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

[Dr. Dro]

it's not that difficult to collect prices online for a couple of regions, that's insanely less effort than testing cards on multiple games and settings. Were talking about a couple of minutes compared to countless hours of testing, would make no difference. That doesn't seem like a valid argument to me. A simple chart would do it, you get charts for everything nowadays, 99% less relevant to viewers than local pricing

Laziness mixed with anglocentrism

The tech market has always placed maximum emphasis in the United States and most of the time, Western Europe. Not saying this is right, I absolutely hate that, but the closest you could get to automating these charts is by using Amazon's API. Then you would basically get real time data from Amazon listings. Which is one large storefront... from America.

MSRP is still the closest we got to official, expected pricing. It isn't perfect, and W1zz at least has been known to go out of his way to do some preliminary market research before writing reviews, at least since Covid and the crypto boom completely threw any balance in the GPU market by the wayside.

 

[Bomby569]

The tech market has always placed maximum emphasis in the United States and most of the time, Western Europe. Not saying this is right, I absolutely hate that, but the closest you could get to automating these charts is by using Amazon's API. Then you would basically get real time data from Amazon listings. Which is one large storefront... from America.

MSRP is still the closest we got to official, expected pricing. It isn't perfect, and W1zz at least has been known to go out of his way to do some preliminary market research before writing reviews, at least since Covid and the crypto boom completely threw any balance in the GPU market by the wayside.

my point is not to complain about no publication in particular, just that some conclusions may not make sense in all regions, but they still want a global audience. That's it.

 

[lilhasselhoffer]

it's not that difficult to collect prices online for a couple of regions, that's insanely less effort than testing cards on multiple games and settings. Were talking about a couple of minutes compared to countless hours of testing, would make no difference. That doesn't seem like a valid argument to me. A simple chart would do it, you get charts for everything nowadays, 99% less relevant to viewers than local pricing

Let me school you on reality. I can buy a card at two Best Buys, 1 Microcenter, 2 main online sites, and a variety of secondary sellers. I live in one state, in one city, and only am willing to travel 30 minutes in any one direction. In those retailers I've got 3 different prices for some of these cards. Anglocentrism is your fancy way of saying racist...when the point is literally that when you press print on a review it should be factually accurate. In a market situation that factual accuracy is literally dying every second it is up...

If you'd like to argue, fine. Make your own review site, and demonstrate the work. There are already services out there that will scrape retailers for pricing...and that's not the point of a review. A review tells you only the base price, generally the manufacturer suggested retail price, because any other thing is going to be inaccurate by virtue of existing. Good luck on that magic wand you require to not be anglocentric...because unreasonable people with unreasonable expectations shouting about racism generally get what they deserve. For the record, that's being ignored because there's literally no pleasing them. That's by definition, because the second you cave to any of their demands the goal posts move farther. Next moaning will be about not having the right retailers, then not having the right currency conversions, then.... It never freaking ends.

 

[Bomby569]

Let me school you on reality. I can buy a card at two Best Buys, 1 Microcenter, 2 main online sites, and a variety of secondary sellers. I live in one state, in one city, and only am willing to travel 30 minutes in any one direction. In those retailers I've got 3 different prices for some of these cards. Anglocentrism is your fancy way of saying racist...when the point is literally that when you press print on a review it should be factually accurate. In a market situation that factual accuracy is literally dying every second it is up...

If you'd like to argue, fine. Make your own review site, and demonstrate the work. There are already services out there that will scrape retailers for pricing...and that's not the point of a review. A review tells you only the base price, generally the manufacturer suggested retail price, because any other thing is going to be inaccurate by virtue of existing. Good luck on that magic wand you require to not be anglocentric...because unreasonable people with unreasonable expectations shouting about racism generally get what they deserve. For the record, that's being ignored because there's literally no pleasing them. That's by definition, because the second you cave to any of their demands the goal posts move farther. Next moaning will be about not having the right retailers, then not having the right currency conversions, then.... It never freaking ends.

what does shopping at the cheapest store has to do with the differences in regional pricing? Nothing.

Btw you should really research the difference between racism and xenophobia, unless you read somewhere i wanted the price of gpus sold only to black people. Not that xenophobia was even an issue, but if you want to be crazy about at least make an effort to be more precise.

 

[95Viper]

Get back on topic.
Stop the bickering/arguing and personal attacks... you don't like something posted, report it and the moderation team will handle the problem.

 

[Mr_Engineer]

AMD should be able to claw back some market share with RX 8000 series especially if the price/performance/power draw is good. If the process used is TSMC 3nm then that will help a lot with efficiency and performance. Also, it will need to have at least similar RT performance comparable Nvidia cards and lower prices.

