AMD: Navi Speculation, Rumours and Discussion [2019-2020]

Status
Not open for further replies.
Ah, yes, I saw that. From your subsequent postings I erroneously assumed, you had done some tests of your own. Sorry.
 
With Zen we always knew what to expect because they told us what to expect, but here we have literally nothing.
AMD told you exactly what to expect from RDNA2, which is RDNA1 + 50% in perf/watt.

Meaning you should expect e.g. 2x Navi 10 performance if you increase power by 33% (1.33 * 1.5 = 2).
On the full Navi 10 you got 225W, so on Big Navi N21 you'd have 300W TBP for twice the performance. 2x Navi 10 would put it about 30% above the 2080 Ti, meaning it'll go against the RTX 3080.


So either AMD lied about power efficiency or Big Navi - if its TBP is set to 300W - will trade blows against the RTX 3080.
 
I don't think they lied, but their estimations at the time can be wrong now I guess. I remember Vega was supposed to bring some ipc gains, and it was very hard to find for exemple. Or the power draw from the 480 were a little too good to be true once in the real world.
Anyway I can't wait, AMD being back in the high end (if they don't screw up) is really a good thing for competition, so, for us.
 
AMD told you exactly what to expect from RDNA2, which is RDNA1 + 50% in perf/watt.

Meaning you should expect e.g. 2x Navi 10 performance if you increase power by 33% (1.33 * 1.5 = 2).
On the full Navi 10 you got 225W, so on Big Navi N21 you'd have 300W TBP for twice the performance. 2x Navi 10 would put it about 30% above the 2080 Ti, meaning it'll go against the RTX 3080.


So either AMD lied about power efficiency or Big Navi - if its TBP is set to 300W - will trade blows against the RTX 3080.
Thats right. But the question is which is the baseline for that 50% perf/watt? Not all the cards based on same arquitecture are the same in perf/watt, moreover according to some people like bondrewd, its not the worst example of rdna1 perf/watt 5700 XT neither one of the best ones like 5600M, RX 5600 XT vs some RDNA2 model...but RDNA1 fmax vs RDNA2 fmax. Aparently, AMD promised the same 50% perf/watt when rdna1 launched, and according to reviews they delivered considering vega64 vs 5700 XT and difference was much higher if you take RX 5700. Cant recall any review talking about fmax back then though...which is vegas fmax?
 
Thats right. But the question is which is the baseline for that 50% perf/watt? Not all the cards based on same arquitecture are the same in perf/watt, moreover according to some people like bondrewd, its not the worst example of rdna1 perf/watt 5700 XT neither one of the best ones like 5600M, RX 5600 XT vs some RDNA2 model...but RDNA1 fmax vs RDNA2 fmax. Aparently, AMD promised the same 50% perf/watt when rdna1 launched, and according to reviews they delivered considering vega64 vs 5700 XT and difference was much higher if you take RX 5700. Cant recall any review talking about fmax back then though...which is vegas fmax?

For the full Navi 10 (5700XT) they used the full Vega 10 (Vega 64) as reference, so for Navi 21 they're most probably using the full Navi 10 SKU as reference.


These comparisons are being made among specific SKUs, not voltages or frequencies or voltage-frequency curves.
I doubt the highest end Navi 21 can reach the power efficiency of the macbook pro's Navi 12, not without underclocking and undervolting at least.
 
I guess it will depend on clocks and voltages, sure, and honestly don´t recall AMD saying it must be at both fmax either.
https://www.techpowerup.com/review/msi-radeon-rx-5700-xt-gaming-x/28.html
Thats 5700 XT vs Vega 64 with 55% better perf/watt at 2160p gaming both at default clocks, and even better at lower res...don´t know about fmax though. And not always highend models have worst perf/watt cause res and quality also plays a role there, look at this good old cypress evergreen 5870 beating even 5770 at 1600p in perf/watt:
https://www.techpowerup.com/review/his-hd-5770/31.html
 
When AMD did those [perf/watt] comparisons to showcase Vega, they simply used TDP numbers. The Footnotes 325 and 362 (which probably is 326, since 362 doesn't exist in the corporate presentation) don't mention the exact methodology either.

which is vegas fmax?
Saw 1752e in some synthies which did not tax power consumption to the max on the 64 LCE.
 
