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

 

[leoneazzurro]

I think this is redacted.No way big navi eat more than 3x more power than consoles.I think igor lost it.Well 8 more days.

AMD Radeon RX 6000 based on Navi 21 XT to feature 320W TBP, 16 Gbps GDDR6 memory - VideoCardz.com

A fresh report from Igor’sLAB puts a new claim on Navi 21 power consumption. AMD Navi 21 XT to feature 320W TGP/TBP Igor Wallossek published his first ‘big leak’ on Big Navi. According to his data, AMD Navi 21 XT will feature TGP at around 320W, not 255W as reported earlier. The difference comes...

videocardz.com

Btw TGP is for entire card.Dont know what the fuck igor doing?

Graphics Cards: TDP and TGP (and don't forget TBP, GCP and MPC...) | Geeks3D

Graphics Cards: TDP and TGP (and don't forget TBP, GCP and MPC...)

www.geeks3d.com

Well he started to give the 3080 24 Gbytes of GDDR6x, makes a lot of arbitrary assumptions and generally he seems too eager to demonstrate something to be credible.
Not saying that the Navi21 cannot reach 320 W but not in the way he's trying to push it.

 

[uzzi38]

Did nobody notice how his board estimate for Navi21 is nearly 2x what you get with Navi10?

5700XT is 180W core power, 225W board.

Estimations are estimations, treat them as such.

 

[Timorous]

Well he started to give the 3080 24 Gbytes of GDDR6x, makes a lot of arbitrary assumptions and generally he seems too eager to demonstrate something to be credible.
Not saying that the Navi21 cannot reach 320 W but not in the way he's trying to push it.

Even if true provided RDNA2 has 1:1 perf/watt scaling like the 5500XT to 5700XT has a 320W part should be 10% ahead of the 3090 if we take the 'TBP' of the Series X GPU to be 150W which is probably slightly inflated. (DF 210W peak from wall means at most 190W used by the components so would mean the CPU, SSD, disc drive etc are only using 40W which is on the lowe end).

 

[dzoni2k2]

AIB cards could be over 300W, why not? The reference ones are definitely staying below though. That 2 slot cooler is not meant for anything above 300W. It's also only 2x8. Ah yes, and we have to account for USB-C.

 

[Saylick]

DF consists of multiple persons, Rich, Dark1x, Dictator...
Dictator on messages boards like Neogaf, ResetEra etc. is a very(!) staunch Nvidia fan, so I personally wouldn't take any classification of RTG products from him without any less biased second opinions. In general Nvidia's following on PC is irrationally strong, that's the uphill battle AMD has to fight in the DIY GPU market.

For what it's worth, Dictator is also a huge MS/Xbox fanboy, which is why he takes a dig at Sony/PS whenever he can. In this case, he's downplaying how fast the PS GPU is.

 

[leoneazzurro]

Even if true provided RDNA2 has 1:1 perf/watt scaling like the 5500XT to 5700XT has a 320W part should be 10% ahead of the 3090 if we take the 'TBP' of the Series X GPU to be 150W which is probably slightly inflated. (DF 210W peak from wall means at most 190W used by the components so would mean the CPU, SSD, disc drive etc are only using 40W which is on the lowe end).

Question is, for 5700XT the "GPU" is rated 180W by AMD: if this number would not have included the VRM losses, then by Igor's calculation it would have accounted for 211W for GPU+VRM only, so Total Board Power would have been in the high 250W. Instead, it is not.

 

[Veradun]

My problem with 40CU N22 is that it is rumoured to be over 300mm² and with the Series X we know AMD could design a 56CU part with a 320bit bus and 64 ROPs in a similar die area.

In fact this is my problem with all the rumours. The die size vs performance vs CU count make little sense. For a 300mm² N22 to make sense it needs to be at 2080Ti/3070 performance, any less and they may as well have made desktop versions of the Series X GPU because 56CUs @ 2.2 Ghz would be 2080Ti level.

If N22 is closer to 350mm² then it needs to perform even better. This would be between 3070 and 3080 performance so if AiBs only have N22 it does line up with the leaked performance expectations but that implies the teased big navi part was the bottom part.

So yea, the rumours make little sense full stop.