I am more excited about Strix Halo with its powerful iGPU, I think it should do very well for laptops and mini-PCs. This APU can help increase AMD market share also.

 

[Scattergrunt]

AMD should be able to claw back some market share with RX 8000 series especially if the price/performance/power draw is good. If the process used is TSMC 3nm then that will help a lot with efficiency and performance. Also, it will need to have at least similar RT performance comparable Nvidia cards and lower prices.

I am more excited about Strix Halo with its powerful iGPU, I think it should do very well for laptops and mini-PCs. This APU can help increase AMD market share also.

Yea, I'm not really worried about Nvidia being topdog forever market share wise (I'm sure they'll always have the dominant share, but not having this sort of level of dominance, is what I mean.) The 90% is only that way for a few reasons, and some of those reasons are not permanent, especially now that Intel has given Nvidia a good kick in the balls with the B580 (and possibly made AMD also more confident in their new midrange centric tactic with the RX 8000 series, who knows). If the RX 8000 series sells well by AMD standards for GPUs then that 90% probably won't last for too many quarters (maybe, just kind of making some hardly educated guesses here which is mostly opinion lol)

 

[Zach_01]

AMD should be able to claw back some market share with RX 8000 series especially if the price/performance/power draw is good. If the process used is TSMC 3nm then that will help a lot with efficiency and performance. Also, it will need to have at least similar RT performance comparable Nvidia cards and lower prices.

I am more excited about Strix Halo with its powerful iGPU, I think it should do very well for laptops and mini-PCs. This APU can help increase AMD market share also.

Apparently, Strix Point/Halo will be RX8000series iGPU (RDNA 3.5) on APUs and RDNA4 RX9000 for desktops dGPUs
Also, very unlikely those to be on 3nm. Word around saying on 5nm. I say maybe 4nm but wont be cheap.

AMD RDNA 3.5 Powers Radeon RX 8000 for Mobile, RDNA 4 Drives RX 9000 Desktop Series

AMD's interim RDNA 3.5 architecture will power the Radeon RX 8000 series integrated graphics in "Strix Halo" mobile processors, while the more advanced RDNA 4 architecture is reserved for the higher-tier Radeon RX 9000 series of discrete graphics, according to @9550pro on X. We previously...

www.techpowerup.com

 

[Dr. Dro]

AMD should be able to claw back some market share with RX 8000 series especially if the price/performance/power draw is good. If the process used is TSMC 3nm then that will help a lot with efficiency and performance. Also, it will need to have at least similar RT performance comparable Nvidia cards and lower prices.

I am more excited about Strix Halo with its powerful iGPU, I think it should do very well for laptops and mini-PCs. This APU can help increase AMD market share also.

I find it rather unlikely it will use a 3 nm node with the prices they are targeting. If it does, you can expect availability to be quite low, and prices higher than they ought to be.

 

[JustBenching]

Which Radeon RX6000/7000 gpus did you use to say something like this ?

Rx6700s. Had it from September 2022, most of the issue were solved in March 2024 drivers. That's 1.5 years...

 

[Frick]

Rx6700s. Had it from September 2022, most of the issue were solved in March 2024 drivers. That's 1.5 years...

I got my 6950xt september last year and the only problems I've had was when I played around with undervolting.

 

[Mr_Engineer]

Also, very unlikely those to be on 3nm. Word around saying on 5nm. I say maybe 4nm but wont be cheap.

For desktop, an RDNA 4 CPU using 4nm monolithic design would be ok (at least much better than rx 7600xt that used 6nm)

But if AMD wants to compete and do well in laptops, Strix Halo has to use 3nm. As I assume it is not only intended as an Intel competitor, but also an Apple competitor who already have their (excellent) M4 chip that uses second generation 3nm process from TSMC (Apple has a lot of cash).

I think Strix Halo falls behind M4 series on performance and energy efficiency if it's not using a 3nm process.

 

[JustBenching]

I got my 6950xt september last year and the only problems I've had was when I played around with undervolting.

That doesn't mean I didn't have issues. I mean i had issues with both nvidia and amd (dx12 crashes, black screens etc.) but with nvidia a new driver within a couple of days usually solves my issue. With amd I had to wait for more than a year.

 

[Frick]

That doesn't mean I didn't have issues. I mean i had issues with both nvidia and amd (dx12 crashes, black screens etc.) but with nvidia a new driver within a couple of days usually solves my issue. With amd I had to wait for more than a year.

Sure, just wanted to give my experience.

