Discussion Ada/'Lovelace'? Next gen Nvidia gaming architecture speculation

[TESKATLIPOKA]

Not enough for handful or 4K, or RT games, which 4060 really isn't powerful enough for that in the first place.

RTX 4060 will be powerful enough for Resident Evil 4 even with RT on in 1440p, but sadly only 8GB Vram will make It unplayable If not at 1080p then certainly at 1440p.

If someone doesn't mind just a bit more than 30FPS, then It would be "powerful enough" even for 4K + RT in this game. Link

How many of the new games will end up the same in the next 2-3 years? And we should pay $350 for this? But to be truthful, RTX 4060Ti would be even worse off.

P.S. Yes, In Hogwarts Legacy this level of performance is not enough even for 1080p + RT, but for that we have DLSS.

 

[Heartbreaker]

How many of the new games will end up the same in the next 2-3 years? And we should pay $350 for this? But to be truthful, RTX 4060Ti would be even worse off.

P.S. Yes, In Hogwarts Legacy this level of performance is not enough even for 1080p + RT, but for that we have DLSS.

Again as I said, 4K or RT, plus RE4 is a buggy release, only just out.

Play at 1440p without RT and it's fine.

Plus RT doesn't have any significant value add in this game, it's hard to spot any differences, and when you do it more often than not, looks better with RT off.

So Buggy release, where you just shut off a useless feature, and it's fine, is your example?

 

[Ranulf]

Great, so more money for more useless features, 3rd gen in of RT tech. If it was really useless or Nvidia didn't care about marketing more RTX, they would name them the GTX 4050/60 etc.. That or just go with the GTX 1750/60/ti/S/ti-200/MX/SE/MX-440.

 

[TESKATLIPOKA]

Again as I said, 4K or RT, plus RE4 is a buggy release, only just out.

Play at 1440p without RT and it's fine.

Plus RT doesn't have any significant value add in this game, it's hard to spot any differences, and when you do it more often than not, looks better with RT off.

So Buggy release, where you just shut off a useless feature, and it's fine, is your example?

Ok. This game wasn't a good example, because of using RT only for reflections. RTX 3070 crashing also didn't help in determining how much impact not enough Vram has on performance.
I will wait for RTX 4060 release and compare It to Its predecessor, but less Vram still leaves a bad impression. Reasonable price could help a lot, but $349 looks pretty likely, and that's a lot.

 

[Heartbreaker]

I will wait for RTX 4060 release and compare It to Its predecessor, but less Vram still leaves a bad impression. Reasonable price could help a lot, but $349 looks pretty likely, and that's a lot.

I think you have high expectations of a $350 card in this era. The 3060 12GB was also an anomaly when the 3070 only had 8GB.

 

[Tup3x]

Not enough for handful or 4K, or RT games, which 4060 really isn't powerful enough for that in the first place.

It's not enough even in Uncharted 4 at 1440p.

 

[Heartbreaker]

It's not enough even in Uncharted 4 at 1440p.

According to DF it's totally fine even with Ultra Textures at 1440p. Ultra is only a problem at 4K, though you could use high at 4K which looks basically identical:

 

[Tup3x]

According to DF it's totally fine even with Ultra Textures at 1440p. Ultra is only a problem at 4K, though you could use high at 4K which looks basically identical:

It's not fine. If you play only one chapter then it's almost certainly fine but once you load next chapter chances are you hit the VRAM limit and performance tanks. Restarting the game will fix it. Situation is worse in The Lost Legacy. One could say that maybe it's a poor port with memory leak of some kind but in any case, it's on the edge and 8GB will cause issues at 1440p.

 

[maddie]

I think you have high expectations of a $350 card in this era. The 3060 12GB was also an anomaly when the 3070 only had 8GB.

Not the reverse?

 

[Heartbreaker]

Not the reverse?

Considering that it went: 8GB (50), 12GB(60), 8GB(60 Ti), 8GB(70), 8GB(70 Ti), 10GB(80), 12GB(80 Ti), 24GB(90).

