Question Speculation: RDNA3 + CDNA2 Architectures Thread

Page 177 - 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

insertcarehere

Senior member
Jan 17, 2013
639
607
136
  • Like
Reactions: scineram

biostud

Lifer
Feb 27, 2003
18,193
4,674
136
With the N32 MIA, and the somewhat lackluster performance (compared to rumors/ expectations) of the N31 is there any doubt that RDNA3 did not turn out the way AMD had hoped? At this point it feels like they will simply skip N32 and let RDNA2 cards fill the lower tiers.
 

biostud

Lifer
Feb 27, 2003
18,193
4,674
136
N32 is definitely a bit of an enigma, but with a 200mm GCD it's gotta be pretty easy to produce in numbers for relatively little cost.

Even if it clocks in at ~6900/6800xt performance AMD should be able to get them out the door.
Maybe performance regression compared to 6950XT?
The 7900XT is not that much faster than the 6950XT.
 

leoneazzurro

Senior member
Jul 26, 2016
905
1,430
136
N33 review (7600S): https://www.notebookcheck.net/Asus-...ook-under-the-auspices-of-the-7.694033.0.html

Unfortunately their game benchmarks is hard to evaluate but it gets 9k in Time Spy Graphics.

This is a single point so it's difficult to judge N33 with only these info, but being 7600S not the full die and with most of tests being CPU/TDP limited I'd say that we'll see something slightly faster than N23 but surely not at N22 level. Edit: which is not exactly the best for the customers but considering the competition it's not too bad as the 4060 parts are not really going to rock either, with the limited specs they are given.
 
Last edited:
  • Like
Reactions: scineram

biostud

Lifer
Feb 27, 2003
18,193
4,674
136
And why then where RDNA2 on sale in US and here (Europe) basically unavailable? I would imagine they would not do a sale and keep a steady supply going if that was the plan.
You might be right, but maybe they simply will keep selling and producing RDNA2 cards at lower prices for a foreseeable future to stay competitive with the Ada stack, until they can sort out the problems of RDNA3 or simply focusing on RDNA3.
 

insertcarehere

Senior member
Jan 17, 2013
639
607
136
This is a single point so it's difficult to judge N33 with only these info, but being 7600S not the full die and with most of tests being CPU/TDP limited I'd say that we'll see something slightly faster than N23 but surely not at N22 level. Edit: which is not exactly the best for the customers but considering the competition it's not too bad as the 4060 parts are not really going to rock either, with the limited specs they are given.

Given AMD are going to market variations of Navi 33 SKUs as successors to 6600/6650m (Navi 23) in laptops its pretty concerning if it turns out that these direct successors aren't clearly faster.

Say what you will about Nvidia but RTX 40xx SKUs on laptop should still be much faster than their direct preceding RTX 30xx SKUs.
 

jpiniero

Lifer
Oct 1, 2010
14,509
5,159
136
And why then where RDNA2 on sale in US and here (Europe) basically unavailable? I would imagine they would not do a sale and keep a steady supply going if that was the plan.

Looked at Newegg... and while the selection isn't great they have several models of N23 and N22. N21 they have a couple but it isn't much. N24 they have a ton of, lol.
 
  • Like
Reactions: scineram

leoneazzurro

Senior member
Jul 26, 2016
905
1,430
136
Given AMD are going to market variations of Navi 33 SKUs as successors to 6600/6650m (Navi 23) in laptops its pretty concerning if it turns out that these direct successors aren't clearly faster.

Say what you will about Nvidia but RTX 40xx SKUs on laptop should still be much faster than their direct preceding RTX 30xx SKUs.