My first estimate for a 505mm2 part was around 100CUs and 384b memory controller.

I agree with you all this is pretty confusing :>

 

[insertcarehere]

If N23 is 236mm² then it needs to be at or slightly above 5700XT.

I really don't get the leaks. Take the series X GPU at around 300mm². Build it using the same library as renoir and it shrinks down to 205mm². That would explain the smaller bus because you would not fit a 320 bit bus on a die that small and 256 might be a challenge so you make it 192bit and those transistors you save on the bus you use for Infinity Cache, and maybe then some so you grow it back to 230mm² ish like the N23 leak but now you have a 2080Ti/3070 tier part in a 230mm² die area. Give it 12GB ram and you are good to go.

An 80CU, 56CU, 40CU, 24CU stack makes sense and it would make Big Navi not that big (350mm²-400mm²) and explain why they have a smaller bus.

I guess I will have to wait till the 28th but I keep hoping something will leak that makes sense of this nonsense.

I think it's a reasonable assumption that there isn't much, if any, leeway for discrete RDNA2 GPUs to be more dense than the console designs we're seeing on contemporary processes. TSMC 7nm fab space hasn't been plentiful or cheap for quite a while, so MS and Sony have probably used every trick in the book to make their APUs die sizes as small as possible to reduce BOM given the tough price point targets the consoles need to hit.

Still, whatever rumors we're seeing on Navi 2x have definitely been plenty confusing, thankfully there's only be a week to go.

 

[Krteq]

https://twitter.com/x/status/1318470450914205696

 

[Geranium]

One thing need to remember, non of the leakers had no info about Ampere's CUDA core number or configuration or even double FP32 output from each CUDA core. Not even proper VRAM configuration.

 

[Glo.]

I think this is redacted .No way big navi eat more than 3x more power than consoles.I think igor lost it.Well 8 more days.

AMD Radeon RX 6000 based on Navi 21 XT to feature 320W TBP, 16 Gbps GDDR6 memory - VideoCardz.com

A fresh report from Igor’sLAB puts a new claim on Navi 21 power consumption. AMD Navi 21 XT to feature 320W TGP/TBP Igor Wallossek published his first ‘big leak’ on Big Navi. According to his data, AMD Navi 21 XT will feature TGP at around 320W, not 255W as reported earlier. The difference comes...

videocardz.com

Btw TGP is for entire card.Dont know what the fuck igor doing?

Graphics Cards: TDP and TGP (and don't forget TBP, GCP and MPC...) | Geeks3D

Graphics Cards: TDP and TGP (and don't forget TBP, GCP and MPC...)

www.geeks3d.com

Don't call Igor that "he lost it". That is shooting the messenger just because we don't like his messages.

And he may just be bringing what AIBs are feeding him with.

320W of power he says?

I can tell you guys that this might be an indication that AMD has decided to go all out, and not leave anything on the table.

When I wrote the performance targets for Navi dies, or rather CU counts, I had information that initial performance targets were for 250W boards. As in 250W in total power drawn by the boards. It was at the time when 2.1 GHz max boost speeds were on the table.

Now we are seeing 2400 MHz clock speeds at 300W board power. If we take this in the context - everything starts to make sense.

How is that 2.4 GHz clock speed going to affect performance?

I don't believe anybody looked at it, but yesterday I posted comparison of performance between what AMD demoed during Zen Keynote, assuming it was full specs for Navi 21 die, with Xbox Series X, wondering if anybody will see something.

52 CUs clocked at 1825 MHz = 12.1 TFLOPs.
80 CUs clocked at 2410 MHz = 24.5 TFLOPs.

2X performance over Xbox Series X, at least - the theoretical maximum performance. 2.4 GHz at 300W board power means that AMD might be going all out. Straight up for win.

Will they win? We'll see...

 

[Krteq]

One thing need to remember, non of the leakers had no info about Ampere's CUDA core number or configuration or even double FP32 output from each CUDA core. Not even proper VRAM configuration.

What are you talking about? They've been pretty close with SM count and power consumption numbers.

Also, we had info about 10GB for RTX 3080 for weeks ahead of launch.

 

[Glo.]

