Question Speculation: RDNA2 + CDNA Architectures thread

Page 200 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

uzzi38

Platinum Member
Oct 16, 2019
2,565
5,572
146
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
 

Glo.

Diamond Member
Apr 25, 2015
5,657
4,409
136
Moore's Law is Dead claims that 6700XT is not going to firmly beat the RTX 3060 Ti, and AMD wants to price this SKU at 350$, and cut down version at sub-300$ price range.

According to his info one of SKUs(there are supposed to be at least two other, cut down SKUs) is built/designed specifically for low cost.

Knowing this - RTX 3060 price of 299$, and the struggle for AIBs to match this price point tells a lot. A LOT!
 

beginner99

Diamond Member
Jun 2, 2009
5,208
1,580
136
Moore's Law is Dead claims that 6700XT is not going to firmly beat the RTX 3060 Ti, and AMD wants to price this SKU at 350$, and cut down version at sub-300$ price range.

If the 6700x is $350 then that makes the 6800(XT) kind of pointless unless they also get a $150+ markup in the actual stores. But even compared on MSRP this doesn't make much sense but I do hope it is actually true.
 

TESKATLIPOKA

Platinum Member
May 1, 2020
2,329
2,811
106
Moore's Law is Dead claims that 6700XT is not going to firmly beat the RTX 3060 Ti, and AMD wants to price this SKU at 350$, and cut down version at sub-300$ price range.

According to his info one of SKUs(there are supposed to be at least two other, cut down SKUs) is built/designed specifically for low cost.
If the performance is comparable, then It's not surprising AMD will price It under RTX 3060ti, but then there will be nothing in the $379-579 range, that's just too big of a blind spot.
I also don't see much sense in having 2 cutdown skus from N22, when they have N23 and N24.
 

moinmoin

Diamond Member
Jun 1, 2017
4,933
7,619
136
Still limited to RDNA2 and Zen 3 so far, but apparently that's only down to validation?
Update on AMD CPU support: SAM will likely be limited to Zen 3 since prior gens were only emulating the PDEP and PEXT instructions (native support made them 250 times faster according to Ian) that may be needed to access all of Resizable BAR. Intel supports them going back to Haswell (there BIOS support is the limiting factor for SAM support).

See eg:
 

Glo.

Diamond Member
Apr 25, 2015
5,657
4,409
136
Yep. And the funniest part is that not Only Asus offers RBAR support for 400 Series Intel Chipsets/Comet Lake CPUs. MSI and Asrock also announced their support for 400 series Intel chipsets.

So if you want RBAR support YOU HAVE TO update to Zen 3, while Intel will offer this on both: 10th and 11th gen, and from 400 series Chipsets.

People who have built 400 series/10th gen CPUs have very, very decent upgrade paths to Rocket Lake.
 

Mopetar

Diamond Member
Jan 31, 2011
7,797
5,899
136
Considering that it's negative performance is some titles I can see why maybe it hasn't been a priority. Maybe optimizations and better developer awareness will resolve that in time.

Are there other applications that will make use of this?
 

moinmoin

Diamond Member
Jun 1, 2017
4,933
7,619
136
And as others already said, it works under Linux since long time ago ...
Reminds me of my previous post regarding this...
That's great information! Hope other players will pick up and support that tech as well. Also intriguing that AMD's GPU driver under Linux apparently supports that for some time already and without the limitations SAM under Windows now has. I guess some reviewer should do some benchmarks under Linux testing how different system configurations (Zen 2 vs 3, PCIe 3 vs 4 etc.) affect performance when using ">4GB MMIO".
So did anybody ever do such benchmarks under Linux?
 
  • Like
Reactions: lightmanek

uzzi38

Platinum Member
Oct 16, 2019
2,565
5,572
146
Reminds me of my previous post regarding this...

