Perf/watt/IHV man hours/posts *split*

[Razor1]

http://radeon.wpengine.netdna-cdn.c...is-Architecture-Whitepaper-Final-08042016.pdf
It's in the Polaris Whitepaper, Page 12. I believe many sites reported the calibration was broken at launch. Seems reasonable to assume a broken feature could be fixed with a revision. If that is what occurred. Also seems reasonable to assume a feature designed to reduce power consumption could reduce power consumption if fixed.

Ok I'm thinking of something else then,

That feature is not broken

Actually is working quite well, this is why you don't have voltage droops when Polaris is pushed, unlike with Fiji. Even with the pci-e slot over volting and drawing too much power droop in core voltage is never seen with Polaris.

Reviewers can't test for this either, they were assuming that was the problem because of the power issues Polaris had.

No, but highly sensitive circuits could be susceptible to the same kinds of interference. Likely not RF in this case, but the same techniques can help with ground isolation. In theory the chip has a mechanism to examine slew rate of various clocks. Likely FINFETs configured with various gains and some logic. I'll admit I'm just guessing as to the cause, but there are only so many explanations for repeated metal spins if that's the case.

Of course its not going to RF that causes the problem, there were a totally of 7 spins, the metal spins would not fix such a problem if there was one, only the mask layer respins can fix such a problem, not to mention, if there was such a problem to begin with the chip probably wouldn't even be functional to say the least.

By a respin no, but we have with software changes. Just go find some benchmarks for power management disabled vs enabled.

Power management isn't disabled, AMD never gave the tools to even look for that, it was assumptions that reviewers were thinking because of Polaris's power problems at launch.

You don't go from a working functional part from carrizo, to a broken part in Polaris (2 polaris variants btw) so you need to figure in those metal respins and mask respins with P11's with P10's. They had many many chances to fix it if it was broken, and that's why its not broken.

 

[Anarchist4000]

Actually is working quite well, this is why you don't have voltage droops when Polaris is pushed, unlike with Fiji. Even with the pci-e slot over volting and drawing too much power droop in core voltage is never seen with Polaris.

In that case nevermind. Rules out any problems, but still possible GF revised the process. Doubtful the results would be that significant, but always possible I guess.

Power management isn't disabled, AMD never gave the tools to even look for that, it was assumptions that reviewers were thinking because of Polaris's power problems at launch.

Haven't followed it that closely, but I thought the calibration was disabled. You asked for examples where there were large power swings though. While not hardware in nature, that situation would provide one. It wouldn't necessarily require all the power features to be disabled either.

 

[CSI PC]

Yeah it is working, just that Nvidia has a jump on AMD due to the team that worked on Maxwell (both architecture and low level silicon-node-circuit design related factors), especially in terms of controlling leakage-thermal relationship/switching activity, speed of operation in terms of dynamic-static power and how well implemented.
From a simplistic view it is incredible how quickly Nvidia Boost technology and power draw-TDP adapts, but that is only one part of the whole, the AMD whitepaper highlights one area where NVidia really has a lead and that is waste energy and the loss between voltage regulator and the GPU (page 13 of the whitepaper), look at the core GPU power demand to TBP as part of it.

All of that has improved further with Pascal as well so it is a challenge for Polaris to catch up (the positivie this is an improvement over previous generations), maybe with Vega they will catch up (I think there is a reasonable chance if those initial figures are correct) but then that looks to be short lived once Volta releases and so the chase continues in terms of overall efficiency-thermals and the power wall.

Cheers