Question Speculation: RDNA2 + CDNA Architectures thread

[uzzi38]

All die sizes are within 5mm^2. The poster here has been right on some things in the past afaik, and to his credit was the first to saying 505mm^2 for Navi21, which other people have backed up. Even still though, take the following with a pich of salt.

Navi21 - 505mm^2

Navi22 - 340mm^2

Navi23 - 240mm^2

Source is the following post: https://www.ptt.cc/bbs/PC_Shopping/M.1588075782.A.C1E.html

 

[zinfamous]

But see this is the part that currently seems unreconcilable with 6800xt being navi21xt, 72cu part.

With TSMCs reported defect density in its 7nm process of 0.09, the ratio of good dies to defective dies is around 2:1.

https://www.anandtech.com/show/16028/better-yield-on-5nm-than-7nm-tsmc-update-on-defect-rates-for-n5
This report indicates and the graphs show that for 7+ nm defect density is at least as good as 7NP, if not better due to less stages in the lithography process.

So, for every 72cu die shipped to an AIB partner, amd are stock piling 2 full 80cu dies?


Below is a list of AMD gpu releases. For each segment, apart from the VII, the full die is always labeled as XT, with the cut XL or previously PRO. The XTX chip
if there has been one has remained in the same segment as the XT die.

View attachment 32278

Why is AMD seemingly changing this around this time, releasing a cut down 72cu die as an XT, keeping the full 80cu die as xtx, and shifting the XTX die to a new product segment, 6900xt

Then there is the report that the 6900xt will be limited in supply, but why would that be the case if it is the 80cu full die of the navi21 chip, when from TSMCs own reported defect density we can conclude that 2 full dies are created to 1 partial die. If the 80cu die is limited, then the 72cu is even more limited.


What does reconcile these differences is that the 6800xt is a different chip altogether, a 64cu ~400mm2 full die, while the 6900xt is the 80cu ~500mm2 full die. This way AMD can choose to produce less 80CU dies, and will be limited in supply as rumour suggests.

The leaked numbers though dont see to line up with a 64cu chip at 2.2 ghz however

I may have missed it, but where are we getting indisputable confirmation that, if 6800 is part of the full die stack, that there really is a 6800XT?

I've seen people use that name when making their tweets or whatever, but other than Lisa Su a few weeks ago showing "Big Navi," and, I recall, nothing about 6800 or 6900 or X or XT, I haven't really see anything confirmed from AMD that "6800XT = Big Navi = 80CU die." I don't care if the best leaks are 100% true; I won't consider any of it unassailable until we get the details next week, direct from AMD as part of an official release.

Again, as true is at all might be right now, we're still getting the exact same arguments of "certainty" that we got with Polaris and especially Vega--I don't care if the architectures are completely different--the same arguments being made over raw calculations that maybe aren't considering proper performance scaling? I mean, paper calculations are one thing, but how does it really work in the end? Anyway, I know that AMD is in a much different place now without Raja running his mouth off, and that now we do have some decent comparisons to make with what (Well, what we think we know) with XboxSX and PS5--still don't know how games actually perform there--again, nothing has been released--but past history is still the better predictor of future results, even if it doesn't always work out that way. It still is.

 

[Glo.]

You catastrophically misunderstand me, as not once did I ever suggest such a thing. Not even close. I am not commenting so much about performance, but about what they can release to market, and how the market responds to those releases. They are still as supply limited as everyone will be and, whether or not you are correct (and I tend to think you are), there is literally no reason to accept rumors and leaks that go against the well-known norm. It doesn't matter how different AMD is now, from 10, even 5 years ago when Zen was brand new, RDNA is looking like the early stages of Zen, just now being able to predictably compete. ...so I think that's where we are.

Plenty of people disbelieved everything being leaked about Zen (I didn't, so picked up a pile of shares at $4.60 at the time, much to my "being very smart"--that's a joke; no one is actually that smart. It's really just luck), and while I *think* we are at a similar moment with RDNA2, there is no logical reason to assume the situation repeats. Even with Zen, almost 5 years ago now, they are only just now catching up to Intel in terms of market. nVidia is not nearly as far behind as Intel has been, (and I mean realistically).

You mistake that I am betting against them. I most certainly am not.

It still is part of my point. We should NEVER judge upcoming release by what was in the past.

Be it looking at the historical performance, be it by how many GPUs AMD will have available.

If it isn't apparent, Ampere release was rushed, hence why the scarce availability. AMD's release is perfectly planned, and perfectly on track since months. So its not that AMD has been caught by Nvidia's release. Those AMD GPUs most likely were in mass production for weeks if not months.

Of course, you still may say something about never underestimating the demand for GPUs, as has been in the past, with crypto crazes. And that will be very fair point . Espcecially if the performance leaks will turn out true.

 

[Vope45]

Which is?

Kicking Intel's butt when they .... up?

So why can't AMD possibly do the same thing to their biggest GPU competitor, when their biggest GPU competitor is doing the same thing, as Intel?

