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

 

[Karnak]

https://twitter.com/x/status/1320327830656483328

 

[BlitzWulf]

https://twitter.com/x/status/1320327891167727616

also tgp should include vram as well

https://twitter.com/x/status/1317847566659223552

 

[dzoni2k2]

Looks like the 3070 reviewers guide leaked

https://videocardz.com/newz/nvidia-geforce-rtx-3070-reviewers-guide-leaked

1 thing I found found interesting is that the 3070 has the same sm count as the 2080 and the 3080 has the same as 2080ti.
I'd bet money that if it werent for AMD being competitive this time around ,that NV would have been selling this configuration or something close to it as the xx80 card and people would have lapped it up
"2080ti perf for only 700 bucks!"

big navi cant come soon enough.

There was probably a late decision to move everything up the stack once they got hints of RDNA2 performance. That's in line with the whole 3080 is our flagship card mantra by Jensen.

 

[dzoni2k2]

These leaked N21 clocks suggest that MacOS spec dump was actually conservative. It won't surprise me if N22 is clocked even higher than 2.5GHz. Who knows where the real limit is.

 

[Glo.]

One thing to note about that 6700XT vs RTX 3070 - I think people should not look at this "battle" this way.

RX 6700 XT may lose 5-10% to RTX 3070, but it will also be massively more efficient.

 

[leoneazzurro]

One thing to note about that 6700XT vs RTX 3070 - I think people should not look at this "battle" this way.

RX 6700 XT may lose 5-10% to RTX 3070, but it will also be massively more efficient.

It should be also cheaper (192 bit bus, smaller die, less components/cooling needs even if with more RAM).

 

[exquisitechar]

One thing to note about that 6700XT vs RTX 3070 - I think people should not look at this "battle" this way.

RX 6700 XT may lose 5-10% to RTX 3070, but it will also be massively more efficient.

The impressive part to me is that it's a relatively small die with a 192-bit bus that punches above its weight class in terms of performance, 4GB more VRAM than the 3070 included. AMD looks like they are extremely well positioned this generation. The only question left is RT performance, where it isn't looking all that rosy according to the new rumors.

 

[Zoal]

One thing to note about that 6700XT vs RTX 3070 - I think people should not look at this "battle" this way.

RX 6700 XT may lose 5-10% to RTX 3070, but it will also be massively more efficient.

If raster performance is comparable NV will heavily push the RT difference in 'modern' games

 

[Glo.]

If raster performance is comparable NV will heavily push the RT difference in 'modern' games

As if there is enough RT'ed games on the marked for it to really matter .

The Ray Traced Games, to non-RayTraced ratio is like, what? 1 billion to one?

 

[Mopetar]

Navi 22 sounds pretty disappointing if it's just 40 CUs and loses to the 3070. I guess AMD made the massive Navi 21 GPU as a halo card to win benchmarks, but the cards they really want to sell are tiny <250mm² dies.

Isn't Navi 21 a 40 CU GPU like Navi 10? In that case the die size is probably pretty similar and it's a 250 mm^2 die against a ~400 mm^2 die.

I'm guessing we'll see the 6700 XT come in at $400 which is $100 less than a 3070. If it's only 10% lower performance on average then it still has a better value. There will probably be a 6700 for $300 - $350 that just had lower clocks since the yields for a smaller die like that should be quite good.

They also have the option of doing a GDDR6X refresh sometime next year when that becomes available if they want to have something to sell for closer to $500, especially if they keep the top 10% of dies in reserve and have a model with higher clocks and memory bandwidth that gets sold as an ultimate 1440p card.

Honestly I think we should be happy if there's a good stock of cards at these prices available for Christmas.

 

[Mopetar]

These leaked N21 clocks suggest that MacOS spec dump was actually conservative. It won't surprise me if N22 is clocked even higher than 2.5GHz. Who knows where the real limit is.

Well Apple sticks the cards in a tiny iMac frame with a cooling solution that cares more about being quiet so it's not surprising they'd have more conservative clock speeds or get better binned chips that don't need as much voltage to hit those speeds

One thing to note about that 6700XT vs RTX 3070 - I think people should not look at this "battle" this way.

RX 6700 XT may lose 5-10% to RTX 3070, but it will also be massively more efficient.

NVidia could stand to drop the voltage a tiny bit and get 95+ percentage of the performance at substantially less power draw. Hell, just make a standard mode and unleashed mode so that the reviews can present more sane power draw figures.

I think AMD will end up using the opportunity to push some cards that are also passed the peak of the efficiency curve, but a 6700 at clock speeds similar to a PS5 could come in close to 100W under a 3070.

