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

[AusWolf]

575w..

That is about 125w more than a 6 core X58 system with a solid tune running a 10GB load in Linpack Xtreme.

Through that little plug lol. I trust it for what I have, but I don't know if I would trust it for that.

It's what all 3 of my PCs consume together while crunching with BOINC.

 

[freeagent]

I am not sure what this system does, I broke my ZM-MFC2 that I used to show me what I would pull. HWiNFO shows 230-240w on the CPU, I would throw in another 100 for the board and mems, and maybe another 50 or so for a GPU in idle. I don't know. It feels like 350-400w just using my spidey senses. I have to finish paying for all this stuff and then I can get a killawatt lol.

 

[AusWolf]

I am not sure what this system does, I broke my ZM-MFC2 that I used to show me what I would pull. HWiNFO shows 230-240w on the CPU, I would throw in another 100 for the board and mems, and maybe another 50 or so for a GPU in idle. I don't know. It feels like 350-400w just using my spidey senses. I have to finish paying for all this stuff and then I can get a killawatt lol.

I just go by my smart meter's display when nothing's running in the house except for the 3 computers. Cool little gadget.

 

[igormp]

is that how it works? i don't think so, a 5090 is not leftover from AI cards, they are completely different. The wafers can be the same but that's another story, when you know the yields are bad you already committed the wafer to either a gaming card or AI card, you can't use a H whatever and make it into a 5090, i don't think so. Maybe i'm wrong and someone can correct me

You are thinking about their x100 chips, those are exclusively for the big AI farms.
However, the rest of the stack serves both the consumer and the enterprise market. There are no GB202 enterprise products yet, but there will soon be.
As an example, the same AD102 chip was used for both the 4090 and also the L40, where the 4090 used a lower bin of that chip.

 

[eidairaman1]

That's my point, if I need the specific television then the problem is either cable or the TV itself, I already said that. It's the same with m LG monitor, it's very picky with the inputs,.

Delusional

It's pretty apparent by now that they've not really cared about gaming graphics since at least Ampere.

They stopped caring with the launch of rtx.

 

[DirtyDingusMcgee]

All i know is this.

I'm glad a bought these two 7900 GRE's for 500 each when i did.

Suits my needs just fine. Go Red team!

 

[AusWolf]

I love the connector though. If the nitro 9070xt has a 12vh, im inclined to support the card (you know, by actually buying it, not with forum posts )

It seems your wish has come true.

Custom AMD Radeon RX 9070 XT Appears Powered by 12V-2x6 Connector

Days before releasing, Chinese leakers on Chiphell are showcasing a custom variant of AMD's upcoming Radeon RX 9070 XT with what appears to be a 12V-2x6 power connector. The custom AIB model is Sapphire Radeon RX 9070 XT NITRO+, which features a triple-fan cooling configuration and a departure...

www.techpowerup.com

 

[Red Hood]

is that how it works? i don't think so, a 5090 is not leftover from AI cards, they are completely different. The wafers can be the same but that's another story, when you know the yields are bad you already committed the wafer to either a gaming card or AI card, you can't use a H whatever and make it into a 5090, i don't think so. Maybe i'm wrong and someone can correct me

The 5090 is made from the GB202 chip which is designed for the workstation cards (6000 blackwell, 5880 blackwell etc or whatever they gonna call them)

The chips which have at least 21760 working cores can be used for the 5090 but the priority is the workstation cards

The 5090 can be used for games but its target is small ai companies and design studios

The B100 and B200 are from a different chip

 

[Bomby569]

You are thinking about their x100 chips, those are exclusively for the big AI farms.
However, the rest of the stack serves both the consumer and the enterprise market. There are no GB202 enterprise products yet, but there will soon be.
As an example, the same AD102 chip was used for both the 4090 and also the L40, where the 4090 used a lower bin of that chip.

it was obviously (apparently not because you didn't understood) an exaggeration to get a point across. Not sure why someone would take that seriously.
But my point stands, if they have some better cards "in the basement" i wouldn't be surprised, and to be clear this time i'm not talking at literal cards or a literal basement.

 

[TRINITAS]

If AMD does not have a significant market share, it is not because of the chips themselves. RDNA architectures are very good.
The problem is on the software side. Although in terms of gaming, AMD does not have too many problems, it is rather on the production side that the shoe pinches.
Just look at the poor performance of Radeons under Blender for example, or the very limited list of GPUs that can be used under Linux and Windows with the HIP API. We can also add the architectural division: RDNA and CDNA are different, and therefore it is not easy to make drivers/software adapted for both at the same time.
NVIDIA's strength comes mainly from its software support, and its universal exploitation on all its GPUs, unlike AMD.
AMD's plan to end the CDNA and RDNA architectures, to replace them with UDNA, in order to standardize everything is a relevant decision, and in my opinion, the final stretch that should not be missed. Intel, for its part, is only just starting out, so it makes sense that the productive performances are still very low. But at least, the Xe architecture is uniform, therefore usable in all sectors, which is a good point.

 

[AusWolf]

But my point stands, if they have some better cards "in the basement" i wouldn't be surprised

They always do. They just call them "Super" and release them a year later for maximum profit.