Its only human mind problem, that it can see only where the hockey Puck was, instead of where it goes.

It's more Intel kicking themselves than AMD doing so. Skylake core was released 2015 and it took AMD 4 years to catch up with Intel having zero changes in architecture. AMD still have a long long way ahead.

AMD can't possibly compete with Nvidia atm, with money from Zen 2 they probably can in a few more years. But at the moment, Nvidia are still leaps and bounds ahead of AMD.

 

[BarkingGhostar]

It's more Intel kicking themselves than AMD doing so. Skylake core was released 2015 and it took AMD 4 years to catch up with Intel having zero changes in architecture. AMD still have a long long way ahead.

AMD can't possibly compete with Nvidia atm, with money from Zen 2 they probably can in a few more years. But at the moment, Nvidia are leaps and bounds ahead of AMD.

When I think of 'leaps and bounds' I think for a given $ that someone if putting out twice the frame rates at the same gaming settings between two comparable products. Is this the case, really?

 

[raghu78]

AMD can't possibly compete with Nvidia atm, with money from Zen 2 they probably can in a few more years. But at the moment, Nvidia are still leaps and bounds ahead of AMD.

In GPU market share and overall revenues/profits yes. On performance lets wait for RDNA2 reviews.

 

[linkgoron]

It's more Intel kicking themselves than AMD doing so. Skylake core was released 2015 and it took AMD 4 years to catch up with Intel having zero changes in architecture. AMD still have a long long way ahead.

AMD can't possibly compete with Nvidia atm, with money from Zen 2 they probably can in a few more years. But at the moment, Nvidia are leaps and bounds ahead of AMD.

Nvidia have had two relatively mediocre launches. Vega was terrible, but AMD wasn't that far behind with Navi10 (5600/5700), they were easily better than 2060(s) and 2070 IMO, and it won't be a shocker if they can compete with the 3080/3070 in both performance and perf/watt while also having significantly more memory.

 

[Vope45]

When I think of 'leaps and bounds' I think for a given $ that someone if putting out twice the frame rates at the same gaming settings between two comparable products. Is this the case, really?

Gaming isn't the most profitable segment for Nvidia anymore.

In GPU market share and overall revenues/profits yes. On performance lets wait for RDNA2 reviews.

Might want to include mindshare as well.

 

[Karnak]

AMD can't possibly compete with Nvidia atm, with money from Zen 2 they probably can in a few more years. But at the moment, Nvidia are still leaps and bounds ahead of AMD.

Few more years? There are just five more days left until wednesday.

 

[Glo.]

But at the moment, Nvidia are still leaps and bounds ahead of AMD.

Only in human mindshare, as we can see.

 

[carrotmania]

 

[moinmoin]

AMD's release is perfectly planned, and perfectly on track since months. (...) Those AMD GPUs most likely were in mass production for weeks if not months.

Really looking forward to seeing what the availability for AMD's CPUs and GPUs will be. The skeptic in me expects way too conservative production numbers that sell out too quickly as AMD calculated demand way too conservatively (like is still the case for e.g. Renoir chips). The optimist in me expects AMD to have planned with RDNA2 and Zen 3 as the breakthrough products from the beginning and allocated accordingly high production numbers. We'll which one of the two is closer to reality.

 

[maddie]

Gaming isn't the most profitable segment for Nvidia anymore.



Might want to include mindshare as well.

True, but if they aren't getting the large revenues from gaming, their R&D will be begin to be affected. Do not assume margins can replace revenue streams.

 

[Vope45]

True, but if they aren't getting the large revenues from gaming, their R&D will be begin to be affected. Do not assume margins can replace revenue streams.

But ultimately margin is where business happens. I suspect their pace in gaming will soon dwindle.

 

[Elfear]

Okay, now someone's just adding bars to the graph for fun. Labels should be "AMD Big Navi Hype Train Edition".

 

[zinfamous]

Really looking forward to seeing what the availability for AMD's CPUs and GPUs will be. The skeptic in me expects way too conservative production numbers that sell out too quickly as AMD calculated demand way too conservatively (like is still the case for e.g. Renoir chips). The optimist in me expects AMD to have planned with RDNA2 and Zen 3 as the breakthrough products from the beginning and allocated accordingly high production numbers. We'll which one of the two is closer to reality.

I think we can predict the availability based on listed MSRP for Zen 3. ...it's not going to be good. But, these are also the X chips. Not sure when they are going to release the non-X Zen chips, maybe early 2nd quarter? but if the current MSRP hold to actual vendor prices by that time, those will probably be only $30-50 less at their respective tiers. I mean, a $250 5600 really isn't the worst price we've seen in relative generational CPU pricing...

 

[Kenmitch]

Lost track as far as if the hype-train is losing or picking up momentum. Not much longer to wait until at least some of the confusion is cleared up. The 28th can't come soon enough.

Fake Rumors Suck: Will the Real RX 6000 Please Stand Up?

