NVIDIA 4000 Series

[keyser van someone]

Outside of game pass, you can get the same games cheaper on a series x as well

 

[Joxeon]

Ultimately I think the real issue is all down to Nvidia and their pricing and product stack vs. what we consider to be entry/mid/high-end. 70 series was never mid tier, it was always high tier - and it absolutely shouldn't be considered mid-tier at 589 quid

The 4090 is twice as fast as a 4070 though and 70% faster than a 4070ti so hard to consider those cards high end.

 

[Grim5]

Emulation makers tell people to stay away from the 4060 and 4060ti, get the 3060ti instead or face a serious downgrade in emulation performance.

The memory system on the 4060 series cards is severely hurting emulation performance

Nintendo Switch emulation team at YUZU calls NVIDIA's GeForce RTX 4060 Ti a 'serious downgrade'

With its downgraded 128-bit memory bus, the Nintendo Switch emulation team at YUZU has shared its thoughts on the new GeForce RTX 4060 Ti.

www.tweaktown.com

 

[GhostDog1981]

The 4090 is twice as fast as a 4070 though and 70% faster than a 4070ti so hard to consider those cards high end.

Tier A is the 4090, you're calling the 4080 and XTX and also the XT midrange when they are in Tier B in GPU Hierarchy thread- not sure why you feel the need to re-write all the gpu names to help you make sense of the current market.

 

[GhostDog1981]

Emulation makers tell people to stay away from the 4060 and 4060ti, get the 3060ti instead or face a serious downgrade in emulation performance.

The memory system on the 4060 series cards is severely hurting emulation performance

Nintendo Switch emulation team at YUZU calls NVIDIA's GeForce RTX 4060 Ti a 'serious downgrade'

With its downgraded 128-bit memory bus, the Nintendo Switch emulation team at YUZU has shared its thoughts on the new GeForce RTX 4060 Ti.

www.tweaktown.com

I think you're from the States?- I see a 7900xt for $779 is that including taxes? If so why does the Pound and Euro match the US price but really we pay a lot more?

 

[TNA]

I think you're from the States?- I see a 7900xt for $779 is that including taxes? If so why does the Pound and Euro match the US price but really we pay a lot more?

He is from Australia.

 

[david25]

Nvidia RTX 4060 benchmark shows it outperforming the RTX 3060 by up to 17%​

Nvidia RTX 4060 benchmark shows it outperforming the RTX 3060 by up to 17% - KitGuru

While originally planned to launch in July, the RTX 4060 will now be arriving before the end of June

www.kitguru.net

And selected comparison here

Nvidia Shares GeForce RTX 4060 Performance Numbers

Nvidia tests its $299 graphics card in 18 games.

www.tomshardware.com

 

[throwaway4372]

Comparing fake frames with no fake frames. Dumb.

 

[TNA]

Comparing fake frames with no fake frames. Dumb.

Is it only beating it because of DLSS 3? If so that is sad. What a crappy gen Ada has been. Only GPU that is worth talking a out is the 4090 and they want silly moneys for it.

 

[Joxeon]

Tier A is the 4090, you're calling the 4080 and XTX and also the XT midrange when they are in Tier B in GPU Hierarchy thread- not sure why you feel the need to re-write all the gpu names to help you make sense of the current market.

Because B tier is generation behind A tier in performance.

 

[Bencher]

You're wasting your time. He's deluded. It's been covered loads of times that you can't really build a console destroying game for under £650

Just look at almost every digital foundry game analysis lately and you will see PC games are plagued with issues and stutter that cancels out a lot of their advantage anyway.

Can you build a console that destroys a 500 pounds PC? Nope. Silly comparison.

 

[Murphy]

Nvidia RTX 4060 benchmark shows it outperforming the RTX 3060 by up to 17%​

Nvidia RTX 4060 benchmark shows it outperforming the RTX 3060 by up to 17% - KitGuru

While originally planned to launch in July, the RTX 4060 will now be arriving before the end of June

www.kitguru.net

Outperforming in compute workloads, AFAIK that's pretty different than gaming.

 

[GhostDog1981]

Because B tier is generation behind A tier in performance.

4090 is a Titan class card so not one for the enthusiast gamer.

Tier B is enthusiast category.

You’ve got it all wrong mate.

You won’t get the magic price to performance you are expecting. Keep that 3080 going for another 5 years maybe? (Or get a 5060ti? cost potentially £600)

 

[Joxeon]

4090 is a Titan class card so not one for the enthusiast gamer.

Tier B is enthusiast category.

You’ve got it all wrong mate.

You won’t get the magic price to performance you are expecting. Keep that 3080 going for another 5 years maybe? (Or get a 5060ti? cost potentially £600)

Its not a Titan class, the die is more cut down than any of the 80ti cards. A titan uses the full die.

Notice where the 4080 sits as well in with the other 70 class cards while the 4070/ti sits in amongst the 60 classes.

