NVIDIA RTX 6000 Ada Generation vs. Radeon PRO Performance On Ubuntu Linux 24.04 LTS

[ktecho]

Ordered one of these in a new server to run AI workloads. Yes, it cost an arm and a leg... but so does a Ferrari ... and to be honest this will generate much more revenue then owning a Ferrari ever will!

I'm curious how will you generate revenue with one of those. Could you elaborate? Thanks!

[iliqiliev]

Can you compare temperatures reported by the GPUs? Different vendors and even different families of the same brand report temperatures differently. I'm not sure if a thermal couple is used tho.

[qarium]

Michael please fix the blender benchmark graphs to split the benchmarks between cuda,optix,hip makes no sense at all.
every scene benchmarkes should contain all graphc from cuda to optix and hip to better compare performance.
also maybe add HIP-RT as alternative to optix...

also the claim of michael that the AMD PRO w7900 costs 4000 is wrong here in germany it costs
only 3238€

https://geizhals.de/amd-radeon-pro-w7900-a2948314.html

if we calculate with the correct price this: "The NVIDIA RTX 6000 Ada Generation was delivering 54% higher performance compared to the Radeon PRO W7900" does not look so good...

3238€*1,54=4986€ this is what the NVIDIA RTX 6000 Ada should cost.. but it costs "7984€"
as one can see here:

https://geizhals.de/pny-rtx-6000-ada-vcnrtx6000ada-sb-a2881054.html

with AMD you literally save 3000€ on the same performance.

[qarium]

Welp, I officially feel good about today

Ordered one of these in a new server to run AI workloads. Yes, it cost an arm and a leg... but so does a Ferrari ... and to be honest this will generate much more revenue then owning a Ferrari ever will!

I have looked longingly at AMD to provide any sort of competition in the GPGPU space but alas they are still quite a ways behind... it took them decades to become competitive and eventually become a threat to Intel, maybe they can eventually do the same in the GPGPU space.

if you check price search engines

https://geizhals.de/pny-rtx-6000-ada-vcnrtx6000ada-sb-a2881054.html

https://geizhals.de/amd-radeon-pro-w7900-a2948314.html

then you discover that at the same performance level nvidia costs 3000€ more clearly not worth it for AI...

for AI workloads its clearly better to just buy more cards and put 8 cards in one machine like tinycorp does it with the AI modell:

https://geizhals.de/gigabyte-radeon-pro-w7900-dual-slot-ai-top-48g-w7900-ai-top-48g-a3270768.html

" GIGABYTE Radeon PRO W7900 Dual Slot AI TOP 48G, 48GB GDDR6, 3x DP, mDP"

[sophisticles]

if we calculate with the correct price: [Bunch of nonsense deleted]

The "correct" price is the point at which profits are maximized by accounting for demand.

Clearly NVIDIA's prices are the correct prices because consumers are willing to pay the asking price to acquire the cards.

What is the "correct" price of a Ferrari?

The reality is that if you had a million dollars in hand and walked into a Ferrari dealership trying to buy a new car, you couldn't.

All new Ferrari cars are sold out 5-8 years in advance, depending on the model, so clearly Ferrari is pricing their cares correctly.

Likewise NVIDIA cards are in high demand and NVIDIA is selling as many as it can make, at significant profit margins, so clearly the prices are correct.

You just want to justify buying a lower performing card.

[sophisticles]

for AI workloads its clearly better to just buy more cards and put 8 cards in one machine like tinycorp does it with the AI modell:

Apparently Tiny has changed it's mind:

https://www.tomshardware.com/pc-components/gpus/tiny-corp-decides-to-make-both-amd-and-nvidia-versions-of-its-tinybox-ai-server-geforce-version-is-67-more-expensive

[qarium]