One thing need to remember, non of the leakers had no info about Ampere's CUDA core number or configuration or even double FP32 output from each CUDA core. Not even proper VRAM configuration.

Yes, there were people talking about 2X the CUDA core numbers before the launch, but the rumor was... unbelievable, so even they were not sure about its credibility.

 

[Head1985]

Don't call Igor that "he lost it". That is shooting the messenger just because we don't like his messages.

And he may just be bringing what AIBs are feeding him with.

320W of power he says?

I can tell you guys that this might be an indication that AMD has decided to go all out, and not leave anything on the table.

When I wrote the performance targets for Navi dies, or rather CU counts, I had information that initial performance targets were for 250W boards. As in 250W in total power drawn by the boards. It was at the time when 2.1 GHz max boost speeds were on the table.

Now we are seeing 2400 MHz clock speeds at 300W board power. If we take this in the context - everything starts to make sense.

How is that 2.4 GHz clock speed going to affect performance?

I don't believe anybody looked at it, but yesterday I posted comparison of performance between what AMD demoed during Zen Keynote, assuming it was full specs for Navi 21 die, with Xbox Series X, wondering if anybody will see something.

52 CUs clocked at 1825 MHz = 12.1 TFLOPs.
80 CUs clocked at 2410 MHz = 24.5 TFLOPs.

2X performance over Xbox Series X, at least - the theoretical maximum performance. 2.4 GHz at 300W board power means that AMD might be going all out. Straight up for win.

Will they win? We'll see...

its 355W for top not 300w.2nd tier is 320w and 3rd tier still crazy 290w.
They already showed performance.Even if its only 2nd tier by igor leak its still 320w and its slower than 3080.
It looks even worse than Ampere alias FERMI 2.0.So i call that redacted.Consoles are super efficient and in desktop they will release something even worse than ampere?


Profanity is not allowed in the tech forums.

AT Mod Usandthem

 

[Glo.]

its 355W for top not 300w.2nd tier is 320w and 3rd tier still crazy 290w.
They already showed performance.Even if its only 2nd tier by igor leak its still 320w and its slower than 3080.

No. They have not shown its performance .

 

[Head1985]

No. They have not shown its performance .

This is 10% slower than 3080.Even if its not top GPU and only 2nd it will be 320w by igor leak.

 

[Glo.]

This is 10% slower than 3080.Even if its not top GPU and only 2nd it will be 320w by igor leak.
View attachment 32060

Ok, fine. It will be bad as everything AMD always does.

 

[Timorous]

Ok, fine. It will be bad as everything AMD always does.

It will be the Bulldozer of the GPU world. Clocks to the moon, uses lots of power but barely performs better than the consoles.

 

[Head1985]

Ok, fine. It will be bad as everything AMD always does.

Well if you believe igors leak(you do) then yes it will be crap.Power hungry like FERMI 2.0(ampere) and slower.I dont believe it and call that leak bull***.

 

[Glo.]

Well if you believe igors leak(you do) then yes it will be crap.Power hungry like FERMI 2.0(ampere) and slower.I dont believe it and call that leak bullshit.

I can believe that AMD decided to go all out and the GPUs will draw up to even 300W of power during gaming.

 

[Head1985]

So you dont believe igor then?Because top card is 355w not 300w.300w is for slowest SKU

 

[Glo.]

So you dont believe igor then?Because top card is 355w not 300w.300w is for slowest SKU

Im trying to see past the rumors that he brings. What is "behind" them, where those information can come from.

Remember, he is only reporting what he has been told by AIBs and what he assumes based on this.

Partially correct, or partially full data can lead you to wrong assumptions.

 

[kurosaki]

So you dont believe igor then?Because top card is 355w not 300w.300w is for slowest SKU

That's way to little. From what Ive heard, the one below top model will draw 500W for 10 TF. Top card is a quad slot, water cooled four chiplet 750W beast, capable of Pong, in 60 fps. Quake 1, without RT almost hits the 30 fps mark.

 

[Stuka87]

Quake 1, without RT almost hits the 30 fps mark.

30? That's got to be with line skipping enabled.

 

[Veradun]

30? That's got to be with line skipping enabled.

I think he was actually referring to Wolfenstein 3D