 

[Glo.]

NVidia could stand to drop the voltage a tiny bit and get 95+ percentage of the performance at substantially less power draw. Hell, just make a standard mode and unleashed mode so that the reviews can present more sane power draw figures.

I think AMD will end up using the opportunity to push some cards that are also passed the peak of the efficiency curve, but a 6700 at clock speeds similar to a PS5 could come in close to 100W under a 3070.

Ampere release is rushed. Hence why they overvolted that silicon, for stability and getting as many stable products out of the gate. I have been telling people for months now, that we will not see Ampere GPUs this year. The only reason why we see Ampere GPUs released is because Nvidia blatantly rushed this release.

Remember. If you have an inferior product, you have to release it before your competition. RDNA2 paints as a very good product from AMD that might be better in some cases than Ampere. Imagine how big damage to mindshare Nvidia would get if they would release those GPUs LATER than AMD would release RDNA2.

 

[BlitzWulf]

https://twitter.com/x/status/1320346548937641986

 

[Glo.]

If its 255 TGP and it still averages 2.3 GHz game clocks then we should be really relaxed about its performance.

It will be absolute monster in efficiency .

 

[reb0rn]

Still top n21 will be slower then 3080, even with cash used more smart benefit for one game will be there and for other it will not

I don`t know why ppl hype it so match most will be hurt when they see there is no way to be faster then 3080

 

[Veradun]

Still top n21 will be slower then 3080, even with cash used more smart benefit for one game will be there and for other it will not

I don`t know why ppl hype it so match most will be hurt when they see there is no way to be faster then 3080

It will more probably trade blows with 3070 with some outliers up to 3080 in AMD biased tests.

It's AMD after all, they cannot possibly be competitive with nvidia

 

[Zstream]

Still top n21 will be slower then 3080, even with cash used more smart benefit for one game will be there and for other it will not

I don`t know why ppl hype it so match most will be hurt when they see there is no way to be faster then 3080

I’m not sure if you’re understanding, but these atm indicate a very close card to the 3080. It will likely lose to RT games, but pure throughput the card will be ok.

 

[dzoni2k2]

Still top n21 will be slower then 3080, even with cash used more smart benefit for one game will be there and for other it will not

I don`t know why ppl hype it so match most will be hurt when they see there is no way to be faster then 3080

Jensen is that you?

 

[andermans]

Well Apple sticks the cards in a tiny iMac frame with a cooling solution that cares more about being quiet so it's not surprising they'd have more conservative clock speeds or get better binned chips that don't need as much voltage to hit those speeds

I'm still not convinced those frequencies are of an actual SKU and instead suspect that these are there purely as placeholder or testing values (or maybe at boot?) The driver when loaded will upload its own frequency tables extracted from the vbios, and AMD tends to share the SMU firmware (from where the tables were extracted) across all SKUs with that die.

 

[TESKATLIPOKA]

If its 255 TGP and it still averages 2.3 GHz game clocks then we should be really relaxed about its performance.

It will be absolute monster in efficiency .

It's still just GPU+VRAM, I think TBP will be 285W or more. If this is just the 72CU version, then the 80CU version will be over 300W TBP even as a reference card.

 

[Veradun]

It's still just GPU+VRAM, I think TBP will be 285W or more.

I'd bet on the usual 295W at max unless they really think they can get the crown pushing the envelop to win against the 3090

 

[Karnak]

If this is just the 72CU version, then the 80CU version will be over 300W TBP even as a reference card.

You sure about that? All that 255W/289W TGP is with clocks somewhere around 2300MHz (or sometimes even a bit higher).

If AMD stays around 20xxMHz for the reference cards, I'm sure the TBP won't exceed 300W. Even for the 80CU part.

 

[TESKATLIPOKA]

You think even 255W TGP model is an OC model, not just the one with 289W TGP?

 

[Karnak]

You think even 255W TGP model is an OC model, not just the one with 289W TGP?

Yes, because the first tweet from Schur in that regard was about an AiB model with 255W TGP and (up to) 2400MHz.

Edit: This one

https://twitter.com/x/status/1317527480866144257

 

[Shivansps]

I dont understand why try to push the Navi 22 agaist the GTX3070 in a dumb corporate attempt to try to get a big margin out of a card that will not sell very well, place it against the 3060 at the same price and win the mainstream sector to gain market share!

If AMD has a superior product they should be using it to provide a significant better performance than they Nvidia counterparts to win market, not get huge margins out of products that will sell in low volume, SPECIALLY with Nvidia having supply issues.