The Real Reason Nvidia Has Abandoned PC Gamers

[loki1944]

Well honestly devs need to make games less limiting to hardware then. AVX and RT only limit massive amounts of hardwarethat is perfectly capable of playing modern titles; case and point my i7 980X is perfectly fine running Robocop/KCD2/Mafia 3/Oblivion Remastered/Far Cry 6/GTA 5 remastered/Payday 3/WH40K Dark Tide/Hitman 3 but is hard locked by AVX in Borderlands 4 /Starfield/The Ladt of Us/Crimson Desert/ 007 (no big loss on this one).

 

[Endymio]

DLSS is just about Nvidia being too lazy to develop hardware

This is so ludicrously inept a statement I literally laughed aloud when reading it. If NVidia has been "too lazy to develop hardware", what's the rest of the world? I doubt even you yourself believe this tripe; it's merely a means to virtue-signal your hatred of a large, successful corporation.

 

[HardReset]

This is so ludicrously inept a statement I literally laughed aloud when reading it. If NVidia has been "too lazy to develop hardware", what's the rest of the world? I doubt even you yourself believe this tripe; it's merely a means to virtue-signal your hatred of a large, successful corporation.

Obviously you have absolutely no clue what you are talking about. Nvidia was supposed to launch Maxwell architecture 2012 but TSMC fkd up 20nm tech. Now Maxwell launched 2014. On 2016 Nvidia launched 95% same Maxwell architecture built on TSMC 16nm node. After 6 years of waiting, Nvidia released RTX 2000 series what at least contained Some architecture improvement. That is 6 year gap between GPU architectures for gamers.

Moving on today, what Nvidia does? We have DLSS version X+1 and some more imaginary frame generation. Where are Real improvements? Outside fake frames, image quality worsening and still too slow RT?

For AMD, Infinity cache and MCD+CCD design are both larger architecture improvements than Nvidia has done overall on last 10 years. AMD at least have have tried although both improvements never really succeeded for many other reasons.

 

[Endymio]

Obviously you have absolutely no clue what you are talking about.

Since DLSS was introduced, NVidia has introduced 3 new GPU generations and improved maximum performance by 330%. They had then, and they still have the fastest cards on the planet, and the most feature rich and capable. Next time think before you post.

For AMD, Infinity cache and MCD+CCD design are both larger architecture improvements than Nvidia has done overall on last 10 years. AMD at least have have tried....

And yet despite all that "trying", AMD still failed to take the crown from the "lazy" NVidia. You make my point for me.

 

[HardReset]

Since DLSS was introduced, NVidia has introduced 3 new GPU generations and improved maximum performance by 330%.

Yeah? See, better process nodes give bigger die sizes and improve performance. Guess why Nvidia is talking about being "so hard" to improve performance as process nodes no longer give huge gains? Because most of Nvidia performance increases come from process nodes, not architecture. No huge process node advancements, no improvements outside fake frames.

They had then, and they still have the fastest cards on the planet, and the most feature rich and capable. Next time think before you post.

You obviously didn't understand was about Nvidia vs Nvidia, not Nvidia vs others. There goes your point. Now I proved you actually don't understand what you are blabbering about.

And yet despite all that "trying", AMD still failed to take the crown from the "lazy" NVidia. You make my point for me.

AMD had much bigger things going on CPU side. Also with RDNA 3 AMD could have easily taken crown from Nvidia but they didn't bother, because people would have bought Nvidia anyway. Nvidia just use same architecture over and over while adding "improvements", like DLSS 5 because they are too lazy to develop real GPU improvements. That is my point and that is also fact.

 

[Endymio]

Yeah? See, better process nodes give bigger die sizes and improve performance.

And yet AMD, Intel, and every other wannabe gpu maker on the planet has also had access to those same process nodes, and they still lose out. You make my point for me.

BTW, who told you that new process nodes "give bigger die sizes"? There's no direct correlation between the two, except that the same transistor count on a newer node yields a smaller die. That is in fact, the only reason that price performance increases over time: new nodes are more expensive to manufacture on, but can use smaller dies.