The "correct" price is the point at which profits are maximized by accounting for demand.
Clearly NVIDIA's prices are the correct prices because consumers are willing to pay the asking price to acquire the cards.
What is the "correct" price of a Ferrari?
The reality is that if you had a million dollars in hand and walked into a Ferrari dealership trying to buy a new car, you couldn't.
All new Ferrari cars are sold out 5-8 years in advance, depending on the model, so clearly Ferrari is pricing their cares correctly.
Likewise NVIDIA cards are in high demand and NVIDIA is selling as many as it can make, at significant profit margins, so clearly the prices are correct.
You just want to justify buying a lower performing card.

nvidia only get this high price because of the workstation market of digital artists who do 3D modelling and 3D CAD designs and so one.

nvidia does not get this high price because of AI... in AI you can perfectly parallel the workload
this means your argument "You just want to justify buying a lower performing card" is bullshit
it is bullshit because you do not save any money you just buy more cards and put more cards into your server.
as you see with tiny-corp they only do AI they just put 8 cards of AMD PRO W7900 AI 2 slots modells in their system to let it compete against the 4090 version.

you are just to stupid to make a difference between workstation market for digital artists who people use OptiX version of Blender and then the card pay itself because it is faster than the AMD card...

and the AI market who no one cares about your bullshit people want more VRAM than the 4090 has and people do not want to save money at all they just buy more cards and let the workloads run in parallel.

nvidia will have a rude awakening in the AI market if they discovering that brand loyalty is zero in this market.
and also they will discover that it does not matter at all if a single card is faster than the competition because they just buy more cards and run the workloads in parallel.

[qarium]

Apparently Tiny has changed it's mind:
https://www.tomshardware.com/pc-comp...more-expensive

what the hell are you talking about? they did only choose the 7900XTX over the AMD PRO W7900 because there is a blower model of 7900XTX and the AMD PRO W7900 is a 3-4 slot design and does not even fit in the case.

its 100% sure that tiny-corp will also offer the AMD PRO W7900 AI 2slot card with blower cooler.

this one:

https://www.gigabyte.com/de/Graphics-Card/W7900-AI-TOP-48G#kf

[sophisticles]

nvidia only get this high price because of the workstation market of digital artists who do 3D modelling and 3D CAD designs and so one.

nvidia does not get this high price because of AI... in AI you can perfectly parallel the workload
this means your argument "You just want to justify buying a lower performing card" is bullshit
it is bullshit because you do not save any money you just buy more cards and put more cards into your server.
as you see with tiny-corp they only do AI they just put 8 cards of AMD PRO W7900 AI 2 slots modells in their system to let it compete against the 4090 version.

you are just to stupid to make a difference between workstation market for digital artists who people use OptiX version of Blender and then the card pay itself because it is faster than the AMD card...

and the AI market who no one cares about your bullshit people want more VRAM than the 4090 has and people do not want to save money at all they just buy more cards and let the workloads run in parallel.

nvidia will have a rude awakening in the AI market if they discovering that brand loyalty is zero in this market.
and also they will discover that it does not matter at all if a single card is faster than the competition because they just buy more cards and run the workloads in parallel.

You are not capable of rational thinking, are you?

Why is it that NVIDIA owns 98% of the AI and Datacenter markets?

Because no one in their right minds is following your advice of putting in 8 AMD cards instead of 1 NVIDIA card, in fact they are doing the exact opposite of what you suggest.

[qarium]

You are not capable of rational thinking, are you?
Why is it that NVIDIA owns 98% of the AI and Datacenter markets?
Because no one in their right minds is following your advice of putting in 8 AMD cards instead of 1 NVIDIA card, in fact they are doing the exact opposite of what you suggest.

looks like you are the one who is not capable of rational thinking because i have bad bad news for you:

the Nvidia CUDA/Gameworks/Parallel Thread Execution PTX/NVPTX Intermediate Language
gets big competition because Khronos Group and Microsoft join forces against Nvidia:

https://www.phoronix.com/news/DirectX-Adopting-SPIR-V

microsoft abolish DXIL ("DirectX Intermediate Language") and go with SPIR-V in the future

this means Microsoft DirectX-Compute and Vulkan-Compute both will use SPIR-V in the future
to compete against Nvidia Parallel Thread Execution PTX ...

you see sophisticles you are the one who is not able to think rationally.