Casting a wary eye

www.tomshardware.com

 

[Karnak]

AMD: Be patient and wait for Benchmarks.
NV: Just buy it.

Tomshardware in a nutshell.

 

[blckgrffn]

Lost track as far as if the hype-train is losing or picking up momentum. Not much longer to wait until at least some of the confusion is cleared up. The 28th can't come soon enough.

Fake Rumors Suck: Will the Real RX 6000 Please Stand Up?

Casting a wary eye

www.tomshardware.com

Take that bucket of cold water somewhere else, man. I don't want it

(Decent article, but there was waaaaaay more AMD "driven" hype on those launches for sure.)

 

[Mopetar]

But see this is the part that currently seems unreconcilable with 6800xt being navi21xt, 72cu part.

With TSMCs reported defect density in its 7nm process of 0.09, the ratio of good dies to defective dies is around 2:1.

Even if a die isn't defective, there's no guarantee that the full 80 CUs will be able to hit a clock target that AMD wants for a product. That gives them the option of having two bins for the full die, but depending on how many chips fall into each they may be better off taking a fair number of those full dies and disabling the worst 8 CUs in order to hit higher clock speeds. Because GPUs rarely scale linearly in terms of performance with such a massive number of CUs, it's entirely possible that disabling some of them and running at a higher clock speed actually results in a GPU with better performance.

AMD might also have a customer like Apple that's bought up a large quantity of their new chips. If that's the case AMD is going to make sure that they get supplied before they worry about getting their full consumer stack launched. I don't know if that's the case and Apple usually doesn't buy top-end GPUs, but it is possible. It makes more sense if we assume that AMD build both types of memory controllers into the chips so they didn't need to create a separate version. Even if that isn't the case, as other users have pointed out AMD makes way more money from a wafer of Zen 3 chiplets than consumer GPUs. However, if these 80 CU dies were going to the pro market where they can easily charge $2000 or more per card, then it makes a lot of financial sense for AMD to put the best chips towards that first.

If the supposed leaks are true, people will wait for the top AMD card, especially if the AMD announces a price that's better than what NVidia is charging. Even if people were tempted to get something else because they can't stand waiting, it isn't like a 3090 or a 3080 are easy to get either. If AMD manages to come out on top, they're probably going to get a lot of people willing to shell out $$$$ for the best card money can buy that never paid attention to them before and they have the same insane demand that NVidia is experiencing. Even if they have more cards at launch, they'll still sell out fast.

 

[leoneazzurro]

Uhmm...

[PATCH] drm/amdgpu: correct the cu and rb info for sienna cichlid

lists.freedesktop.org

What does it mean?

 

[soresu]

We're quickly approaching our destination.

Brace yourselves, leaks are becoming torrential.

You're traveling through another dimension, a dimension not only of sight and sound but of mind. A journey into a wondrous land whose boundaries are that of imagination. That's the signpost up ahead - your next stop, the Twilight Zone!

 

[soresu]

AMD can't possibly compete with Nvidia atm, with money from Zen 2 they probably can in a few more years. But at the moment, Nvidia are still leaps and bounds ahead of AMD.

In software yes, sadly they cannot possibly hope to catch up alone.

In hardware on the other hand it is an entirely different ballgame.

AMD and nVidia have effectively reversed positions now with AMD concentrating on gaming only GPU's while nVidia is sticking many more transistors than are actually being used on any given gaming GPU.

That's not to say that nVidia can't take AMD's path of diverging compute/HPC/ML and gaming uArch's - but there is little evidence of that as yet.

 

[lightmanek]

Uhmm...

[PATCH] drm/amdgpu: correct the cu and rb info for sienna cichlid

lists.freedesktop.org

What does it mean?

I fear we will soon discover that Hype-Train have one more gear it can engage to go even faster!
I always liked Jawed's speculation about CU count not matching the claimed die size. I might not quite believe in 160CU's but a number in between, like 112CU's would still make more sense than "just" 80 ...

This is one of the worst years for human civilisation and yet one of the best years I can remember when comes to hardware lanches. After a decade of boring and medicore produt updates, I'm more than exited for what we are getting this year on PC / Consoles / Mobiles

 

[DDH]

I fear we will soon discover that Hype-Train have one more gear it can engage to go even faster!
I always liked Jawed's speculation about CU count not matching the claimed die size. I might not quite believe in 160CU's but a number in between, like 112CU's would still make more sense than "just" 80 ...

This is one of the worst years for human civilisation and yet one of the best years I can remember when comes to hardware lanches. After a decade of boring and medicore produt updates, I'm more than exited for what we are getting this year on PC / Consoles / Mobiles

Well, not if you're a NVIDIA fan

But anyway, the die sizes definitely didn't add up to the rumoured cu counts. But that was assumed to be due to the infinity cache. Still, this seems to be next level sandbagging AMD has engaged in

 

[maddie]

Uhmm...

[PATCH] drm/amdgpu: correct the cu and rb info for sienna cichlid

lists.freedesktop.org

What does it mean?

Ha. They've done a Nvidia.