Guess why Nvidia is talking about being "so hard" to improve performance as process nodes no longer give huge gains? Because most of Nvidia performance increases come from process nodes, not architecture.

Untrue, and irrelevant to this argument even were it true. All firms have access to these same nodes, yet no one has improved performance more than NVidia.

You obviously didn't understand was about Nvidia vs Nvidia, not Nvidia vs others.

There is no region of the multiverse in which this word salad makes sense.

Also with RDNA 3 AMD could have easily taken crown from Nvidia but they didn't bother

Ah, they "didn't bother"? Sounds like laziness to me.

 

[HardReset]

And yet AMD, Intel, and every other wannabe gpu maker on the planet has also had access to those same process nodes, and they still lose out. You make my point for me.

Once again, Nvidia vs Nvidia. No Nvidia vs others.

BTW, who told you that new process nodes "give bigger die sizes"? There's no direct correlation between the two, except that the same transistor count on a newer node yields a smaller die. That is in fact, the only reason that price performance increases over time: new nodes are more expensive to manufacture on, but can use smaller dies.

Better manufacturing processes generally enable bigger die sizes too and/or more transistor count.

Untrue, and irrelevant to this argument even if it were true.

True, read article. It clearly says Nvidia has been pretty much brute forcing more performance lately. Better manufacturing processes and still more power consumption. Now you are disagreeing with article without any real points.

There is no region of the multiverse in which this word salad makes sense.

Article is about Nvidia.

The Real Reason Nvidia Has Abandoned PC Gamers

See? How Nvidia has progressed. Not "against others" but "against itself".

Ah, they "didn't bother"? Sounds like laziness to me.

Because development costs were too high vs sales. Simple.

And because Nvidia know they can sell old stuff with minimal improvements, Nvidia is also lazy.

 

[Endymio]

Once again, Nvidia vs Nvidia. No Nvidia vs others.

At least attempt to make sense. Which of those NVidia is the lazy one, and which is the one that dramatically increased performance to keep the title of fastest GPU?

Better manufacturing processes generally enable bigger die sizes too and/or more transistor count.

This isn't even remotely correct. They enable more transistors per unit die size. And initially, they demand smaller dies, because yields are lower, and yields depend heavily on overall die size. But even on mature nodes, its false to say that more advanced nodes somehow "enable" larger dies.

True, read article. It clearly says Nvidia has been pretty much brute forcing more performance lately.

Oops, you didn't read the article. It stated that the 3090 "brute forced" much of its gains through higher power consumption ... but that the 4090 did the exact opposite -- increased performance 60% while reducing power consumption.

And you still ignore the point. Even were your claim true, it's irrelevant to this argument. AMD and Intel haven't taken the crown from NVidia, despite all their purported "architectural gains".

 

[HardReset]

At least attempt to make sense. Which of those NVidia is the lazy one, and which is the one that dramatically increased performance to keep the title of fastest GPU?

Um, Nvidia has had fastest CPU over a decade. And AMD abandoned high end over a decade go. How much Nvidia needs to do to keep title of fastest GPU when other competitors do not even try to have fastest GPU

This isn't even remotely correct. They enable more transistors per unit die size. And initially, they demand smaller dies, because yields are lower, and yields depend heavily on overall die size. But even on mature nodes, its false to say that more advanced nodes somehow "enable" larger dies.

What I say is correct. There is generally drawback of larger die having more defects and so often small dies are preferred. But still if large monolithic die is needed, modern processes often allow it better.

Oops, you didn't read the article. It stated that the 3090 "brute forced" much of its gains through higher power consumption ... but that the 4090 did the exact opposite -- increased performance 60% while reducing power consumption.

3000-series used pretty bad Samsung "8nm" tech whereas 4000 series used much better TSMC 5nm. 4000 series also mostly brute forced but manufacturing tech was so much better. Exactly like 980 vs 1080.

4090 vs 5090 had no big difference on process node. And surprisingly power vs performance looks bad.

Brute force, just like I said.

