AMD: Navi Speculation, Rumours and Discussion [2019-2020]
- Thread starter Kaotik
- Start date
- Status
Bondrewd
Veteran
Vega marketing was very detached from the harsh reality.
I remember looking at a benchmark where Vega 10 was compared to Fiji at similar clocks and b/w and was loosing.
https://www.computerbase.de/2017-08...amm-anno-2205-rx-vega-64-vs-vega-56-vs-fury-x
https://www.computerbase.de/2018-05/amd-radeon-rx-vega-untersucht/2/
Was faster than Fiji same clocks & BW at launch and more so a year after, probably there's cases where it was slower too, but not on average. They do mention Vega would be similar to Polaris based on Polaris being faster than Tonga and Vega being faster than Fiji, though.
This is the best example of quality marketing of Raja's era.
* It's not a mere GCN CU, now it's a new Next-Gen CU!
* Higher IPC, higher frequency, nice!
* Shiny graphics, really simple. In one clock, there are twice that many ops! Twice the IPC!
Reality was a bit different.
We got full-speed 16b packet math and some frequency boost. The IPC gain was really super impressive - about mighty 5% compared to Fiji. I repeat, compared to Fiji, not Polaris.
There were another tests comparing Fiji vs Vega, some of those got negative IPC, some of them got 10% uplift.
Since this refers to rapid packed math, one could argue, that the number of instructions are the same, just not the number of data/threads.
This is my interpretation as well. When they said compared to last gen, I took it to mean the X1 architecture not last-gen RDNA. I think the biggest gains in RDNA2 are in the performance/watt metric. These are the gains that are allowing both MS and Sony to hit console power budgets at 10+ Tflops.
It should bode well for consumer GPU's as I thought the initial Navi products were a little too power hungry.
Bondrewd
Veteran
If only you could believe how well N22/23 duo and Smartshift V2 are received along laptop OEMs.
:^)
Hopefully not.
All those +50% efficiency metrics most often come from wider chips working at lower clocks and voltages. In the best case, there are annotations mentioning "iso level of performance", in the worst case there is hype / foolishly optimistic projections and disappointment once final products hit shelves.
Bondrewd
Veteran
Oh no, they're all speed daemons.
And that's why you will never see those +50% efficiency gains in practice.
Real efficiency gains at F max, ISO process and chip config will be much less than advertised 50% at ISO performance and vastly wider chip config.
Bondrewd
Veteran
That's exactly what you're getting.
For RDNA1/Navi10 that +50% perf/watt was if anything under estimated number (it was based on Division 2 @ 1440p Ultra, just like RDNA2's claimed +50% is)
Bondrewd
Veteran
Yeah, same as AMD IPC numbers for Zen iterations.
Looks pretty close to actual according to this: https://tpucdn.com/review/amd-radeon-rx-5700-xt/images/performance-per-watt_2560-1440.png
Well, I would say 58% is quite a bit more than 50% already. Computerbase has 5700 XT offering ~63% better perf/watt than Vega 64 https://www.computerbase.de/2019-07...st/3/#diagramm-performance-pro-watt-2560-1440
Right, and just +40% against Vega 56. Which is what OlegSH is saying.
The claim was never against Vega 56, and we do see in practice those claims on the cards AMD made the claim on and opposed to his post, Navi 10 is thinner and higher clocks than Vega 64
Bondrewd
Veteran
We're comparing fmax part to fmax part here, laddo.
Don't move the goalposts.
You are getting your 50%.
@Bondrewd can you please expand upon your replies and aim for actual replies that progresses the open discussion forward for everyone? As it stands now, the majority of them are seen as meaningless fragments without any factual or source backing. Your posting style is alienating the vast majority of long time users. If you continue on your current path and it becomes a choice between them or you, the choice is obvious.
- Status