Nvidia's GPU market share hits 90% in Q4 2024 (gets closer to full monopoly)

[Rover4444]

I hope AMD and Intel gets a higher marketshare in the following months, the have a very good chance. Battlemage and 9070 has so far been a great success and Nvidia struggles wirh production due to gddr7.

Is it really due to GDDR7? That'd be a relief. I hope NVIDIA cards come back in stock at MSRP, and then for a lot of things after that. ROCm support for the 9000 series especially.

WHERE IS THE COMPUTE AMD IT'S BEEN COOKED FOR THREE MONTHS

 

[Cowboystrekk]

Is it really due to GDDR7? That'd be a relief. I hope NVIDIA cards come back in stock at MSRP, and then for a lot of things after that. ROCm support for the 9000 series especially.

WHERE IS THE COMPUTE AMD IT'S BEEN COOKED FOR THREE MONTHS

From what I've rrad gddr7 supply is limited atm and nvifia prioritizes AI over gamers. The node they use from tsmc is mature and has good availability from what I've heard.

 

[igormp]

A new architecture changes the internal workings of a chip. A refresh only extends on some functionality. X3D can be considered a refresh. Or RDNA 3.5 over RDNA 3. But XT CPUs and Super GPUs cannot, imo, because nothing changed internally. I guess it's semantics from this point on.

Well, we have different understandings then. But yeah, continuing this would be just bickering over semantics.

 

[WhenMusicAttacks]

5070 Ti is definitely the better value at $150 more than the 9070 XT. Imagine being on RDNA 3 and learning you can't run FSR 4 while every card since Turing can do DLSS 4. Maybe it won't finewine like AMD does, but it'll be able to run ANYTHING you can think of and you can count on support a decade from now. AMD can't even support ROCm day one and keep it up for more than five years for most of their cards.

RDNA 3 can run fsr4 already

 

[Rover4444]

RDNA 3 can run fsr4 already

Can it?

 

[Event Horizon]

Can it?

I'm pretty sure it can't unless they develop an FSR4-lite version that doesn't require the new instructions that RDNA4 has. Even then the quality and/or performance uplift may suffer.

 

[AusWolf]

Well, we have different understandings then. But yeah, continuing this would be just bickering over semantics.

Some perfect examples are the Radeon R9 280X and the HD 7970. Or the GTX 770 and 680. I don't consider them a refresh, either. But yeah, let's leave it at that.

 

[igormp]

Some perfect examples are the Radeon R9 280X and the HD 7970. Or the GTX 770 and 680. I don't consider them a refresh, either. But yeah, let's leave it at that.

The specific examples you have are often just referred as "rebrands".
The convention to call something a "refresh" is regarding minor improvements that don't bring any architectural change. You are entitled to your opinions, but the convention is already in place and is how most people communicate:

Intel 14th Gen Raptor Lake Refresh Full Lineup Leaks Out

Six Intel 14th Gen Core Raptor Lake-S Refresh SKUs were prematurely listed by an e-tailer late last month, with the already leaked Core i9-14900K CPU model popping up again (a week later) via CPU-Z and Geekbench tests. This higher end K and KF product lineup is expected to launch at some point...

www.techpowerup.com

AMD to refresh socket AM4 with Ryzen '5000 XT' CPUs — two new iGPU-less Ryzen 8000-series processors also announced

AMD is still committed to AM4, for the budget sector.

www.tomshardware.com

 

[AusWolf]

The specific examples you have are often just referred as "rebrands".
The convention to call something a "refresh" is regarding minor improvements that don't bring any architectural change. You are entitled to your opinions, but the convention is already in place and is how most people communicate:

Intel 14th Gen Raptor Lake Refresh Full Lineup Leaks Out

Six Intel 14th Gen Core Raptor Lake-S Refresh SKUs were prematurely listed by an e-tailer late last month, with the already leaked Core i9-14900K CPU model popping up again (a week later) via CPU-Z and Geekbench tests. This higher end K and KF product lineup is expected to launch at some point...

www.techpowerup.com

AMD to refresh socket AM4 with Ryzen '5000 XT' CPUs — two new iGPU-less Ryzen 8000-series processors also announced

AMD is still committed to AM4, for the budget sector.

www.tomshardware.com

That's misleading as heck, imo. I'll stick to my own terms if you don't mind.

 

[Visible Noise]

RDNA 3 can run fsr4 already

Citation needed.

 

[AusWolf]

Citation needed.

Until someone proves otherwise:

I'm pretty sure it can't unless they develop an FSR4-lite version that doesn't require the new instructions that RDNA4 has. Even then the quality and/or performance uplift may suffer.