And you still ignore the point. Even were your claim true, it's irrelevant to this argument. AMD and Intel haven't taken the crown from NVidia, despite all their purported "architectural gains".

I don't. Using same GPU architecture with bigger die area gives more speed. However designing big chip still costs money. Intel is still reluctant to release bigger GPUs, not because they cannot but because designing bigger GPU costs money. Same for AMD.

 

[Endymio]

What I say is correct. There is generally drawback of larger die having more defects and so often small dies are preferred. But still if large monolithic die is needed, modern processes often allow it better.

Again: this isn't even remotely correct. In theory, any process node can be used for any die size. In practice, viability depends on yields, at which older, more mature nodes always perform better. Period.

In the last decade, "Lazy" NVidia has improved GPU performance more than anyone on the planet. Facts matter.

 

[Rapob]

Steam deck has almost 70% of the handheld PC market, I want to know the mental gymnastics you used to come to your conclusion that it didn't sell well. The market might not be huge, but it dominated the market it created. There are better handhelds, but they cost as much as legitimate gaming laptop, thermal throttle like crazy and have battery life issues because of the hardware you use. People look at the steam deck and only look at performance numbers. I hate to tell people this, but those performance numbers don't matter if the battery doesn't last long enough to make your first save.

The switch 2 didn't sell well because it's the better hand held. Most users actually keep it docked like a console. It sold well because Nintendo hordes it's IP. They didn't buy a switch because it was better, though bought it because it's the only way to play Mario Kart, Zelda and Pokemon.

Steam Deck sold like 4 million units in 4 years - That is like nothing compared to regular consoles sales. Nintendo Switch sold 4 million units in a matter of days when it launched.

Steam Deck don't hold 70% of the handheld PC market or even close. There is better handhelds than Deck and there is 100s if not 1000s of different handhelds now, which are just small form factor PCs anyway. Mobile PC market is growing fast, improving fast too, and 4 million units of Steam Deck don't account for 70% of the market - laughable claim.

Steam Deck is nothing but a low-end PC (extremely low-end you might say). I had a Steam Deck OLED and mostly used it for indies and emulation of old consoles and amiga games. Literally ran like pure crap in AAA games and the lack of FSR 4 is terrible, FSR 3.1 and older is useless, especially at low res which Deck uses. Looks so bad when enabled. FSR 3.1 is garbage for anything else than 4K/UHD, even 1440p looks too bad. DLSS is so much better. DLSS 2 looks far better than FSR 3.1

However, sold it again, for what I paid, after price hike. So cost me nothing to test it for months and months. Do you even have one? Or do you just defend it, because I have massive experience with Steam Deck v1 and I was not impressed. Just too slow.

Steam Deck was good for one thing: Streaming games using my Nvidia GPU from my primary PC. Maxed out AAA games streamed directly to the Deck and with OLED screen, looked good and battery lasted vastly longer this way.

The hardware on Steam Deck is just so bad, in so many ways. Too slow really.

Sream Deck 2 with full FSR 4+ support and beefier hardware can't come soon enough.
Rumours say 2027-2028 just in time for next gen consoles like PS6.

There is better handhelds than Deck but the real problem is AMD Z1/Z2 whichs which are not very powerful and lacks FSR 4 which is a must have on a handheld like this.

Hence why Switch 2 has DLSS. Gives better perf, saves power, or both. Steam Deck and current handhelds not having FSR 4 is a huge problem.

Tons of people use Switch 2 in handheld mode, sadly screen is kinda crap and we need an OLED version or Pro version ASAP.

Nvidias upcoming APUs with ARM/RTX combo is going to do damage in both handheld and mobile/laptop market soon.

If Nvidia truly wanted the console market, they could get it. They just did not bother about it yet and AMD took the measly scraps. Even AMD don't need the console market these days, they earn more per wafer on CPUs and enterprise GPUs. Lets see if AMD will bother to even renew the contract for next gen. AMD went from a small player to a big player in the last 7 years, before AMD market cap increased bigtime - CPUs is their prime focus today. PS5 and XSX contributed nicely in AMD financials years ago but today, not that much and they probably don't even need this market or AMD will demand a higher premium next time - with Nvidia having APUs too, things could get spicy - lets see if console market will go ARM next gen (or the generation after that)

Nintendo has shown that using ARM CPU is very much possible. x86 might face a problem in this market going forward. ARM eats into server market too, little by little.

