NVIDIA GeForce RTX 5080 Founders Edition

[Bugsy004]

Nvidia's RTX 5080 is Actually an RTX 5070

Take a hard look at the specs, and you'll see the RTX 5080 is just an RTX 5070 in disguise. Nvidia's greed may be at play, and...

www.techspot.com

Nvidia's RTX 5080 is Actually an RTX 5070
Wait, What?

 

[Krit]

Nvidia's RTX 5080 is Actually an RTX 5070

Take a hard look at the specs, and you'll see the RTX 5080 is just an RTX 5070 in disguise. Nvidia's greed may be at play, and...

www.techspot.com

Nvidia's RTX 5080 is Actually an RTX 5070
Wait, What?

It's extremely slow **70 class gpu for a **80 Super price

 

[x4it3n]

http://imgur.com/EI7BFTX

this is the _average_ across the whole benchmark, not just the highest single frequency seen

I still think that the 5090s could have had the same clocks if not limited to 600W...

Also are you going to receive some AORUS RTX 5090 MASTER, AORUS RTX 5090 XTREME WATERFORCE (Liquid Cooled) or ASUS Astral LC 5090 OC to review them?

 

[lexluthermiester]

The guy on High Yields says that yields with a full GB202 enabled are probably about 56% or more, so the 5090 with only 170SM out of 192SM definitely has a lot more yields... Maybe 75-80% would be a good number knowing that TSMC usually have great yields 1 year+ after release.

You might have missed the point of what I was saying..

probably about 56% or more

56% is terrible. But again, 56% of what? Perfect dies? I doubt that. Imperfect but usable dies, ok that's more plausible.

 

[x4it3n]

You might have missed the point of what I was saying..


56% is terrible. But again, 56% of what? Perfect dies? I doubt that. Imperfect but usable dies, ok that's more plausible.

Watch the video man... Instead of asking make your own research!
He said that among all the GB202 chips available at least 56% (probably more) can have a full GB202 (192SM) enabled. So the amount of perfertly working chips with 170SM (RTX 5090) probably gets around 80% yields.

 

[RUSerious]

RTX 2080 Ti, 3090, 4090 and now 5090 all had issues at release! Seems like a constant pattern at Nvidia...

NV always goes for large dies on their top GPU - yields are always poor when pushing closer to the reticle limit. But, as you point out, binning works just great for salvaging dice.

 

[lexluthermiester]

Watch the video man...

What video?

 

[RUSerious]

What video?

I would guess this one:

I haven’t watched it yet myself.

 

[Sir Beregond]

Nvidia's RTX 5080 is Actually an RTX 5070

Take a hard look at the specs, and you'll see the RTX 5080 is just an RTX 5070 in disguise. Nvidia's greed may be at play, and...

www.techspot.com

Nvidia's RTX 5080 is Actually an RTX 5070
Wait, What?

In the die it uses, it's still an 80 card which has been using the midrange die since the 600-series. With the exceptions of the 780 and 3080, they were using the x04 dies until Ada they changed up the numbering to x03.

On the other hand, this gen they have literally cut-down the 5080 to almost exactly half of the 5090, and typically its the 70 card that's half of the flagship.

 

[x4it3n]

What video?

Search for High Yield GB202 on YouTube

 

[Bwaze]

On the other hand, this gen they have literally cut-down the 5080 to almost exactly half of the 5090, and typically its the 70 card that's half of the flagship.

Also, not beating previous generation x090, and not by a small margin:

RTX 5080 is 12% slower than RTX 4090.

RTX 4080 is 30% faster than RTX 3090.

RTX 3080 is 31% faster than RTX 2080 Ti.

Look how they massacred my boy...

 

[lexluthermiester]

I would guess this one:

I haven’t watched it yet myself.

Interesting. TY. Just watched it.

Search for High Yield GB202 on YouTube

That guy stated himself that he was effectively guesstimating. That is not a citable source of info.

 

[x4it3n]

Interesting. TY. Just watched it.


That guy stated himself that he was effectively guesstimating. That is not a citable source of info.

He's guesstimating but all the other information online say that the yields are around 80-90% so he can't be that far off with the 5090.
TSMC 4NP node is still an Enhanced 5nm so the yields are definitely not bad. If Apple and Qualcomm are having good 3nm yields right now then older nodes definitely have better yields. If Nvidia chose a 3nm node then maybe availability would be an issue.

 

[lexluthermiester]

He's guesstimating but all the other information online say that the yields are around 80-90% so he can't be that far off with the 5090.

Sorry, I'm not buying it. At this stage of the 3nm process development, 70% viability is the absolute limit that can be reasonably expected, and I'm being generous. Until TSMC themselves verify yields levels and their method of breaking down the binning, there is no way to know for sure.