So did anybody ever do such benchmarks under Linux?
Benchmarks no, but Nemes tried setting it up and said it was a huge pain in the arse just to get working. There is certainly more to it than just a BIOS toggle.

But again, those two instructions don't really change anything in the process.
 

Glo.

Diamond Member
Apr 25, 2015
5,657
4,409
136
Isn't Vulkan not yet supporting this feature, and it makes it impossible, at this point to use this, since DXVK - translation layer from DX calls to Vulkan calls - would have to compensate for this?
 

Glo.

Diamond Member
Apr 25, 2015
5,657
4,409
136

N23, 2350 MHz max boost clock, and it appears that it is at 1.05v.

If this is also at 90W TGP, its pretty incredible efficiency.


Some more details.
 

Glo.

Diamond Member
Apr 25, 2015
5,657
4,409
136
Navi 23 has 8x PCIe 4.0.

You know what this means?

Its supposed to replace RX 5500XT die.

NOW I understand why Nvidia is releasing 299$ 12 GB RTX 3060 with 30 SMs, and 3050 Ti with 6 GB VRAM, with just 28 SMs.

Jesus christ. This thing will be mental in terms of efficiency.
 

sze5003

Lifer
Aug 18, 2012
14,177
622
126
Yes, the only way to get a new navi GPU is to get a laptop lol
A buddy of mine sold his 1080ti before the 3 series came out. He had to buy a whole new desktop just to get a 3080. Returned the first desktop becuause it had a gigabyte 3080, the second computer had an Asus model so he kept that one.

Meanwhile I'm still waiting on that speculated 3080Ti. If true I should have one by the spring time or summer lol.
 

TESKATLIPOKA

Platinum Member
May 1, 2020
2,329
2,811
106
Patrick schur
Navi 22 (Mobile)
NV22 TBD: - 146 W TGP - 192-bit wide memory interface (up to 6 memory chips)
NV22 TBD: - 90 / 110 / 135 W TGP - 160-bit wide memory interface (up to 5 memory chips)
Die Size: 18.31 mm × 18.27 mm ≈ 334.52 mm² Package: 40 mm × 40 mm
For a mobile part pretty high TGP, for a desktop It would be much better. ;)
 
Last edited:

TESKATLIPOKA

Platinum Member
May 1, 2020
2,329
2,811
106
N23, 2350 MHz max boost clock, and it appears that it is at 1.05v.

If this is also at 90W TGP, its pretty incredible efficiency.
It looks like N24 has better efficiency.
N24 has 50W max TGP, so this N23 with 90W max TGP doesn't look so incredible in comparison, If the parameters are these:
N23: 32CU, 128TMU, 64ROPs, 128bit GDDR6, 64MB IC
N24; 24CU, 96TMU, 32-48ROPs, 64-96bit GDDR6, 32-48MB IC

On the other hand, If 90W TGP(not TBP) is enough for 2350Mhz max boost, that would be great compared to older products.
I couldn't find a screenshot of Frequency for RX6800 in this MorePowerTool Igor's edition, so we don't know If 2350Mhz is max turbo or not, but for a mobile part It's very very high.
 
Last edited:

Glo.

Diamond Member
Apr 25, 2015
5,657
4,409
136
It looks like N24 has better efficiency.
N24 has 50W max TGP, so this N23 with 90W max TGP doesn't look so incredible in comparison, If the parameters are these:
N23: 32CU, 128TMU, 64ROPs, 128bit GDDR6, 64MB IC
N24; 24CU, 96TMU, 32-48ROPs, 64-96bit GDDR6, 32-48MB IC

On the other hand, If 90W TGP(not TBP) is enough for 2350Mhz max boost, that would be great compared to older products.
I couldn't find a screenshot of Frequency for RX6800 in this MorePowerTool Igor's edition, so we don't know If 2350Mhz is max turbo or not, but for a mobile part It's very very high.
N24 may have much, much lower clock speeds than Navi 23.
 