Judging from this first test the 7600 will be clearly faster than the 6600, but not by an huge amount.
Consider that 7600S is not full N33 die but 6800S is full N23 die, clocks and TDP are nominally at the same level and many tests at 1080 are mainly CPU limited when the tested laptop had a slightly slower CPU than the one used on the 6800S laptop. This does not mean that N33 can trounce N23 by 50% or more, but it will for sure be slightly faster and with all probability cheaper, at least when the novelty effect will subside.
And which facts lead you thinking that Nvidia parts (4060) will be much faster? Did you see the specs of these parts? Did you see the first test where the 4060 mobile (full AD107 die) is not much faster than a 3060 mobile?

NVIDIA GeForce RTX 4060 (AD107) Laptop GPU has already been tested - VideoCardz.com

In this test apparently the GPU was configured with a 140W TDP (115W base +25W dynamic boost) and the CPU was faster than the one used on the 3060, which by contrast is limited to a max of 115W plus a max dynamic boost of 15W)
 
Last edited:
  • Like
Reactions: Kaluan and scineram

insertcarehere

Senior member
Jan 17, 2013
639
607
136
Judging from this first test the 7600 will be clearly faster than the 6600, but not by an huge amount.
Consider that 7600S is not full N33 die but 6800S is full N23 die, clocks and TDP are nominally at the same level and many tests at 1080 are mainly CPU limited when the tested laptop had a slightly slower CPU than the one used on the 6800S laptop.

I don't know why you're focusing on comparing with 6800S when the Zephyrus G14 in the comparison has an identical CPU to the tested laptop and is using 6700S (28 CU RDNA 2). The difference between that G14 and the tested laptop could very well be attributable to margin of error and the fact that the 14-inch G14 has tighter thermal constraints.

Note that for larger gaming laptops 7600m SKUs will replace 6600m SKUs wholesale, so its not like N23 was a high-end product there anyway.

And on which facts lead you thinking that Nvidia parts (4060) will be much faster? Did you see the specs of these parts? Did you see the first test where the 4060 mobile (full AD107 die) is not much faster than a 3060 mobile?

NVIDIA GeForce RTX 4060 (AD107) Laptop GPU has already been tested - VideoCardz.com

Uh..your link where the 4060 mobile is on average 30+% faster than 3060 mobile and puts up a good fight against the 3070ti mobile?

1676461948467.png
1676461978338.png
1676462042258.png
1676462118642.png
 

leoneazzurro

Senior member
Jul 26, 2016
905
1,430
136
You missed the part where the 4060 has a higher power available, and a faster CPU, and 3070Ti mobile cards are not that much faster than 3060 cards at the same TDP, maybe? Not even considering that the 3070Ti in that test is paired with a even slower CPU for gaming than the 3060?
Lol. Just looking at the 3060Ti desktop performance (paired with a 5700X) in that very graph should give you a measure of the real value of the 4060 (probably slightly faster than the 3060 12G desktop -which is indeed also present in the graph with again a slower CPU-, and that in scenarios where VRAM is not the limit).
 
Last edited:

insertcarehere

Senior member
Jan 17, 2013
639
607
136
You missed the part where the 4060 has a higher power available, and a faster CPU, and 3070Ti mobile cards are not that much faster than 3060 cards at the same TDP, maybe?
Screenshot_2023-02-15-21-49-58-943_com.sec.android.app.sbrowser.jpg
Some very thorough reviewers have debunked this.
Not even considering that the 3070Ti in that test is paired with a even slower CPU for gaming than the 3060?
Apparently 12800HX is a dog slow cpu that is heavily bottlenecking 3070Ti here but you do you.

Anyways this is an RDNA3 thread and this is what 7600S (28CU RDNA3 on mobile) seems to perform like:
Screenshot_2023-02-15-22-04-31-654_com.sec.android.app.sbrowser.jpg
Love to know how that's clearly faster than RDNA2 equivalents in 6700/6800S.
 

leoneazzurro

Senior member
Jul 26, 2016
905
1,430
136
View attachment 76555
Some very thorough reviewers have debunked this.