 

[x4it3n]

Sorry, I'm not buying it. At this stage of the 3nm process development, 70% viability is the absolute limit that can be reasonably expected, and I'm being generous. Until TSMC themselves verify yields levels and their method of breaking down the binning, there is no way to know for sure.

Except Blackwell is NOT on 3nm, it's on N4P an Enhanced 5nm that has been ouf for about 2 years already, therefore the yields are much higher! And the 5090 only has 170SM enabled out of 192SM... it's 22SM less than a full die! So of course the yields are going to be good.

 

[lexluthermiester]

Except Blackwell is NOT on 3nm, it's on N4P an Enhanced 5nm that has been ouf for about 2 years already, therefore the yields are much higher!

Incorrect. Blackwell is using an NVidia requested custom process based on the 3N lith.

NVIDIA Blackwell Architecture

Catapulting generative AI to trillion-parameter scale.

www.nvidia.com

It is newer than 3N and as such the yields are lower. It doesn't matter if it's a technically larger process, it's not as mature.

 

[Sir Beregond]

Except Blackwell is NOT on 3nm, it's on N4P an Enhanced 5nm that has been ouf for about 2 years already, therefore the yields are much higher! And the 5090 only has 170SM enabled out of 192SM... it's 22SM less than a full die! So of course the yields are going to be good.

Incorrect. Blackwell is using an NVidia requested custom process based on the 3N lith.

NVIDIA Blackwell Architecture

Catapulting generative AI to trillion-parameter scale.

www.nvidia.com

It is newer than 3N and as such the yields are lower. It doesn't matter if it's a technically larger process, it's not as mature.

Pretty sure 50-series is still on 4N. Can't speak for these others.

 

[lexluthermiester]

Pretty sure 50-series is still on 4N. Can't speak for these others.

You didn't actually read did you? Scroll up the page and soak in the context of the conversation. The Blackwell lith process is a custom modified 4nm version of the 3N process TSMC is currently manufacturing.

 

[Sir Beregond]

You didn't actually read did you? Scroll up the page and soak in the context of the conversation. The Blackwell lith process is a custom modified 4nm version of the 3N process TSMC is currently manufacturing.

Ah I did read, but re-reading I misunderstood what you were saying.

 

[lexluthermiester]

Ah I did read, but re-reading I misunderstood what you were saying.

Fair enough. Thought you were trolling me. It's been happening a lot with all the releases and event going on.

 

[Contra]

You didn't actually read did you? Scroll up the page and soak in the context of the conversation. The Blackwell lith process is a custom modified 4nm version of the 3N process TSMC is currently manufacturing.

Dear Sir, you have neither the information nor the terminology.
TSMC N3 technological processes are in no way connected with N5 and its derivatives N4. Yes, rtx50 is manufactured on exactly the same N4 as rtx40. Only and exclusively Blackwell server chips are manufactured on N4p, but not client ones, as you are trying to claim.
And stop spreading a your missinformation, when talking about your or someone else's dreams about rtx50 on 3nm

 

[lexluthermiester]

Dear Sir, you have neither the information nor the terminology.
TSMC N3 technological processes are in no way connected with N5 and its derivatives N4. Yes, rtx50 is manufactured on exactly the same N4 as rtx40. Only and exclusively Blackwell server chips are manufactured on N4p, but not client ones, as you are trying to claim.
And stop spreading a your missinformation, when talking about your or someone else's dreams about rtx50 on 3nm

Read, learn.

NVIDIA Blackwell Architecture

Catapulting generative AI to trillion-parameter scale.

www.nvidia.com

I'm not going hold your hand.
NVidia has CLEARLY stated that they're using a custom lith process for Blackwell. It is known that that process is both different from the process used for RTX4000 AND derived from TSMC 3N.

 

[x4it3n]

Incorrect. Blackwell is using an NVidia requested custom process based on the 3N lith.

NVIDIA Blackwell Architecture

Catapulting generative AI to trillion-parameter scale.

www.nvidia.com

It is newer than 3N and as such the yields are lower. It doesn't matter if it's a technically larger process, it's not as mature.

Nvidia always use Custom nodes, but N4P is a 4nm node aka an Enhanced 5nm node.

https://www.hardwareluxx.de/index.p...rchitektur-und-den-geforce-rtx-50-karten.html

TSMC 3nm node has a 200M+/mm² density.

 

[Contra]

then compare with.. 0% improve arch and perf/$|mm|tr|w and all other

 

[x4it3n]

then compare with.. 0% improve arch and perf/$|mm|tr|w and all otherView attachment 383756

He's the only one not getting it, even with proof he still won't believe us lol