TESKATLIPOKA

Platinum Member
May 1, 2020
2,329
2,811
106
I don't see why would be there a big difference in frequency.
Those values are max TGPs for each mobile GPU.
N23 has 65-90W TGP.
N24 has 35-50W TGP and one SKU with only 25W.
I can imagine that those with the lowest TGP will have much lower clocks, but I expect the highest ones to have comparable clockspeed between N23 and N24.
 

Glo.

Diamond Member
Apr 25, 2015
5,657
4,409
136
I don't see why would be there a big difference in frequency.
Those values are max TGPs for each mobile GPU.
N23 has 65-90W TGP.
N24 has 35-50W TGP and one SKU with only 25W.
I can imagine that those with the lowest TGP will have much lower clocks, but I expect the highest ones to have comparable clockspeed between N23 and N24.
I'd say that there might be reshuffle in AMD stack.

N24 may be a competitor to MX450 in terms of power, while delivering way higher performance.

While N23 may be 1:1 Navi 14 replacement.
 

TESKATLIPOKA

Platinum Member
May 1, 2020
2,329
2,811
106
In my opinion 25W N24 SKU should be more than enough against MX450, the rest of N24 SKUs should perform better than GTX 1650Ti, but I won't quess by how much, It all depens on clockspeed.
What is certain is that N23 should butcher Navi14 in performance. Even with the same clocks we have 45% higher TFLOPs(32CU vs 22CU) and Texture Fillrate(128TMU vs 88TMU), Pixel fillrate could be 50-100% better depending If there is 48 or 64 ROPs. Now add higher clocks and we will end up with a beast which should be faster than even RX5600M. :D
BTW RX5600M is only 1/3 faster than RX 5500M in Borderlands 3 as shown here.
 
Last edited:
  • Like
Reactions: Tlh97 and Mopetar

Glo.

Diamond Member
Apr 25, 2015
5,657
4,409
136
Basic maths suggests that if those clocks are correct, for desktop GPU, then Navi 23 SKU, with 128 bit bus, and 32 CUs should be around RTX 2070 Super performance with 120W total board power.

:)
 
  • Like
Reactions: Tlh97 and Mopetar

leoneazzurro

Senior member
Jul 26, 2016
905
1,430
136
BTW RX5600M is only 1/3 faster than RX 5500M in Borderlands 3 as shown here.

And that's mainly due to the very low clocks of the 5600M respect to the 5500M (1,265 GHz vs 1,650 GHz, boost clock vs boost clock). In this case we have 5,829 TFLOPS vs 4.632 TFLOPS (a 26% advantage - so here the 5600M scales more than linearly)-
 

TESKATLIPOKA

Platinum Member
May 1, 2020
2,329
2,811
106
And that's mainly due to the very low clocks of the 5600M respect to the 5500M (1,265 GHz vs 1,650 GHz, boost clock vs boost clock). In this case we have 5,829 TFLOPS vs 4.632 TFLOPS (a 26% advantage - so here the 5600M scales more than linearly)-
I looked at the review once more and I found out Witcher 3 has better performance, they also mention the frequency.
Witcher test:
RX 5600m - 1352MHz(6.23TFLOPs) and 66.1 FPS(146%)
RX 5500m - 1657MHz(4.67TFLOPs) and 45.4 FPS(100%)
Difference in TFLOPs is 34%, but performance increased by 46%.

For comparison:
N23 32CU at 1800-2000Mhz is 7.37-8.19TFLOPs.
N24 24CU at 1800-2000Mhz is 5.53-6.14TFLOPs.
Even N24 could be pretty close to RX 5600M, If the clockspeed will be very high( at least 2GHz) and nothing else will be a mayor bottleneck.

BTW AMD Adrenalin reported 90W power consumption during Witcher for 5600m, probably It's only the GPU without VRAM.
 
Last edited:
  • Like
Reactions: Tlh97 and Mopetar