 

[Bomby569]

They always do. They just call them "Super" and release them a year later for maximum profit.

and "super ti's"

 

[wolf]

If you were pointing at me with that...... If you weren't pointing at me

I mean... Gestures broadly at thread. I wasn't pointing at you specifically, but you did see fit to answer which I think says something. I know you're one of the more rational here but can also perhaps be dragged down by others as can I.

In any case it was mostly meant in jest to poke fun at the heated nature of the debate which appeared to be going nowhere constructive.

 

[AusWolf]

I mean... Gestures broadly at thread. I wasn't pointing at you specifically, but you did see fit to answer which I think says something. I know you're one of the more rational here but can also perhaps be dragged down by others as can I.

In any case it was mostly meant in jest to poke fun at the heated nature of the debate which appeared to be going nowhere constructive.

Every conversation seems to get heated sooner or later the days. Such a shame, especially considering the lacklustre offerings these days.

 

[DAPUNISHER]

Mobile is 80% of PC sales. Nvidia has it on locked down. They dominate prebuilts too.

 

[Krit]

Big thumbs up for AMD that they decided to not overprice gpus this time around. As a result market share should increase now.

RX 9070 XT 100$ more expensive than RX 7800 XT and around 51% performance boost.
When RTX 5070 Ti is only 15-17% faster than RTX 4070 Ti Super.

 

[Scattergrunt]

Mobile is 80% of PC sales. Nvidia has it on locked down. They dominate prebuilts too.

AMD doesn't have a good reputation with laptop makers for some reason and the only AMD hardware I can distinctly remember being popular in prebuilts that wasn't a CPU was the RX 6600

 

[vacsati]

You might want to be more concerned about this

People tend to overreact to everything.
I guess this is a similar case to the one when they cried because of a few dozen melted connectors, while in reality not even a thousandth of a percent of the cards were affected.

 

[Dr. Dro]

Look, I'm not expecting Nvidia, or any other hardware manufacturer to test their products in every imaginable hardware configuration. But if something works, it works, and if something doesn't work, it doesn't work. And my TV doesn't work well with Nvidia, but it does with AMD and Intel just fine. It doesn't matter who you blame, the end result is the same: I have to use an AMD or Intel card if I want 4k 60 with true 8-bit colour.

So yes, I'm blaming the Nvidia driver. And no, I'm not saying that it's a universal issue, just my own. And that's that. If you don't like it, it's not my problem.

I think I know what you're talking about, I have a 2022 The Frame, similar to this one but being the 32 inch model it's comparably basic, 1080p 60 Hz only and the One Connect box is a simpler model with only 2 HDMI inputs and apparently a lot less processing speed (in other words, it lags balls).

I've had an issue with it that it often doesn't work correctly with RGB/Full range input. Colors look completely off with heavy RGB crush, if you try a calibration chart it'll be unable to reproduce almost half of the gradient, the three workarounds I found were to keep toggling between resolutions until it somehow looks normal, to enable HDR (doesn't fix but helps somewhat) or to use 4:4:4 limited color. I wonder if it's a bug, but it definitely affects modern Nvidia GPUs. Now that you mention it, it doesn't seem to look as bad on my old Mac mini (320M on 309.08 driver) with the same settings, haven't really tried it with my laptop's AMD graphics, I should do that sometime. Unlike the HDMI port, the USB-C port is controlled by the 5600H's graphics instead of the RTX 3050 for some reason, which gives me that option.

 

[AusWolf]

I think I know what you're talking about, I have a 2022 The Frame, similar to this one but being the 32 inch model it's comparably basic, 1080p 60 Hz only and the One Connect box is a simpler model with only 2 HDMI inputs and apparently a lot less processing speed (in other words, it lags balls).

I've had an issue with it that it often doesn't work correctly with RGB/Full range input. Colors look completely off with heavy RGB crush, if you try a calibration chart it'll be unable to reproduce almost half of the gradient, the three workarounds I found were to keep toggling between resolutions until it somehow looks normal, to enable HDR (doesn't fix but helps somewhat) or to use 4:4:4 limited color. I wonder if it's a bug, but it definitely affects modern Nvidia GPUs. Now that you mention it, it doesn't seem to look as bad on my old Mac mini (320M on 309.08 driver) with the same settings, haven't really tried it with my laptop's AMD graphics, I should do that sometime. Unlike the HDMI port, the USB-C port is controlled by the 5600H's graphics instead of the RTX 3050 for some reason, which gives me that option.

Exactly, thank you!

The workaround I'm using is the Intel Xe iGPU for image output, while still having the dGPU in the system. Nvidia's saving grace in my case is that dual graphics works flawlessly even on desktop. If I had an F-SKU Intel CPU, though, I'd be F*ed. Ironic.

 

[WhenMusicAttacks]

Oh god, i was not expecting these awful 1% lows because of vram issues from the 5070 12gb compared to the 9070 .. these are happening in older games, sometimes even without ray tracing, and GamersNexus also pointed out that these issues get worst the longer you play!

Some examples will be added with link to the right timestamp :

Cyberpunk:

Indiana Jones:

now, these are just noticed because the game actually end up stuttering&crashing, some other games just don't load textures and might end up even performing better that way without showing in the benchmark charts, as shwon with the HWUnboxed video about the 3070 vs 6800xt 2 years ago.