Chart makes no sense. Too many SKUs with CU counts close to each other. If Navi is barely equal to Vega 20 clock for clock and cannot clock much higher then how can a 40 CU Navi 12 equal a 56 CU Vega?
Chart makes no sense. Too many SKUs with CU counts close to each other. If Navi is barely equal to Vega 20 clock for clock and cannot clock much higher then how can a 40 CU Navi 12 equal a 56 CU Vega?
If the above chart is legit, I dont see why NAVI is a fail.
Yes, his Navi leak never made sense, and I don't trust his new video either. Quite interesting how we know absolutely nothing about Navi officially, and the "leaks" aren't leaks, but actually nonsense. The opposite of Vega, which is not a surprise since RTG's marketing people are at Intel now. At least the LLVM commits seem to suggest that significant changes will be made to the CUs for the first time in a while, which is promising.Chart makes no sense. Too many SKUs with CU counts close to each other. If Navi is barely equal to Vega 20 clock for clock and cannot clock much higher then how can a 40 CU Navi 12 equal a 56 CU Vega?
He could still be right regarding power consumption and thermals. Won't be surprising if that is indeed the case.Yes, his Navi leak never made sense, and I don't trust his new video either. Quite interesting how we know absolutely nothing about Navi officially, and the "leaks" aren't leaks, but actually nonsense. The opposite of Vega, which is not a surprise since RTG's marketing people are at Intel now. At least the LLVM commits seem to suggest that significant changes will be made to the CUs for the first time in a while, which is promising.
Why the disbelief surrounding these rumors that Navi could end up being a turd?
Chart makes no sense. Too many SKUs with CU counts close to each other. If Navi is barely equal to Vega 20 clock for clock and cannot clock much higher then how can a 40 CU Navi 12 equal a 56 CU Vega?
The rumoured SKUs might not make any sense, but it could still be true that the cards are hot and power-hungry and do not clock very high.You answered your own question. The disbelief was about the leak being true and not Navi being a turf. AMD won't make 3 dies with so small of a CU difference on expensive 7nm. That's is given UNLESS they have magical performance and can sell so many cards that 3 dies would be worth it over more cutting. Possibility: near 0.
Bullshit (or FUD, whether accudental or intentional, no difference) needs to be plausible.Why the disbelief surrounding these rumors that Navi could end up being a turd?
You wish.He could still be right regarding power consumption and thermals.
You doubt RTG's capacity to underdeliver?Bullshit (or FUD, whether accudental or intentional, no difference) needs to be plausible.
You wish.
#1, RTG doesn't exist anymore.You doubt RTG's capacity to underdeliver?
Chart makes no sense. Too many SKUs with CU counts close to each other. If Navi is barely equal to Vega 20 clock for clock and cannot clock much higher then how can a 40 CU Navi 12 equal a 56 CU Vega?
It supposedly bringing no uArch improvements besides the node, which is both stupid and outright false.what makes NAVI fail in comparison to Polaris/Vega and Turing ??
Whether it's true or not it doesn't matter anymore, at this point the quality of the Navi silicon is independent of the leak info.but it could still be true that the cards are hot and power-hungry and do not clock very high.
Based on a qualification sample's PCI-ID there might be APUs with Navi 10 Lite running at 1.8Ghz.As you can see the Vega replacements would have to clock > 2GHz if they're to achieve their respective performance targets. Either this is true or it would mean that the rumor regarding clock speeds barely hitting their targets must be false.
At this point all I'm waiting for is another leak for Zen 2 that essentially does what this 2nd Navi leak did: cover the other end of the prediction spectrum.
Well, honestly, throwing enough redacted at the wall works.Leak everything to cover all possibilities
It supposedly bringing no uArch improvements besides the node, .
That would probably be a 384b setup, closer to Tahiti than anything, so 12GB memory.Big Navi
56/64 CUs :: 3585/4096 Shaders :: 8GB GDDR 6 :: 200/250 Watts :: $300 / $350
It supposedly bringing no uArch improvements besides the node, which is both stupid and outright false.
View attachment 5898
This is what the theoretical clock speeds should be like based on the rumored performance tiers. As you can see the Vega replacements would have to clock > 2GHz if they're to achieve their respective performance targets. Either this is true or it would mean that the rumor regarding clock speeds barely hitting their targets must be false.
You don't even need brains to look recent LLVM commits.The reason some people believe it is because AMD/RTG has done it before
That's 4tris@clock, the actual setup is 16px/SE, thus 16ROPs@SE tops.Since GCN has hard limits that bottleneck its per-clock performance (4 shader engines, and thus a maximum of 4 pixels/clock, plus the 64-ROP and 64-CU limit)
Chart makes no sense. Too many SKUs with CU counts close to each other. If Navi is barely equal to Vega 20 clock for clock and cannot clock much higher then how can a 40 CU Navi 12 equal a 56 CU Vega?
That would probably be a 384b setup, closer to Tahiti than anything, so 12GB memory.
16/20/36/40 or 16/20/40/44, and nothing about Navi20 because it doesn't exist yet outside of some schizofrenic brains.Alright, what would you correct in the following chart
For one, you never apply the term "IPC" to GPUs, these aren't ILP-centric machines.Although the bit about IPC being worse than Vega seems odd at first
Why are you applying CPUs to GPUs?Intel once built an architecture designed for really high clock speeds that never materialized