 

[igormp]

For desktop, an RDNA 4 CPU using 4nm monolithic design would be ok (at least much better than rx 7600xt that used 6nm)

But if AMD wants to compete and do well in laptops, Strix Halo has to use 3nm. As I assume it is not only intended as an Intel competitor, but also an Apple competitor who already have their (excellent) M4 chip that uses second generation 3nm process from TSMC (Apple has a lot of cash).

I think Strix Halo falls behind M4 series on performance and energy efficiency if it's not using a 3nm process.

Strix Halo won't really be competing with Apple's offerings tho. A M4 Max is a sub-80W chip with double the memory bandwidth, whereas strix halo is 100W+. Power efficiency wise those are not in the same class at all.
It'll likely have better MT performance (which is the bare minimum given the power target and the extra cores), but will still fall behind in ST and GPU perf.

The Pro chips are sub 30W, so totally out of range for the halo lineup.

 

[Mr_Engineer]

Strix Halo won't really be competing with Apple's offerings tho. A M4 Max is a sub-80W chip with double the memory bandwidth, whereas strix halo is 100W+. Power efficiency wise those are not in the same class at all.
It'll likely have better MT performance (which is the bare minimum given the power target and the extra cores), but will still fall behind in ST and GPU perf.

The Pro chips are sub 30W, so totally out of range for the halo lineup.

Reportedly there will be cut down versions of Strix Halo, for example 12 cores/32 CUs (below 100W?), or less, compared to the maxed out Strix Halo with 16 cores / 40 CUs.

Regarding memory bandwidth, I didn't check what it will be for M4 Max, but if Strix Halo will uses a 256bit bus, then the memory bandwidth should be sufficient to feed the cores and the powerful iGPU.

We'll have to wait and see.

 

[igormp]

Reportedly there will be cut down versions of Strix Halo, for example 12 cores/32 CUs (below 100W?), or less, compared to the maxed out Strix Halo with 16 cores / 40 CUs.

Regarding memory bandwidth, I didn't check what it will be for M4 Max, but if Strix Halo will uses a 256bit bus, then the memory bandwidth should be sufficient to feed the cores and the powerful iGPU.

We'll have to wait and see.

Even a cut down version would likely be over 50W, which makes it way more power hungry than an M4 Pro, as an example.
The M4 Max has a 512-bit memory bus, double that of strix halo. The M4 Pro has 256-bit, but it's also a sub-30W chip.

For strix halo to me more efficient, AMD would need to make it monolithic, which I'm not sure it's a design cost they want to get into at this time.

 

[Mr_Engineer]

I did some checking, and found testing done by this website that mentions that M4 Pro uses up to 46W, but averages 40W.

Depending on the application tested, the M4 Pro (12 and 14 cores) was anywhere between 10% to 50% faster in CPU Multicore than Ryzen AI 370 (12 cores). But the iGPU is where it was much faster and much more power efficient than the Ryzen AI 370 which uses a monolithic design on 4nm (compared to M4 Pro at 3nm).

I agree for Strix Halo to be anywhere near M4 Pro/Max in performance and power efficiency, it has to be monolithic and use a TSMC 3nm process.

It's amazing to me how a 40W M4 chip has an iGPU as fast or faster than a mobile RTX 4060/4070 dGPU (up to 115W).

Power efficiency is very important for tablets/laptops and Apple seems to be way ahead, and will probably eat up more market share from Intel/AMD due to that.

Imagine Apple releasing a discrete graphics card, even a 200W card may end up faster than 300/400W cards from Nvidia and AMD.

 

[Bomby569]

Imagine Apple releasing a discrete graphics card, even a 200W card may end up faster than 300/400W cards from Nvidia and AMD.

it would cost 5000usd, no partner models, construction quality would be shit despite the price tag and you couldn't repair it.
Ah, the dream

 

[JustBenching]

I did some checking, and found testing done by this website that mentions that M4 Pro uses up to 46W, but averages 40W.

Depending on the application tested, the M4 Pro (12 and 14 cores) was anywhere between 10% to 50% faster in CPU Multicore than Ryzen AI 370 (12 cores). But the iGPU is where it was much faster and much more power efficient than the Ryzen AI 370 which uses a monolithic design on 4nm (compared to M4 Pro at 3nm).

I agree for Strix Halo to be anywhere near M4 Pro/Max in performance and power efficiency, it has to be monolithic and use a TSMC 3nm process.

It's amazing to me how a 40W M4 chip has an iGPU as fast or faster than a mobile RTX 4060/4070 dGPU (up to 115W).

Power efficiency is very important for tablets/laptops and Apple seems to be way ahead, and will probably eat up more market share from Intel/AMD due to that.