 

[LimyG]

Since RDNA3 was chiplet design, AMD could have surpassed Nvidia quite easily. Equally sized CCD + MCD chiplets vs equally sized (vs CCD) Nvidia monolithic die would have been easy win for AMD. However developing huge GPU is expensive and AMD have long time just concentrated on mid end market where major profits are. AMD just decided to stick with strategy where high end niche is abandoned.

It remains to be seen how long this AI bubble will hold on. As usual, certain things are "allowed" to operate on loss but at some time investors want profit too.

Yeah, that's the kind of thing I was getting at. Agree fully.
It's a shame for us gamers - but, well we can blame A.I. No doubt a lot of profit there!

 

[Rapob]

Frankly speaking I've never figured 3060Ti as a "gem" - maybe it's just retrospective thoughts inspired by GT4050 aka "RTX4060"? OFC that thing would never rival midtier GA104 with given AD107, even with help of miraculous TSMC 5nm.
I thought 3080 was more of a "gift", when nvidia tear flagship GA102 from its heart for a "mere" 700$.

3080 is nothing like a 3090 and Samsung 8nm (10nm in reality, just renamed and on par with TSMC 12nm) was very cheap and cut down due to poor yields at Samsung. 3000 series was priced low mostly due to using Samsung - Nvidia came crawling back to TSMC with 4000/5000 series and they perform massively better when looking at perf/watt, which is the most important metric for a GPU.

3090 = ~21% more CORES and bigger bus too, 384 bit.
3090 Ti had like 25% more cores.

10GB vs 24GB, massive difference however 10GB was/is plenty for most gamers still, especially when you don't max out new AAA games completely at 1440p/4K, which no-one probably does on a 6+ year old GPU but at least it has 100% support for DLSS 4, unlike AMD cards from that time.

Good upscaling (with broad support) beats VRAM in terms of longevity. Any day of the week. You can always get around lack of VRAM. You can't get around slow GPU performance and bad upscaling.

I will take 10-12GB VRAM with good upscaling any day over 16GB VRAM with crap upscaling, especially if the latter has mediocre RT perf and lack of features as well.

Cards like Radeon 6800/6900 series aged like milk even with 16GB. Arch was poor, RT slammed the cards and RT elements are forced in many games and have for years. No AI cores to bruteforce FSR 4, hence why Radeon 6000 first get FSR 4 support sometime in 2027, they are probably to slow to run it well, bruteforcing FSR 4 with INT8 and perf takes a huge hit, GPU will buckle.

3080 10GB vs Radeon 6800 XT 16GB - Which aged better? 3080 did. Upscaling really was gold, and Radeon 6000 still don't have FSR 4 but will "maybe" get it in 2027 sometime, no FP8 means INT8 bruteforcing so it will probably not work good. 3080 have Tensor cores which does the work. Nvidia futureproofed RTX, hence why DLSS 4 works on all RTX cards, native FP8 or not, yet RTX 4000/5000 does it better than 2000/3000 due to native FP8.

Lacking FP8 and AI/MATRIX CORES is AMDs big problem with FSR 4 on older cards. Can't fix what is not present. Nvidia had Tensor cores since first generation of RTX, 2000 series from 2018. Hence DLSS 4 support is no problem at all.

Why yap about old crap tho. People should be using RTX 5000/4000 or Radeon 9000/7000. Everything else is meh and dated. If I had to choose anything older, RTX 3000 is the only cards I would touch with a ten feet pole. RDNA 2 is dead in the water and everything before RDNA 2 is literally trash bin worthy.

If you asked some random dude to pick a 3080 or 6800XT, or 2070/2080 vs 5700XT, absolutely no-one would pick the AMD cards. That is how shitty they aged and with no FSR 4 support, dead in the water today.