I never put stock into IHV claims of power improvements. It's often at some completely meaningless comparison point. Ampere is a good example.
 
Except RDNA2 also has whatever units they are using for RTRT, which I presume need to be powered too, somehow. The final equation may end up being RDNA x2 x66% (Raster) + X (RTRT) = > 300W in RayTraced games. I wonder is that's why a certain IHV is so concerned with super-precise power measurements all of the sudden....
 
Except RDNA2 also has whatever units they are using for RTRT, which I presume need to be powered too, somehow.

Part of the TMUs, no? And for minor silicon cost, if the Series X HotChip presentation is anything to go by.
 
Well lets look at the xbox systems
You have
XSX 12tflop 52 CUs at 1.825ghz = DF says the 2 week old port was running at the same settings and frame rates as a 2080 using the internal benchmark
XSS has 20CUs at 1.55ghz this will have to run that same game at 1440p

But some how you have a 80 CU navi 2 having trouble hitting 2080ti performance levels on the desktop while running with crazy power requirements ?

Something some where is not adding up at all
 
if its TBP is set to 300W

Depends if they go with 300w gpus then. Most likely they do, in the PC space this isn't so much of a problem.

XSX 12tflop 52 CUs at 1.825ghz = DF says the 2 week old port was running at the same settings and frame rates as a 2080 using the internal benchmark

Already a big feat if RDNA2 can match a 2080 with 12TF, they are neck on neck then with Turing and Ampere, ballpark. All they have to do is release a 80CU version of the XSX GPU, clock it accordingly and they got a perfect 3080 alternative or even beater. RT seems to work as shown in XSX Minecraft and Forza examples. Only DLSS alternative missing, but i assume AMD has something in the works for their pc gpus too in that area.

To match the 3090, they just have to go monsterly big, i have no doubts AMD can do that, but offcourse with a price to match. I think they will/should do this, even if it's a smaller market, it's the company range topping product they can boast with everywhere (we are the fastest or whatever).

AMD gives some heat to NV=lower prices as we see already with RTX3000, but also the push for more performance/features from competing companies. Now Intel left to join the party.
Intel seems to try to recover, their next CPU's seem intresting, they feel the Zen3 heat probably. Nice times like this :p Also AMD probably has to have something SSD IO tech too, as an answer to NV's 14GB/s RTX IO direct to GPU.
 
Isn't FidelityFX a DLSS alternative? And there's also DirectML, which i doubt RDNA2 wouldn't support.
If you're referring to CAS, not really. I believe all it does is sharpen a TAA image. DLSS takes over TAA and tells the game engine to render at a lower resolution, taking motion vectors and its AI reconstruction algorithms to produce the final frame.
 
I don't think they lied, but their estimations at the time can be wrong now I guess. I remember Vega was supposed to bring some ipc gains, and it was very hard to find for exemple. Or the power draw from the 480 were a little too good to be true once in the real world.
Anyway I can't wait, AMD being back in the high end (if they don't screw up) is really a good thing for competition, so, for us.
AMD claimed higher IPC in “Vega NCU”. Most reviewers seemed ended up trying to quantify it through normalised performance of the complete GPU system, in which CUs are just one among many other kinds of cogs.

This time they claimed explicitly 50% performance per watt improvements over RDNA 1, not just a specific metric (IPC) of a specific component (CU).
 
Last edited:
This time they claimed explicitly 50% performance per watt improvements over RDNA 1, not just a specific metric (IPC) of a specific component (CU).

Well looking at the X Box Series X 315W PSU, the power supply is nominally 255W (12V main output) + 60W (others).
So with the caliber rated at about 255W and the actual operating power consumption at about 140W (multiplied by the efficiency of DC-DC (usually about 80%))... gets us to about 120W. That is 33% lower than 5700XT and slightly higher than 5700. The performance then is around 30% higher than 5700XT, coupled with the energy efficiency ratio 66% higher than that of 5700XT which in turn leads to the confidence that AMD claims to increase energy efficiency by 50%.
 
Status
Not open for further replies.
Back
Top