In your graph you are comparing graphics score of a 3060 at 140W with a 3070Ti. Where is the graphics score of the 4060 in that graph @140W? Quite lower than a 3070Ti and onty 19% faster than the 3060 mobile. But cherry picking, you know... Also, the 3070Ti in your own graph @140W is lower than 12000 points, while the 3060 is just below 8500. That is around 45% but at lower TDPs (which are often found in the laptop market) the difference shrinks a lot. In any case, the 4060 GS lays exactly in the between 3060 mobile and 3070Ti mobile, that is where I said it is.

Apparently 12800HX is a dog slow cpu that is heavily bottlenecking 3070Ti here but you do you.

It is not a slow dog in absolute terms but it is noticeably slower than the 13900HX in gaming, you can just google around and find by yourself the difference between an Alder lake and a Raptor Lake in this regard. Which, at 1080p where there are way more situations where CPU is a limiting factor, skews the results in the direction of the faster CPU. To see where AD107 stays, you can look easily at the results against the 3060 12G desktop, which is barely slower than the 4060 even with a slower CPU (13400KF vs 13900HX). Trying to compare mobile GPUs on different notebooks is always tricky as there is no platform uniformity: CPU, power management, memory, cooling. Every one of this factors can skew the results in a direction or the other. I.e. just trying to compare the results on a barebone like the one where the 4060 mobile was tested with the GPU on a light& thin gaming like the Zephirus G14 is simply stupid. This is also why AMD when launching the N33 mobile lineup chose to make a comparison with 3060 desktop cards. Those results should be taken with a grain of salt, however that does not mean they are false by default.
Anyways this is an RDNA3 thread and this is what 7600S (28CU RDNA3 on mobile) seems to perform like:
View attachment 76556
Love to know how that's clearly faster than RDNA2 equivalents in 6700/6800S.

First of all, you are again making the mistake of taking results on different platforms as representative of every case out there. 7735HS is basically the equivalent of 6800HS but the 7600S has 10W less TDP available than the 6700S, not even speaking about the 6800S on the other notebook which has a faster processor (which a 1080p still counts) and it is the full N23 die even if with similar TDP. And these differences come before considering the mainboard/power management, RAM, cooling differences that could be there. You can also see that in the individual test in almost every case the 7600S performs like the 6800S while having less CUs and a slower CPU. The comparison with the 3060 is stupid in this case because not only the CPU is quite faster on the Nvidia laptop but it has also a 140W GPU power limit versus 95W in the 7600S case.
So yes, from these data you can deduct that the N33 (which in the 7600XT version will have the full die and up to 120W power limit) will be likely be faster than the similarly configured N23 even if not by miles and even if it could end slower than a 4060 mobile it can for sure compete in both performance and especially price as a 200mm^2 N6 die costs much less than a 147 mm^2 N4.
 
Last edited:
  • Like
Reactions: Tlh97

biostud

Lifer
Feb 27, 2003
18,193
4,674
136
The 7900XTX is roughly 62% faster than the 6800XT in 4k, the N31 die has 60% more CUs than the N32, so if clock speeds are the same as N31, the N32 should be in the range of (6800XT - 6950XT). So if they cannot increase the clock speed significantly or reduce the price, then there is very little reason to release it.
 

leoneazzurro

Senior member
Jul 26, 2016
905
1,430
136
The 7900XTX is roughly 62% faster than the 6800XT in 4k, the N31 die has 60% more CUs than the N32, so if clock speeds are the same as N31, the N32 should be in the range of (6800XT - 6950XT). So if they cannot increase the clock speed significantly or reduce the price, then there is very little reason to release it.