AMD should literally just focus on Radeon 9000/7000 and the future. No need to waste ressources on getting FSR 4 to Radeon 6000 series, too slow and old to use it anyway.

People ramble about VRAM like the most important metric for longevity. It's not and it won't matter at all if GPU arch is pure crap with no support for good upscaling.

RTX 2000/3000 vs Radeon 5000/6000 is the proof. 8-10 years old now, Nvidia aged like wine compared to AMD, due to DLSS 4 support on day one and the fact that AMD GPUs back when buckled when just a tiny bit of RT elements were present. GPU was pure crap and only able to do simple rasterization. This was partly fixed with Radeon 7000 and completely fixed with 9000 series. Future looks bright for AMD GPUs, if they just forget about old crap archs and look forward instead of backward. No-one freaking cares about Radeon 5000/6000 that old trash is dumpster worthy tech.

Can't wait to see RDNA 5 / UDNA, if AMD takes this serious, they could shine, now that upscaling is fixed with FSR 4.x - I sadly think they won't try much and insist on staying in the "low to mid-end market" like Radeon 9000 is aimed at. Because they are CPU first and makes more money here. Making cheap gaming GPUs is not their goal. Eats away at their TSMC output for little margins.

AMDs priority list:

1. Enterprise CPUs
2. HEDT and desktop CPUs
3. Enterprise GPUs, AI etc.
4. APUs, and custom APUs (OEM PCs, Consoles etc)
5. Lunch break
6. Gaming GPUs

If in doubt, just read their financial reports.

 

[HardReset]

Again: this isn't even remotely correct. In theory, any process node can be used for any die size. In practice, viability depends on yields, at which older, more mature nodes always perform better. Period.

In the last decade, "Lazy" NVidia has improved GPU performance more than anyone on the planet. Facts matter.

Again, no. There is hard limit how big single die can be. For example: https://www.eetimes.com/understanding-the-big-spend-on-advanced-packaging-facilities/

The third reason is perhaps less obvious. The maximum exposure field size of a contemporary EUV scanner is 26 mm (slit) by 33 mm (scan), or 858 mm² (which is sometimes called maximum reticle size). Next-generation High-NA EUV (0.55 numerical aperture) scanner will retain a 26mm slit, but will halve the scan to 16.5mm, so the exposure field size will be 429 mm2.

So yes, there are practical hard limits for die size that come directly from manufacturing equipment used. Perhaps not on theory but on practice yes.

Once again you are comparing Nvidia to others rather than looking how much less Nvidia has achieved lately. Now, with RTX 4090 Nvidia used TSMC 5nm tech, with RTX 5090 they used custom TSMC 5nm tech. Now, there is small difference between those two manufacturing techs so if Nvidia wants improvement, they need better architecture because this time manufacturing tech makes small difference. From article:

RTX 5090 is 30% faster than RTX 4090 but also consumes 28% more power too (450W vs 575W). Basically performance per watt stays same with somewhat same process node. In other words, Nvidia just recycled 4090 for new process node and that's it.

 

[HardReset]

Nintendo has shown that using ARM CPU is very much possible. x86 might face a problem in this market going forward. ARM eats into server market too, little by little.

Well, PS3, Xbox 360 and Wii showed that PowerPC is possible. Apple showed that switching from 68K to PowerPC to Intel to ARM is possible.

Closed system that does not need to maintain backwards compatibility can use whatever CPU ISA available. There are reasons why Sony and MS both decided to go with x86, biggest one is undoubtedly backwards compatibility.

 

[Endymio]

Again, no. There is hard limit how big single die can be. For example: https://www.eetimes.com/understanding-the-big-spend-on-advanced-packaging-facilities/ ... So yes, there are practical hard limits for die size that come directly from manufacturing equipment used. Perhaps not on theory but on practice yes.

