The AMD Radeon VII Review: An Unexpected Shot At The High-End
by Nate Oh on February 7, 2019 9:00 AM ESTRadeon VII & Radeon RX Vega 64 Clock-for-Clock Performance
With the variety of changes from the Vega 10 powered RX Vega 64 to the new Radeon VII and its Vega 20 GPU, we wanted to take a look at performance and compute while controlling for clockspeeds. In this way, we can peek at any substantial improvements or differences in pseudo-IPC. There's a couple caveats here; obviously, because the RX Vega 64 has 64 CUs while the Radeon VII has only 60 CUs, the comparison is already not exact. The other thing is that "IPC" is not the exact metric measured here, but more so how much graphics/compute work is done per clock cycle and how that might translate to performance. Isoclock GPU comparisons tend to be less useful when comparing across generations and architectures, as like in Vega designers often design to add pipeline stages to enable higher clockspeeds, but at the cost of reducing work done per cycle and usually also increasing latency.
For our purposes, the incremental nature of 2nd generation Vega allays some of those concerns, though unfortunately, Wattman was unable to downclock memory at this time, so we couldn't get a set of datapoints for when both cards are configured for comparable memory bandwidth. While the Vega GPU boost mechanics means there's not a static pinned clockspeed, both cards were set to 1500MHz, and both fluctuated from 1490 to 1500MHZ depending on workload. All combined, this means that these results should be taken as approximations and lacking granularity, but are useful in spotting significant increases or decreases. This also means that interpreting the results is trickier, but at a high level, if the Radeon VII outperforms the RX Vega 64 at a given non-memory bound workload, then we can assume meaningful 'work per cycle' enhancements relatively decoupled from CU count.
As mentioned above, we were not able to control for the doubled memory bandwidth. But in terms of gaming, the only unexpected result is with GTA V. As an outlier, it's less likely to be an indication of increased gaming 'work per cycle,' and more likely to be related to driver optimization and memory bandwidth increases. GTA V has historically been a title where AMD hardware don't reach the expected level of performance, so regardless there's been room for driver improvement.
SPECviewperf is a slightly different story, though.
Facebook
Twitter
RSS
289 Comments
View All Comments
Dr. Swag - Thursday, February 7, 2019 - link
If I had to guess, those tests probably are more dependent on memory capacity and/or memory bandwidth.Klimax - Friday, February 8, 2019 - link
Could be still difference between AMD's and Nvidia's OpenCL drivers. Nvidia only fairly recently started to focus on them. (Quite few 2.0 features are still listed as experimental)tipoo - Thursday, February 7, 2019 - link
That they changed the FP64 rate cap entirely in BIOS makes me wonder, should the iMac Pro be updated with something like this (as Navi is supposed to be launching with the mid range first), if it would have the double precision rate cap at all as Apple would be co-writing the drivers and all.tvdang7 - Thursday, February 7, 2019 - link
I feel AT needs to update the game list. I understand that these are probably easier to bench and are demanding but most of us are curious on how it performs on games we actually play. Lets be real how many of you or your friends play these game on the daily? BF1 and MAYBE GTA are popular but not on the grand scheme of things .Manch - Thursday, February 7, 2019 - link
7 DX 111 DX 12
1 Vulcan
Need a better spread of the API's and denote which games are engineered specifically for AMD or Nvidia or neither. I think that would be helpful when deciding which card should be in your rig.
TheinsanegamerN - Thursday, February 7, 2019 - link
Perhaps tell game developers to get with the times then? You cant test what isnt there, and the vast majority of games with repeatable benchmarks are DX11 titles. That is not Anandtech's fault.Manch - Friday, February 8, 2019 - link
Didn't say it was. Merely a suggestion/request. There are around 30 games that are released with DX 12 support and about a dozen with Vulkan. Some of the DX 11 titles tested for this review offer DX 12 & Vulkan supt. They exist and can be tested. If there is a reason to NOT test a DX version or Vulkan version, for example RE2's broken DX12 implementation, OK fair enough. I think it would offer a better picture of how each card performs overall.Manch - Friday, February 8, 2019 - link
DX11 DX12 Vulkan
BF1 Tested Yes No
FC5 Tested No No
AotS Yes Tested Yes
Wolf Yes Yes Tested
FF Tested Maybe? No
GTA Tested No No
SoW Tested No No
F1 Tested No No
TW Tested Yes No
4 of the games tested with DX11 have DX 12 implementations and AotS has a Vulkan implementation. If the implementation is problematic, fair enough. Put a foot note or a ** but there are games with DX 12 and Vulkan out there on current engines so it can be done.
Ryan, perhaps and article on the games, the engines, their API implementations and how/why you choose to use/not use them in testing? Think it would be a good read.
Manch - Friday, February 8, 2019 - link
Sorry bout the format didn't realize it would do that to it.eddman - Friday, February 8, 2019 - link
"about a dozen with Vulkan"What are these dozen games? Last time I checked there were only three or four modern games suitable for vulkan benchmarking: Wolfenstein 2, Doom, Strange Brigade and perhaps AotS.
IMO Wolfenstein 2 is enough to represent vulkan.
"Wolf Yes Yes Tested"
Wolfenstein 2 is vulkan only; no DX12.
As for DX12, yes, I too think they could add more.