Hmm.. probably the reason is about costs (N21 is 520 mm^2 N7, N32 according to Angstronomics who was correct up to date is 200mm^2 N5 plus 4x37,5mm^2 N6 dies (150mm^2) so a lot less silicon and probably higher yields which probably offset the higher N5 cost. Then there will be probably higher clocks, and there is still the "finewine" unknown factor. Another reason is shifting down a notch the market segmentation with a slightly more scalable product. Last generation we had a large gap between the 6800XT/6900XT/6950XT cards and the 6700XT, which was covered by the 6800 which was a cutdown N21 but a quite large die nevertheless.
Now we have a slighty "better" situation with the N31 at top end with 7900 series, N33 will cover the low end with probably +15% over the N23 and N32 becomes the "new middle" with 7800 and 7700 series. Of course having the same graphics die as the 7800 on the 7700 could seem stupid, but once one factors the N5 costs in and sees what the competition is doing on the same market range, it could be more reasonable. After all, the main die cost come from the N5/N4 dies, and a N32 should be 200mm^2 N5 while AD104 is rumored to be 295mm^2 N4. In this regard, even when adding the cost of the cache/MC dies, N32 should be quite competitive on costs and, even if it comes around the 6800XT/6900XT performance, it could still be OK if priced right.
Note that I am speaking about costs (favoring AMD) and not price, because anyway it's clear that the prices are going to be higher and higher.
 
  • Like
Reactions: Tlh97 and Joe NYC

insertcarehere

Senior member
Jan 17, 2013
639
607
136
In your graph you are comparing graphics score of a 3060 at 140W with a 3070Ti. Where is the graphics score of the 4060 in that graph @140W?
Your argument was that a 3060 and 3070ti mobile aren't much different w/ different TDPs, which that graph disproves. There's an NDA for RTX 4050/60/70 mobile so of course those aren't gonna show up here now.

Trying to compare mobile GPUs on different notebooks is always tricky as there is no platform uniformity: CPU, power management, memory, cooling. Every one of this factors can skew the results in a direction or the other. I.e. just trying to compare the results on a barebone like the one where the 4060 mobile was tested with the GPU on a light& thin gaming like the Zephirus G14 is simply stupid. This is also why AMD when launching the N33 mobile lineup chose to make a comparison with 3060 desktop cards. Those results should be taken with a grain of salt, however that does not mean they are false by default.

Who was trying to compare a 4060 mobile with T&Ls? The main comparison on NBC was between a TUF A16 w/ 7600S and Zephyrus G14 w/ 6700S + Legion S7 w/ 6800HS, all of which are on all-AMD platforms with significant emphasis on portability. The review itself specifically notes that the 7600S fails to distinguish itself from its AMD predecessors.

After the load of nonsense that was Navi 31 performance slides from AMD (54% PPW, 50-70% faster in raster lol), their comparisons stop meaning anything, especially when there are independent reviews out.

First of all, you are again making the mistake of taking results on different platforms as representative of every case out there. 7735HS is basically the equivalent of 6800HS but the 7600S has 10W less TDP available than the 6700S, not even speaking about the 6800S on the other notebook which has a faster processor (which a 1080p still counts) and it is the full N23 die even if with similar TDP. And these differences come before considering the mainboard/power management, RAM, cooling differences that could be there. You can also see that in the individual test in almost every case the 7600S performs like the 6800S while having less CUs and a slower CPU. The comparison with the 3060 is stupid in this case because not only the CPU is quite faster on the Nvidia laptop but it has also a 140W GPU power limit versus 95W in the 7600S case.
So yes, from these data you can deduct that the N33 (which in the 7600XT version will have the full die and up to 120W power limit) will be likely be faster than the similarly configured N23 even if not by miles and even if it could end slower than a 4060 mobile it can for sure compete in both performance and especially price as a 200mm^2 N6 die costs much less than a 147 mm^2 N4.
So you do acknowledge that there are significant flaws in comparing performance between mobile GPUs given platform inconsistencies. Enough so to dismiss on the surface very notable 25+% gains between 4060 mobile and 3060 mobile. But with the 7600S being ~5% faster than 6700S in a (limited) set of games, you're suddenly very confident in concluding that N33 should be clearly faster than equivalent N23... I wonder why that is.
 

leoneazzurro

Senior member
Jul 26, 2016
905
1,430
136
Your argument was that a 3060 and 3070ti mobile aren't much different w/ different TDPs, which that graph disproves. There's an NDA for RTX 4050/60/70 mobile so of course those aren't gonna show up here now.