Oops! You should read your own links. You're speaking of max reticle size, which doesn't directly correlate to the declining feature size of advancing nodes. From 130nm down to 2nm EUV, max reticle size remained unchanged. But, as the article says:

The maximum exposure field size of a contemporary EUV scanner is 26 mm (slit) by 33 mm (scan), or 858 mm² (which is sometimes called maximum reticle size). Next-generation High-NA EUV (0.55 numerical aperture) scanner will retain a 26mm slit, but will halve the scan to 16.5mm, so the exposure field size will be 429 mm2.

So your original claim is diametrically reversed from the truth: the more-advanced high-NA EUV node reduces maximum die size. But of course, that hasn't even happened yet ... over the period of time in question, maximum die size hasn't changed.

Once again you are comparing Nvidia to others rather than looking how much less Nvidia has achieved lately.

Oops again. As my first post on the subject stated, since the introduction of DLSS, NVidia has increased maximum consumer card performance by more than 300% Compared to Nvidia.

 

[Rudyduu]

all we got was 9070 XT that don't even beat 5070 Ti. Another lost oppotunity.

Better check current benchmarks and look at the massive differences in price. The 9070XT doesn't beat the 5070 ti in every title, but for $300 less it doesn't need to, since it's priced at a tier lower.

 

[HardReset]

Oops! You should read your own links. You're speaking of max reticle size, which doesn't directly correlate to the declining feature size of advancing nodes. From 130nm down to 2nm EUV, max reticle size remained unchanged. But, as the article says:

The maximum exposure field size of a contemporary EUV scanner is 26 mm (slit) by 33 mm (scan), or 858 mm² (which is sometimes called maximum reticle size). Next-generation High-NA EUV (0.55 numerical aperture) scanner will retain a 26mm slit, but will halve the scan to 16.5mm, so the exposure field size will be 429 mm2.

So your original claim is diametrically reversed from the truth: the more-advanced high-NA EUV node reduces maximum die size. But of course, that hasn't even happened yet ... over the period of time in question, maximum die size hasn't changed.

Again, maximum die size is not constant, it depends on tools used. Perhaps max size will get smaller as it also means tools may be much simpler. And why make process that could do big dies as AMD only make max medium size dies, Intel has also moved on multiple smaller dies, Nvidia too, phones have never made huge dies etc. There is not huge demand for big dies so what follows next is not hard to guess.

My claim is more than valid. Care to explain why Nvidia paid tons of money for TSMC just to make custom process based on N4P? Easy answer:

RTX 4090 die size is 609mm2 and RTX 5090 die size is whopping 750mm2!

And since that process is based on TSMC N4P that Should be faster and more efficient version of original 5nm (Nvidia custom seems not to be), there is only one explanation that makes any sense: Nvidia wanted process that allows them to make bigger die.

Oops again. As my first post on the subject stated, since the introduction of DLSS, NVidia has increased maximum consumer card performance by more than 300% Compared to Nvidia.

And? RTX 2080 Ti had around 19M transistors. RTX 5090 has whopping 92M transistors.

So Nvidia "only" got 3X more performance while transistor count was upped 4.8X ""

 

[Endymio]

Again, maximum die size is not constant, it depends on tools used.

The die size enabled by the "tools used" didn't change from 130mm to 2nm. It's about to change -- but when it does, maximum die size will decline. Your claim that "more advanced process nodes allow larger die sizes" was wrong. Own it and move on.

So Nvidia "only" got 3X more performance while transistor count was upped 4.8X ""

You're becoming increasingly detached from reality. NVidia upping transistor counts by 4.8X -- I.e. a far more complex design -- is the exact opposite of proving they are "lazy" ... especially when over the same period AMD upped transistor counts 4.3X, and still couldn't take the performance crown.

As you yourself admit: AMD has surrendered the fight entirely, and is no longer even attempting to make high-end cards. It's pretty clear here who, if any, of these firms is "lazy".

 

[HardReset]

The die size enabled by the "tools used" didn't change from 130mm to 2nm. It's about to change -- but when it does, maximum die size will decline. Your statement that "more advanced process nodes allow larger die sizes" was wrong. Own it and move on, please.