The 60 stood out as the anomaly. It's clear they consider 8GB adequate, but the 60 series they had to choose between 6GB and 12GB because of bus size.

 

[TESKATLIPOKA]

Videocardz​

RTX 4070 supposedly costs only $599. It supposedly has 46SM(5888 Cuda) and 12GB 192bit, 36MB L2, 200W TBP.

I think RTX4060 can't cost $349 If this is true, more like $299.

 

[Heartbreaker]

I think RTX4060 can't cost $349 If this is true, more like $299.

There you go setting yourself up for disappointment again.

3070 also had 5888 Cuda cores, but cost only $499, and yet still charged $329 for the 3060.

Given the significant price inflation at every tier of the 4000 series lineup, I think it's extremely unlikely that the 4060 will cost less than the 3060.

 

[TESKATLIPOKA]

There you go setting yourself up for disappointment again.

3070 also had 5888 Cuda cores, but cost only $499, and yet still charged $329 for the 3060.

Given the significant price inflation at every tier of the 4000 series lineup, I think it's extremely unlikely that the 4060 will cost less than the 3060.

RTX 3070 has 52% more SM(Cuda) and 52% more performance at 4K(TPU) while costing 52% more.

RTX 4070 has 92% more SM(Cuda) and let's say 85-92% more performance at 4K while costing 85-92% more, so that's $312-323.

This would point out to $329, which is less than $349 but more than $299.

Why I think $299 is still possible?
RTX 3060(276mm2) -> RTX 3070(392.5mm2) ended with 4GB less Vram and 42% bigger die.
RTX 4060(~160mm2) -> RTX 4070(294.5mm2) ended with 4GB more Vram and 84% bigger die.
Production cost relative difference is bigger in this gen than the previous one.

 

[Heartbreaker]

RTX 3070 has 52% more SM(Cuda) and 52% more performance at 4K(TPU) while costing 52% more.

RTX 4070 has 92% more SM(Cuda) and let's say 85-92% more performance at 4K while costing 85-92% more, so that's $312-323.

This would point out to $329, which is less than $349 but more than $299.

Why I think $299 is still possible?
RTX 3060(276mm2) -> RTX 3070(392.5mm2) ended with 4GB less Vram and 42% bigger die.
RTX 4060(~160mm2) -> RTX 4070(294.5mm2) ended with 4GB more Vram and 84% bigger die.
Production cost relative difference is bigger in this gen than the previous one.

All the theory crafting and wishful thinking in the world won't get around the fact that NVidia can and almost certainly will charge more than that.

Most RTX 3050 are still selling for over $300. You have to be dreaming in technicolor to think NVidia is going to price the 4060 at the going rate of the old 50 series cards from the previous generation. Especially when every other 4000 segment increased in price.

They still need room for new overpriced 4050 to slot in under the 4060, after all.

 

[SteveGrabowski]

Again as I said, 4K or RT, plus RE4 is a buggy release, only just out.

Play at 1440p without RT and it's fine.

Plus RT doesn't have any significant value add in this game, it's hard to spot any differences, and when you do it more often than not, looks better with RT off.

So Buggy release, where you just shut off a useless feature, and it's fine, is your example?

If RT is a useless feature what's the point of buying Nvidia?

 

[Heartbreaker]

If RT is a useless feature what's the point of buying Nvidia?

It's a useless feature in that game. Not every game. In some games it actually improves the visuals.

 

[In2Photos]

If RT is a useless feature what's the point of buying Nvidia?

Because NVidia tells us to buy them, of course!

 

[Ajay]

Because NVidia tells us to buy them, of course!

Well, while being an excellent engineering company, Nvidia are great at marketing. They do put the money where there mouths are - spending allot of $$s on game co-development. The original rtx 2000 series ray tracing performance was pitiful, but those three letters 'rtx' are really helping to sell allot of cards today. Nvidia changed consumer’s focus. I don’t think AMD can catch up in the short term because their efforts are focused elsewhere and logically so. They’ll need to spend their way into a significant market share, if and when they want to.