If you could read, the GS of the 4060 is available in the Videocardz link, as well as the 4060 TDP.

Who was trying to compare a 4060 mobile with T&Ls? The main comparison on NBC was between a TUF A16 w/ 7600S and Zephyrus G14 w/ 6700S + Legion S7 w/ 6800HS, all of which are on all-AMD platforms with significant emphasis on portability. The review itself specifically notes that the 7600S fails to distinguish itself from its AMD predecessors.

After the load of nonsense that was Navi 31 performance slides from AMD (54% PPW, 50-70% faster in raster lol), their comparisons stop meaning anything, especially when there are independent reviews out.

The fact is that ONLY the AMD platforms in the notebookcheck test have limited variability as they use i.e. the same kind of processor even at different bins and thus similar behavior in gaming workloads. Instead, all the other test you want to use to prove the "astonishing gains" of the 4060 have big variations especially in the CPU department that is quite significant at 1080p. And with all these "astonishing gains" the full die 4060 mobile is barely more powerful that a desktop 3060 12 Gbytes when the latter is paired with a sensibly less powerful CPU.

So you do acknowledge that there are significant flaws in comparing performance between mobile GPUs given platform inconsistencies. Enough so to dismiss on the surface very notable 25+% gains between 4060 mobile and 3060 mobile. But with the 7600S being ~5% faster than 6700S in a (limited) set of games, you're suddenly very confident in concluding that N33 should be clearly faster than equivalent N23... I wonder why that is.

The problem is that you are trying to use the platform differences to push the 4060 way further than it can (by ignoring the huge CPU variations, in example), as saw by the 3DMark graphics score (which is almost independent by the CPU used), and you are dismissing the parts that go against your flawed narrative. I.e. you continue to speak about the 7600S being barely faster than the 6700S but you are ignoring the +15W TDP of advantage the 6700S (+16%) has, or the +15% more CUs the 6800S has. Not even speaking about the fact there is a +45W advantage on the 3060 side in the tests reported. I wonder why that is. And when I said faster, I also stated it will not be for sure +50% or even +30% faster. My bet is around +15% clock for clock, watt per watt.
 
  • Like
Reactions: Tlh97

TESKATLIPOKA

Platinum Member
May 1, 2020
2,329
2,811
106
N33 doesn't look that great from a performance perspective, but we still don't know the price. If It's priced lower than RTX 4060 then I am ok with It.
I think performance won't be that much different.

P.S. Adrenalin drivers were 22.40.01.11, which are not the latest.
Adrenalin 23.2.1 improvement was shown only with RDNA2 and not RDNA3.
AMD page
 
Last edited:

beginner99

Diamond Member
Jun 2, 2009
5,208
1,580
136
The 7900XTX is roughly 62% faster than the 6800XT in 4k, the N31 die has 60% more CUs than the N32, so if clock speeds are the same as N31, the N32 should be in the range of (6800XT - 6950XT). So if they cannot increase the clock speed significantly or reduce the price, then there is very little reason to release it.
Lower production cost and sell at same or higher price?
New features especially in terms of encoding? (not sure if both 6000 and 7000 support AV1 for example)
 
  • Like
Reactions: Tlh97 and Joe NYC

biostud

Lifer
Feb 27, 2003
18,193
4,674
136
Lower production cost and sell at same or higher price?
New features especially in terms of encoding? (not sure if both 6000 and 7000 support AV1 for example)

It’s just that if it doesn’t offer a significant perf/price jump then the reason to upgrade is simply not there, even if there is new encoding features.

So I think AMD will make the switch to N32 at some point, they’re simply not in a hurry because it is going to be roughly the same performance as the N21. Or they might as the speculation goes try to do respin of RDNA3. Which I’m less incli to believe.