TSMC 28nm tech was used for looong time. Despite that biggest die actually made was around 600mm2 that was released 4 years after process started shipping on volume, so process was Very mature.

However there are over 800mm2 die size products actually made on both TSMC 7nm and 5nm.

Care to explain that?

You're becoming increasingly detached from reality. NVidia upping transistor counts by 4.8X -- I.e. a far more complex design -- is the exact opposite of proving they are "lazy" ... especially when over the same period AMD upped transistor counts 4.3X, and still couldn't take the performance crown.

And -- as you yourself admitted -- AMD has surrendered the fight entirely, and is no longer even attempting to make high-end cards. It's pretty clear here who, if any, of these firms is "lazy".

AMD upped transistor counts against what? AMD starting point was mid range chip. Nvidia starting point was high end chip. Invalid comparison.

AMD being lazy or not does not make Nvidia less lazy.

 

[Endymio]

TSMC 28nm tech was used for looong time. Despite that biggest die actually made was around 600mm2 that was released 4 years after process started shipping on volume, so process was Very mature.

However there are over 800mm2 die size products actually made on both TSMC 7nm and 5nm.

Care to explain that?

You're doubling down on this silliness? Again: the reticle limit has exactly zero to do with the defining characteristic of a process node: the feature size. 800mm is right at the reticle limit : we could have made dies that large back then. Had there been a need to do so, we could have easily expanded that limit to a larger size, or -- using maskless lithography -- we could make any size die we want, up to the size of an entire siicon wafer: 70,000mm, for a 300mm wafer.

Your statement was wrong. Drop it and move on.

AMD being lazy or not does not make Nvidia less lazy.

So in your opinion, every GPU maker on the planet is lazy? If you truly believe this, why not start a chip company, and make your own graphics cards?

 

[HardReset]

You're doubling down on this silliness? Again: the reticle limit has exactly zero to do with the defining characteristic of a process node: the feature size. 800mm is right at the reticle limit : we could have made dies that large back then. Had there been a need to do so, we could have easily expanded that limit to a larger size, or -- using maskless lithography -- we could make any size die we want, up to the size of an entire siicon wafer: 70,000mm, for a 300mm wafer.

Your statement was wrong. Drop it and move on.

My original claim was:

Better manufacturing processes generally enable bigger die sizes too and/or more transistor count.

TSMC 40nm: Biggest die made around 530mm2
TSMC 28nm: Biggest die made around 600 mm2
TSMC 20nm: It was trash
TSMC 16nm: Biggest die made over 800 mm2
TSMC 7nm: Biggest die made over 800 mm2
TSMC 5nm: Biggest die made over 800 mm2

See? Better processes DO allow bigger die sizes vs old onees. Not that I said sizes will get bigger and bigger and bigger but 28nm is still being used and Nvidia used it too with GTX980Ti. Modern processes allow much bigger dies.

So in your opinion, every GPU maker on the planet is lazy? If you truly believe this, why not start a chip company, and make your own graphics cards?

If market leader is lazy and no-one actually cares, then there is probably not much money to be made..

 

[Rapob]

Well, PS3, Xbox 360 and Wii showed that PowerPC is possible. Apple showed that switching from 68K to PowerPC to Intel to ARM is possible.

Closed system that does not need to maintain backwards compatibility can use whatever CPU ISA available. There are reasons why Sony and MS both decided to go with x86, biggest one is undoubtedly backwards compatibility.

Biggest one was price, those cheap AMD APUs made perfect sense at the time.

CPU was a major problem with PS4/XB1, those garbage jaguar cores is the reason most games ran 30 fps (with dips), even when PS4/XB1 launched, jaguar was slow and dated, used in ultra low-end stuff.

Custom APU = Mostly cheap crap, just look at Steam Machine if in doubt. Claims custom but pure garbage compared to off the shelf APUs. Only reason its "custom" is because AMD did not have anything this low-end and they had to make it, to hit budget.