 

[Bencher]

Its not a Titan class, the die is more cut down than any of the 80ti cards. A titan uses the full die.

Notice where the 4080 sits as well in with the other 70 class cards while the 4070/ti sits in amongst the 60 classes.

That's because the Ada die is much much much bigger than anything before it, in terms of transistor count. Giving you 50% of 10k euros is not the same as give you 50% of a million, right?

 

[t2na]

Its not a Titan class, the die is more cut down than any of the 80ti cards. A titan uses the full die.

Notice where the 4080 sits as well in with the other 70 class cards while the 4070/ti sits in amongst the 60 classes.

Christ, that chart really puts things into context with how much Nvidia are trying to screw everyone this generation.

 

[GhostDog1981]

Its not a Titan class, the die is more cut down than any of the 80ti cards. A titan uses the full die.

Notice where the 4080 sits as well in with the other 70 class cards while the 4070/ti sits in amongst the 60 classes.

No matter what patterns this OP has found it doesn’t change the fact Nvidia is happy to call the 4080 an 80 class card and they’re not going to re-write that for you or any gamer.

The 4070ti- 4080 12gb was a step too far and they realised they were being silly.

Still think they’ll carry on offering smaller dies and cut down bus width for 80 class and 70 class cards next gen.

This gen might be the best of a bad next couple of gens.

Moaning about die sizes might get awfully tiring next gen and the gen after lol

 

[Murphy]

That's because the Ada die is much much much bigger than anything before it, in terms of transistor count. Giving you 50% of 10k euros is not the same as give you 50% of a million, right?

Transistor count is meaningless in terms of size, cost, performance and pretty much everything other than 'look how good our fabrication process is'.

 

[Armageus]

Its not a Titan class, the die is more cut down than any of the 80ti cards. A titan uses the full die.

The yields on the full die part are terrible - you only have to look at the L40 Datacenter and RTX 6000 ADA Workstation parts to see this (and these are intended for the current cutting edge AI market, that wants every bit of power that they can get) - Even those aren't using the "full" die (although they are a lot nearer).

Look at the price premium for such parts

Table created to visualise the BS

	Model	Die	CUDA Cores	Memory Bus (bits)	Bandwidth (GB/s)	VRAM (GB)	TDP (W)	Price
Mobile	4050 Laptop	AD107	2560	96		6	35-115
Mobile	4060 Laptop	AD107	3072	128		8	35-115
Mobile Workstation	RTX 2000 ADA	AD107	3072	128		8	35-140
Desktop	4060	AD107	3072	128	272	8	115	$299
Desktop	4060Ti	AD106	4352	128	288	8/16	160	$399/$499
Mobile	4070 Laptop	AD106	4608	128		8	35-115
Mobile Workstation	RTX 3000 ADA	AD106	4608	128	256	8	35-140
Theoretical	AD106 Max	AD106	4608	128
Fake/Leak	4060 Super		5120	192	432	12	200	$399
Mobile Workstation	RTX 3500 ADA	AD104	5120	192	432	12	60-140
Desktop	4070	AD104	5888	192	504.2	12	200	$599
Workstation	RTX 4000 SFF ADA	AD104	6144	160	320	20	70
Mobile	4080 Laptop	AD104	7424	192		12	80-150
Mobile Workstation	RTX 4000 ADA	AD104	7424	192	432	12	80-175
Desktop	4070 Ti	AD104	7680	192	504.2	12	285	$799
Datacenter	L4	AD104	7680	192	504.2	24	285
Theoretical	AD104 Max	AD104	7680	192
Fake/Leak	4070 Super		8192	256	672	16	280	$699
Mobile	4090 Laptop	AD103	9728	256		16	80-150
Mobile Workstation	RTX 5000 ADA	AD103	9728	256		16	80-175
Desktop	4080	AD103	9728	256	761.8	16	320	$1199
Theoretical	AD103 Max	AD103	10240	256
Fake/Leak	4080 Super		13312	320	840	20	350	$1199
Desktop	4090	AD102	16384	384	1008	24	450	$1599
Datacenter	L40	AD102	18176	384	864	24/48	300	$8000
Workstation	RTX 6000 ADA	AD102	18176	384	960	48	300	$6800
Theoretical	AD102 Max	AD102	18432​	384
Fake/Leak	4090 Super		18432​	384	1152	24	450	$1999

 

[GhostDog1981]

The yields on the full die part are terrible - you only have to look at the L40 Datacenter and RTX 6000 ADA Workstation parts to see this (and these are intended for the current cutting edge AI market, that wants every bit of power that they can get) - Even those aren't using the "full" die (although they are a lot nearer).

Look at the price premium for such parts

So it costs more to make a true old fashioned 80 class gpu- hence it’s not economically achievable for Nvidia which explains why comparing to past gen die sizes and scrutinising value for money might be inaccurate?

What’s AMDs excuse then?