AMD Radeon VII Linux Benchmarks - Powerful Open-Source Graphics For Compute & Gaming

[ms178]

What is the reason for the increased performance of Radeon 7 (3840 stream processors) over Vega 64 (4096 stream processors)?

Baseline: Vega 64
From the article: Harmonic mean of FPS is increased by (90.901-71.441)/71.441 = 27%
Frequency is increased by (1750-1546)/1546 = 13%
Stream processors are decreased by (3840-4096)/4096 = -6%

The unexplained part is 27-13+6 = 20%. Is this caused just by the greater memory bandwidth, or has the stream processor efficiency been improved as well?

It was stated that it is not just a die shrink but they also fixed a couple of HW bugs along the way. Maybe these account for a meaningful portion of that part.

[mibo]

Good hardware. Good software. Good test.
Thank you all - engineers, developers, and Michael :-)

The high temperature you saw seems to be a new additional reading from the hottest of the 64 temp sensors in the chip. There shoud be one other temperature number provided. At least it is under Windows (I read).

"Orthogonal, but closely related, is improved thermal monitoring, where the number of temperature diodes has been doubled to 64 sensors. As a consequence, AMD is now fully utilizing junction temperature monitoring instead of edge temperature monitoring. Junction temperature measurements were used in Vega 10 (showing up as ‘hotspot’ temperature) but Vega 20 has made the full jump to junction temperature for the full suite of fanspeeds, clockspeeds, and such..."

https://www.anandtech.com/show/13923/the-amd-radeon-vii-review/2

[edwaleni]

No doubt about it, the HBM is making those OpenCL numbers look really good.

[TazKhaelyor]

Good hardware. Good software. Good test.
Thank you all - engineers, developers, and Michael :-)
[...]

Same here, even if it's not really what I'm looking for, I'm glad I was wrong about this card's performance ! I just got more excited about Navi !

[tildearrow]

Typos:

One of the areas where the Radeon GPUs have been lagging behind for a while in competitiveness has been in the performnace-per-Watt area with Polaris and Vega,

>On the NVIDIA side,

If the card really hit 109°C, I am surprised how it didn't affect/destroy the 7nm chip with ease. I've heard that the smaller the transistor size, the lower the peak temperature can be...

[dungeon]

The slightly less competitive Vulkan performance may be due using RADV as the more common but less official AMD Vulkan Linux driver.

He, he so we have one of two the more common but less official and second first which is less common but the more official

Joke a side, what to say other than we finally get to that 7nm For the price of 7 again

[Termy]

i first saw some figures on a windows-benchmark and was slightly "disappointed", but seeing these linux-numbers is a really positive surprise...go AMD!

[riklaunim]

If the card really hit 109°C, I am surprised how it didn't affect/destroy the 7nm chip with ease.

It's from the on-chip temperature sensors. The usual temperature from the edge of the die is lower. The chip starts to throttle when the hotspot exceed 110C and in JayztwoCents review it reached up to 114C.

In general it doesn't really look good right now. The FP64 increase is good but for gaming it won't be a great value, not until prices go down if at all. Now all hopes will be in Navi - just as they were in Polaris or Vega. They really have to step up their game if they want to make great GPUs.

[IvantheDugtrio]

Wonder if overclocking/undercooking has been tried in Linux. Would be nice to have a tool like Wattman in Linux.

[Brisse]

Wonder if overclocking/undercooking has been tried in Linux. Would be nice to have a tool like Wattman in Linux.

This isn't Linux-related, but Sweclockers tested undervolting and it saved them a whopping 60W under load while bringing energy efficiency right in line with Turing and Pascal.