ARM will easily run most x86 games, you will see very soon when Nvidia N1X goes live. Windows 11 has massive ARM support and most x86 games will run no problem (Steam game catatog with an ARM CPU on Windows is going to be much bigger than the Steam game catalog you get in Linux using x86 - Hence Steam Deck verification stamp on games)

AMD needs to speed up their Zen 5 + RDNA 4 APUs, or they will face trouble real soon.

Sadly it looks like AMD cares very little about PC gamers, they never miss an opportunity to miss an opportunity in the PC gaming GPU market.

Leaving high-end GPU market before RDNA 4 launches. Cancelling the top RDNA 4 SKUs, etc. All AMD does is undercut Nvidia by 50 dollars and call it a day. They likely don't care much, as CPUs is their prime field. CPUs is the reason AMD woke up from the dead.

 

[Rapob]

Better check current benchmarks and look at the massive differences in price. The 9070XT doesn't beat the 5070 ti in every title, but for $300 less it doesn't need to, since it's priced at a tier lower.

9070 XT is not 300 dollars less. What shrooms did you eat this morning?

Sapphire Radeon RX 9070 GRE Pulse Review

The Sapphire Radeon RX 9070 GRE Pulse is the quietest card we've reviewed today. Even at full load it runs whisper quiet, yet still packs enough punch for a great 1440p gaming experience. Priced at $550, the GRE is shaping up to be a decent alternative to NVIDIA's GeForce RTX 5070.

www.techpowerup.com

Their new 9070 GRE is slower than 5070 while costing the same. Oh yeah DAT SOME GOOD VALUE. In terms of features and game support, Nvidia wins with ease, as usual.

Also, there’s much more than just rasterization performance to consider when buying a GPU today.

Features, like upscaling, frame gen, downsampling, HDR support (RTX HDR), Reflex, Shadowplay - just to name a few - plus RT and Path Tracing advantage on the Nvidia side - AMD don't even come close in terms of features and game support. This is the main reason why AMD GPUs are cheaper.

Absolutely no-one will pay Nvidia prices for an AMD GPU. Not even AMD GPU fanbois.
AMD is forced to undercut Nvidia to even sell. That is reality for you and the reason why AMD officially left the high-end GPU market with Radeon 9000 series. Expensive AMD GPUs just don't sell.

Like 90% of AMD GPU sales, is in the sub 500 dollar segment. AMDs best selling GPUs of all time, is dirt cheap GPUs like RX 580, 570, 480, 470 - 200-300 dollar SKUs. 5700XT sold well too, priced at 300-400 dollars. Launched at 449, cut down 399 almost instantly post launch.

If they can keep improving on FSR 4.x and actually hits the nail with RDNA 5/UDNA, future look better but I have my doubts, as AMD probably don't care much for gaming GPUs. Simply eats from their TSMC output, which is limited. They earn much more on CPUs, enterprise GPUs and APUs.

AMD is CPU/APU first. When it comes to GPU market, they aim for enterprise GPU market. Gaming GPUs is not a big focus. Little money here. Hence why Nvidia rocks a 85-90% dGPU marketshare (without even trying) and AMD is struggling to even hold 10%

Nvidia don't care much for gaming GPUs either. Obviously they can make more money in the enterprise space. Does not change the fact that Nvidia dominates the gaming GPU market anyway.

9070 XT is a fine mid-end card but no-one would buy it if 5070 Ti was the same price. 5070 Ti sold far better, while costing more. People simply want the RTX features and don't care to pay extra. Nvidia GPUs hold their value far better than AMD GPUs anyway, much bigger demand in the used market and Nvidia don't lower prices constantly like AMD.

 

[Rudyduu]

Oh for ****s sake. Is tech spot really so desperate for revenue that they have taken to doomposting?

Nvidia has not stopped making GeForce cards
Nvidia has not stopped development of future generations
Nvidia has not stopped driver development.
Nvidia has not abandoned anything.

Steve, your article is bad and you should feel bad.

Feels like you read the title and nothing else. Read the whole article. Look at the data. Look at the hardware, the performance gains from the previous generation and look at their prices. They are putting so much less R&D into gaming and so much more into AI. They clearly don't care that much about gaming anymore.