It would be nice to use RDNA 3's AI cores for something at least. FSR 3.5 perhaps?

 

[igormp]

Until someone proves otherwise:
View attachment 389593


It would be nice to use RDNA 3's AI cores for something at least. FSR 3.5 perhaps?

RDNA 3 doesn't really have "AI cores". It has instructions to operate on matrices, but it runs on the regular vector execution units.

 

[AusWolf]

RDNA 3 doesn't really have "AI cores". It has instructions to operate on matrices, but it runs on the regular vector execution units.

I've heard people saying that before, but then what's this?

 

[Visible Noise]

I've heard people saying that before, but then what's this?
View attachment 389597

You were lied to by a corporation.

 

[AusWolf]

You were lied to by a corporation.

Really? What's this, then?

 

[Hecate91]

Really? What's this, then?
View attachment 389601

FSR4 would have to be ported to bf16, iirc rdna3 doesn't support fp8.

AMD did claim they might be able to add FSR4 to rdna3, but AMD never lied about it, at least they show block diagrams and don't hide the depreciation of features.

 

[wolf]

at least they show block diagrams and don't hide the deprecation of features.

Who doesn't show block diagrams, and who is hiding the deprecation of features?

 

[Visible Noise]

Really? What's this, then?

WMMA matrix math instructions running on a bog standard vector unit. The next slide in the deck explains it.

AMD doesn’t have dedicated tensor hardware.

Tensor - Wikipedia

en.wikipedia.org

 

[igormp]

I've heard people saying that before, but then what's this?
View attachment 389597

Misleading marketing, those still live within the regular ALUs, hence why their rate of execution is not much faster than the regular fp32 rate.
That "64" currelates to the number of WGPs (or is it CUs? I don't recall exactly).

WMMA matrix math instructions running on a bog standard vector unit. The next slide in the deck explains it.

AMD doesn’t have dedicated tensor hardware.

Tensor - Wikipedia

en.wikipedia.org

RDNA 4 now has it, and CDNA has had those for quite some time as well.

 

[Visible Noise]

RDNA 4 now has it, and CDNA has had those for quite some time as well.

Yes, I assumed the discussion was limited to RDNA 3.

Who is hiding the deprecation of features?

You know where he’s going. Don’t try to fix stupid as they say.

 

[oxrufiioxo]

Yes, I assumed the discussion was limited to RDNA 3.


You know where he’s going. Don’t try to fix stupid as they say.

Radeon Lovers finally get a 4070ti super like gpu and they bow out their chest and say come at me bruh.

I'm happy for them man they get to see what it's like to have a competent gpu without any major weaknesses.

 

[igormp]

Radeon Lovers finally get a 4070ti super like gpu and they bow out their chest and say come at me bruh.

I'm happy for them man they get to see what it's like to have a competent gpu without any major weaknesses.

Don't quote me on that, but I've heard from some acquaintances that work with GPU kernels that it seems to be impossible to achieve the max rated memory bandwidth for the 9070xt, in a similar fashion to what happened to Intel's Alchemist GPUs.

However, it could very well be just a skill issue in their part without the proper optimizations for the new ISA, but time will tell.

 

[Visible Noise]

Radeon Lovers finally get a 4070ti super like gpu and they bow out their chest and say come at me bruh.

I'm happy for them man they get to see what it's like to have a competent gpu without any major weaknesses.

Not entirely though. AMD still isn’t doing BVH traversal in hardware. Your phone probably is doing hardware traversal, but AMD GPUs are still doing traversal in shader code and repurposing the back end hardware for raytracing.

 

[dont whant to set it"']

Radeon Lovers finally get a 4070ti super like gpu and they bow out their chest and say come at me bruh.

I'm happy for them man they get to see what it's like to have a competent gpu without any major weaknesses.

Enjoy your washed up colours .

 

[Visible Noise]

Don't quote me on that, but I've heard from some acquaintances that work with GPU kernels that it seems to be impossible to achieve the max rated memory bandwidth for the 9070xt, in a similar fashion to what happened to Intel's Alchemist GPUs.

However, it could very well be just a skill issue in their part without the proper optimizations for the new ISA, but time will tell.

AMD GPUs can’t raytrace and texture at the same time due to back-end resource contention, maybe that keeps them from using full bandwidth.

I wish 3D mark still had fill rate and texture rate tests.

Enjoy your washed up colours .

We will, thank you. Did you have anything actually useful to add to the discussion, other than taking a dump in the thread?