what postpatltkeaauu **** f.

[zexelon]

zexelon already addressed the annihilation part.
Regarding data centers and supercomputers, I don't understand by the way you write whether you're saying they use nvidia or AMD/Intel, but the reality is they use mostly nvidia and most GPGPU programs use CUDA. Even the likes of PyTorch don't support OpenCL and just recently started supporting ROCm.
In those segments it's not that much about control as long as things are fast, they'll be running in shared clusters in most cases anyway.

Not certain if the response was directed my way or not, but in the interest of clarity, something like 97% of the GPUC market uses Nvidia and as you said, its because of the shear performance. There are some tests being done on a large cluster scale using the AMD instinct cards but they are so niche and experimental atm that they can hardly be called a production system. I think it was Sandia National Labs that was deploying this AMD based system, but dont quote me on that one!

On a personal level I will freely admit that I am 100% a Nvidia user and that is 80% driven by CUDA requirements atm. I would very much like AMD to be part of my companies GPUC systems, but unfortunately its just not there. There is nothing out there atm that can touch a 3090 for performance in AI/ML training... except the A5000 (and higher) Nvidia cards. Now that Nvidia also supports VM'ing their cards with no hacks its actually delightful to setup everything from dev to deployment with them.

Now when it comes to the perfect system... AMD Threadripper with several Nvidia 3090 cards can not in any way shape or form be touched... unless you can get a DGX system... but they literally are gold colored because mortals cant obtain them except perhaps through a Faustian bargain with Jensen Huang.

[sinepgib]

Not certain if the response was directed my way or not, but in the interest of clarity, something like 97% of the GPUC market uses Nvidia and as you said, its because of the shear performance. There are some tests being done on a large cluster scale using the AMD instinct cards but they are so niche and experimental atm that they can hardly be called a production system. I think it was Sandia National Labs that was deploying this AMD based system, but dont quote me on that one!

On a personal level I will freely admit that I am 100% a Nvidia user and that is 80% driven by CUDA requirements atm. I would very much like AMD to be part of my companies GPUC systems, but unfortunately its just not there. There is nothing out there atm that can touch a 3090 for performance in AI/ML training... except the A5000 (and higher) Nvidia cards. Now that Nvidia also supports VM'ing their cards with no hacks its actually delightful to setup everything from dev to deployment with them.

Now when it comes to the perfect system... AMD Threadripper with several Nvidia 3090 cards can not in any way shape or form be touched... unless you can get a DGX system... but they literally are gold colored because mortals cant obtain them except perhaps through a Faustian bargain with Jensen Huang.

It wasn't directed to you, I mentioned you because there was no point in me looking for the actual numbers after you posted them, but I'm glad you made this post because it's damn interesting.

[sinepgib]

what postpatltkeaauu **** f.

Hmmmm I have no idea what you're trying to say, sorry.

[birdie]

Alas, the reality is being proprietary has cost Nvidia nothing. They have the largest market share (by far), and the highest profit margins (its embarrassing!) in the GPU or GPUC markets**!

Discrete GPU market:

	Q2'20	Q2'21	Q3'21
AMD	20%	17%	17%
Nvidia	80%	83%	83%

Nvidia has absolutely no need to support FOSS! As you can see Nvidia's market share has actually grown while the Linux community has been standing in the corner wishing they could even be close to the stage but not understanding why no one wants to bring the stage to them.

Its even more tilted in the GPUC market where last stat I saw Nvidia holds something like 97% market share. Ironically that is also an almost exclusively Linux market... and no one there cares at all about FOSS religious nuttery, they just need something that works and crunches numbers... if AMD would even show up at the table with ROCm they would be very welcomed (FOSS or not). Actually on that note its been very refreshing to see AMD actually putting effort into ROCm, I hope it reaches something even close to CUDA some day!

**Discounting Intel's embedded GPUs, its a tough one to get clear stats on as Intel by far dominates in shipments but a large number of those onboard GPUs get augmented with an AMD or Nvidia card.

THIS IS FAKE NEWS

(Linux lovers are a lot like Trump supporters: they hate facts and live in alternative Trumpesque/Linux reality where if you don't support Linux you're losing).

[Volker Schmidt]

This phrase is so wrong SO MUCH.
Disclaimer, using Radeon 5700 XT currently.
But previous to that ALL my GPU were nvidia. Starting first generation Nvidia cards. For over a decade of all together use of Nvidia cards, they had drivers that just worked.
I switched from Nvidia to AMD, and some games in Steam stopped working all together. And still don't work.
So NO. Nvidia's closed sourced driver is GOOD.

With that said, i can live without couple of games and prefer to have freedom over functionality.

Being a 5700XT user myself, can you list a few games in steam which worked under Nvidia, but not with the 5700 XT? I currently have quite a good user experience at this point.