An article about RTX 4080 16/12GB

[birdie]

@W1zzard

Would be nice if your editors ran an article on these two GPUs RTX 4080 16/12GB comparing them to previous generations of NVIDIA GPUs in terms of

• die size
• bus width
• CUDA cores count in relation to the top tier product

And then made a conclusion where they actually stand name-wise.

Starting with e.g. the GTX 10 series or even the GTX 900 series.

 

[W1zzard]

NVIDIA Cancels GeForce RTX 4080 12GB, To Relaunch it With a Different Name

NVIDIA has decided to cancel the November 2022 launch of the GeForce RTX 4080 12 GB. The company will relaunch the card under a different name, though it didn't announce the replacement name just yet. The naming of the RTX 4080 12 GB was cause for much controversy. With the RTX 40-series "Ada,"...

www.techpowerup.com

 

[agent_x007]

AD103 is a new chip number (usually we had xx2 or xx4 up to this point), and to me it's around where "G92" was back in 2007 (with G80 being "top dog"), so... xx70 series ?
Price point is simply adjusted up, because nvidia margins/quality.

Regardless of how it's named :
If performance will be where they (buyers) want it to be and the price is right - they will buy it.

 

[dj-electric]

I was talking about a deep dive into previous generation cards and how tiers stacked up earlier. This article doesn't mention any previous gen products or their characteristics.

I know about the announcement.

A pertinent discussion and comments on reddit. People argue RTX 4080 16GB is actually a xx60 Ti class product at best.

Tiers of products is something that marketing departments are responsible for, not egineers.
When NVIDIA has migrated from Fermi to Kepler, they have named the GTX 680 like they did because of its performance, not die size, obviously.

You always have the TPU GPU database for these kinds of comparisons. I believe that personal opinions on the matter could only take away from what people may or may not think of NVIDIA's marketing strategies.

 

[Aquinus]

You always have the TPU GPU database for these kinds of comparisons. I believe that personal opinions on the matter could only take away from what people may or may not think of NVIDIA's marketing strategies.

^ This. @birdie you can crunch these numbers yourself and report back if it's something you're interested in. You have all of the tools necessary to do it.

 

[Vayra86]

I was talking about a deep dive into previous generation cards and how tiers stacked up earlier. This article doesn't mention any previous gen products or their characteristics.

I know about the announcement.

A pertinent discussion and comments on reddit. People argue RTX 4080 16GB is actually a xx60 Ti class product at best.

The stack changes. I mean its clear that the Geforce stack wants to accommodate different types of user segments, it always has. Over time we saw (semi) pro markets invade the Geforce stack, but we also have a much wider spread of GPU performance throughout, to cater to several resolutions that also require wildly different performance. 1080p > 4K is x4. That's a major departure from the days of, say, Kepler, where people might alter between 720p/1080p at best, no ultrawides, 1440p was only just beginning with korean 27 inchers, etc. Even just for gaming.

People argue, but all of this isn't rocket science. The positioning of the 4090 versus the 4080 16GB and the huge gap between them, makes it pretty clear where the rest of the stack lands. Is it an xx60ti class product at best, yeah? If so, how on earth are they going to fit meaningful product underneath? Its the exact reason that 12GB card got axed.

We can already predict Ada is going to have to be price competitive to even be more than an afterthought of Ampere. Because that's really what it is. Some proprietary bullshit features on top of a shrink of Ampere, and the only way to go faster is by going bigger. The 4090 is the sorry excuse that results from that lack of progress and the 4080 16GB confirms this gen won't do a whole lot for us compared to the last.

However, that 4080 16GB does compare to 3090/3090ti levels of performance. Calling that an xx60ti is not being fair about it. Its an x70, at least, and an x70ti at best. One could also defend the idea that its pretty amazing how they get the results they do on such a small bus/low amount of hardware resources. Big part of that is of course clocking which is miles better than anything Samsung.

^ This. @birdie you can crunch these numbers yourself and report back if it's something you're interested in. You have all of the tools necessary to do it.

This is my TL DR

fwiw, we don't really need any kind of consensus on this, do we. In the end its just a name, people pay for price/perf.

 

[wolf]

I'm not sure I see the merit in W1zzard covering this in reviews, to an extent it's immaterial to what he wants to show us.

It has also changed many times over the years, what really matters the most is Price and Performance, the product name, die it uses etc is interesting and academic at best but not exactly all that relevant in the context of a GPU review, at least in the sense you mean where it is compared against products of old and I'd imagine has it's own full page dedicated to that.

Perhaps just start a thread about it with your own comparison of dies over the generations and numbers to keep the discussion focussed on the topic you're interested in?