 

[A///]

AMD needs to deliver something better than a lukewarm day old grainy baked potato to be relevant in the gaming space as much as nvidia is.

 

[TESKATLIPOKA]

AMD needs to deliver something better than a lukewarm day old grainy baked potato to be relevant in the gaming space as much as nvidia is.

AMD is pretty much out this generation. They Redacted up pretty badly with RDNA3.
So what we are left is only overpriced Nvidia and let's hope RDNA4 will do better.

AD106 and AD107 don't look very good, only 8GB Vram and high price, at best the same perf/$ ratio as the previous gen. RTX 4070 for $599 doesn't look that bad in comparison, ~80-85% of RTX 4070 Ti performance + 12GB Vram.

Profanity is not allowed in the tech dorums.
admin allisolm

 

[A///]

AMD is pretty much out this generation. They Redacted up pretty badly with RDNA3.
So what we are left is only overpriced Nvidia and let's hope RDNA4 will do better.

AD106 and AD107 don't look very good, only 8GB Vram and high price, at best the same perf/$ ratio as the previous gen. RTX 4070 for $599 doesn't look that bad in comparison, ~80-85% of RTX 4070 Ti performance + 12GB Vram.

They did a lot of work but you can get as close a performance if you go with a 6900xt or 6950xt. amd gpu divisions fine wine turns to be fine vinegar.

Please do not quote profanity.
admin allisolm

 

[Aapje]

AMD is pretty much out this generation. They Redacted up pretty badly with RDNA3.
So what we are left is only overpriced Nvidia and let's hope RDNA4 will do better.

AD106 and AD107 don't look very good, only 8GB Vram and high price, at best the same perf/$ ratio as the previous gen. RTX 4070 for $599 doesn't look that bad in comparison, ~80-85% of RTX 4070 Ti performance + 12GB Vram.

I actually think that Nvidia also messed up badly, by underestimating how fast VRAM demands would grow and by misjudging the willingness of gamers (rather than miners) to spend. I foresee dramatic sales across the board in 2023.

I'd strongly advise people to wait until the next generation if they can. If you can't, then RDNA2 is actually the most attractive for pure gamers. You either have more VRAM or cheaper prices, so either the card should last longer or it should be less costly if you have to upgrade again relatively soon.

Please do not quote profanity.
admin allisolm

 

[TESKATLIPOKA]

I actually think that Nvidia also messed up badly, by underestimating how fast VRAM demands would grow and by misjudging the willingness of gamers (rather than miners) to spend. I foresee dramatic sales across the board in 2023.

I think limited Vram is mainly due to laptops, where space is limited and where OEMs want to keep BOM as low as possible, and It's likely a bigger market than desktop for Nvidia.

On the other hand, they could offer desktop versions with 2x more memory in clamshell.
RTX4050 -> 2*6=12GB
RTX 4060(Ti) -> 2*8=16GB
RTX 4070(Ti) -> 2*12=24GB
RTX 4080 -> 2*16=32GB
RTX 4090 -> 2*24=48GB
But this amount of Vram would be an overkill for Ad104 and higher. The best thing would be If 24Gbit(3GB) memory chips were released.

 

[Heartbreaker]

I think limited Vram is mainly due to laptops, where space is limited and where OEMs want to keep BOM as low as possible, and It's likely a bigger market than desktop for Nvidia.

On the other hand, they could offer desktop versions with 2x more memory in clamshell.

Certainly the limit on desktop is simply NVidia not allowing it. It would be nice if they simply allowed OEMs to build 16GB-24GB cards for 4060-4070Ti as an option.

 

[jpiniero]

It's not so simple. You're effectively doubling the memory cost and memory power consumption. Not to mention any of the additional costs of doing the clamshell on top of that.