Imagine Apple releasing a discrete graphics card, even a 200W card may end up faster than 300/400W cards from Nvidia and AMD.

M4 pro and max are way, way, way way bigger than the Ryzen 370. There is no such thing as magic, and you can't compare chips based on core counts but based on transistors. The m4 pro is not as efficient as it looks, it's just a HUGE damn chip. More transistors = more performance at similar power.

 

[Zazigalka]

AMD will compete in mid-range, but we've been in a situation like this before (rdna1 vs turing), let's not kid ourselves that focusing on mid-range really helps them claw back marketshare.

 

[Mr_Engineer]

M4 pro and max are way, way, way way bigger than the Ryzen 370. There is no such thing as magic, and you can't compare chips based on core counts but based on transistors. The m4 pro is not as efficient as it looks, it's just a HUGE damn chip. More transistors = more performance at similar power.

I disagree. I mentioned the number of cores as a reference point, at the end of the day the consumer doesn't care if the chip inside is 10 times bigger or smaller as long as it fits the laptop/machine. They care about price, performance, and energy efficiency. Yes, more transistors can mean higher performance, that's why I mentioned that Apple has an advantage as they are using a denser and more power efficient 3nm process compared the 4nm process used for Ryzen AI 370.

While the M4 Pro is probably much larger as you mentioned, the laptops tested were very similar in weight (MacBook Pro M4: 3.4lb - 1.6kg) and (ASUS Zenbook S 16: 3.31lb - 1.5kg). So the MacBook using the "huge" M4 Pro chip was faster, more power efficient, and weights about the same:

I think AMD is on the right track with their newer APU releases for handhelds/laptops (Z1, Z2, Ryzen AI 370, Strix Halo etc) should be able to increase their market share in these segments.

Also, Strix Halo should be a better competitor for Apple M4 Pro/Ultra/Max, but it needs to be built on 3nm, otherwise it will fall short against competing solutions from Apple and ARM in general.

 

[JustBenching]

I disagree. I mentioned the number of cores as a reference point, at the end of the day the consumer doesn't care if the chip inside is 10 times bigger or smaller as long as it fits the laptop/machine. They care about price, performance, and energy efficiency. Yes, more transistors can mean higher performance, that's why I mentioned that Apple has an advantage as they are using a denser and more power efficient 3nm process compared the 4nm process used for Ryzen AI 370.

While the M4 Pro is probably much larger as you mentioned, the laptops tested were very similar in weight (MacBook Pro M4: 3.4lb - 1.6kg) and (ASUS Zenbook S 16: 3.31lb - 1.5kg). So the MacBook using the "huge" M4 Pro chip was faster, more power efficient, and weights about the same:

View attachment 378551

I think AMD is on the right track with their newer APU releases for handhelds/laptops (Z1, Z2, Ryzen AI 370, Strix Halo etc) should be able to increase their market share in these segments.

Also, Strix Halo should be a better competitor for Apple M4 Pro/Ultra/Max, but it needs to be built on 3nm, otherwise it will fall short against competing solutions from Apple and ARM in general.

You are comparing the 14" macbook vs a 16" amd laptop. The 16" macbook weights over 2kg (2128 to be exact). The base version is also more expensive while having less ram and storage. So yes, it does matter that it's a much bigger chip because it's' more expensive, a lot more expensive. If you go with matching specs the m4 16" costs 3.5k$ vs 2k for the zenbook.

AMD will compete in mid-range, but we've been in a situation like this before (rdna1 vs turing), let's not kid ourselves that focusing on mid-range really helps them claw back marketshare.

I don't even understand what focusing on mid range even means. Does that mean that they weren't focused there before and so their midrange offerings were crap?

 

[Zazigalka]

I don't even understand what focusing on mid range even means. Does that mean that they weren't focused there before and so their midrange offerings were crap?

a nicer way to say they're not gonna make anything faster than what they already have.

 

[Mr_Engineer]

You are comparing the 14" macbook vs a 16" amd laptop. The 16" macbook weights over 2kg (2128 to be exact). The base version is also more expensive while having less ram and storage. So yes, it does matter that it's a much bigger chip because it's' more expensive, a lot more expensive. If you go with matching specs the m4 16" costs 3.5k$ vs 2k for the zenbook.

So AMD needs a larger chip to compete in this segment and increase market share, especially that their laptop dGPUs are not selling much compared to Nvidia laptop chips. For this reason I see AMD having a better chance competing in laptops and gain market share with an APU that includes a powerful iGPU, which is what they are doing with Strix Halo (they should have done that 10 years ago).

Now we'll have to wait and see how it performs.