So that was unique to that specific engine and cannot be replicated by others. Bf1 is an interesting candidate for the next one, for some reason the 6900 is a beast but ryzen is at i5 lv
Enviado desde mi HTC One mediante Tapatalk
https://fgiesen.wordpress.com/2013/01/29/write-combining-is-not-your-friend/
All games do write combined writes when writing to mapped GPU buffers (such as constant buffers). Most other applications never do write combined writes. Getting write combining perfect is comparable to black magic. Sometimes you need to fight the compiler. Thankfully it's much easier with modern OoO x86 CPUs compared to last gen PPC cores. OoO window is usually big enough to fill the write combine buffers completely before they are written. I have no idea how for example HT/SMT interacts with write combining. I can easily imagine that small scheduling differences can lead to potentially very big problems.
I nowadays never directly write to write combined memory. I prepare an aligned struct that I memcpy at once. This is a safe way to avoid problems.
Last edited:
itsmydamnation
Veteran
This:
In many review its on pair with i5 4690 / 6600.
And now go play BF1 how people actually play BF1.
Then add ~30 chrome tabs behind it, mumble/discord/teamspeak/skype , steam/origin , av and all the other random crap people have open ( eg. i have my guitar amp modeling software and vst's open like 24/7).
Some one did that with a ryzen and i think a 6700 ( maybe 7700) on anandtech and it wasn't pretty for the quad core. i cant find the post and i need to run off to work. hopefully i can find it in the evenings.
The other thing is memory, people are starting to figure it out , support is getting better and 3200+ really does make a big difference to ryzen ( prob more from the reduced fabric latency then anything).
It makes a huge difference. The single player doesn't stress the cpu much at all compared to the multiplayer. The consistent frametimes on Ryzen make a much smoother overall gaming experience that a simple average fps on a single player built-in benchmark just cannot show.
Computerbase.de have full MP tests for BF1 and other MP games with excellent graph systems to isolate specific games as desired.
I think you missed the point... IT was not about which CPU is better is was about what improvement could ryzen get there with a software update.
What's left if you don't use the integrated benchmark of Ashes Of The Singularity but test ingame:
vs. the integraded benchmark:
They run this stuff four times and have two different Ryzen systems (but with the same Mainboard model) and the results are practically the same.
Today they will update the results with RX 480 numbers but they already said that they aren't better.
In addition they want to upload the save-file so everyone who has this game can jump in and compare.
Source (German):
http://www.pcgameshardware.de/Ryzen...ecials/AMD-AotS-Patch-Test-Benchmark-1224503/
vs. the integraded benchmark:
They run this stuff four times and have two different Ryzen systems (but with the same Mainboard model) and the results are practically the same.
Today they will update the results with RX 480 numbers but they already said that they aren't better.
In addition they want to upload the save-file so everyone who has this game can jump in and compare.
Source (German):
http://www.pcgameshardware.de/Ryzen...ecials/AMD-AotS-Patch-Test-Benchmark-1224503/
DavidGraham
Veteran
Unfortunately, for consistency sake, I think CB only tests multiplayer on empty maps, So there is little useful data here.
Another data point worth pondering is the disparity between DX11 and DX12 scores: The minimal difference between the 7700K and the 1800X in DX11 (1800X is 5% faster). And the huge difference between them in DX12 (7700K is 35% faster).
CB testing shows all AMD CPUs taking a massive hit under DX12, here the API appears to be bound by single threaded performance as the 7700K climbs it's way to the top, while In DX11 core count seems to be the determining factor (weird I know!). This could be a direct consequence to them testing empty maps. As DX12 doesn't really exploit it's multi-threading capabilities when not much is going on on the screen. For comparison sake, PCGH managed to show that DX12 is heavily CPU dependent in Battlefield 1(scales more with CPU cores than DX11).
While there wasn't the RX 480 update yet, PCGH included Broadwell-E numbers, the min FPS seem to fall apart when using CPUs with many cores:
Source:
http://www.pcgameshardware.de/Ryzen...ecials/AMD-AotS-Patch-Test-Benchmark-1224503/
Source:
http://www.pcgameshardware.de/Ryzen...ecials/AMD-AotS-Patch-Test-Benchmark-1224503/
Well that sucks, not really an MP test then lol. It's what I could find in short notice but unfortunately for obvious reasons reviewers don't really benchmark MP much.
edit: It's interesting though that it's obviously a huge difference between the SP built-in bm and what they're doing.
I wonder if these two things are related?
Last edited:
3d Mark API overhead test paints a complete different picture of Ryzen draw call performance:
http://www.tomshardware.com/reviews/amd-ryzen-7-1700x-review,4987-3.html
Ryzen has a 16% lead over 7700K in the DX12 draw call test. It also beats 7700K in DX11 multithreaded test (24%), but lags significantly behind in DX11 single threaded test (-45%). So there doesn't seem to be a generic Ryzen DX12 performance issue. It must be something engine/game specific.
DX11 single threaded draw call performance is pretty bad and that's alarming, since most DX11 games use single render thread (deferred contexts are usually a bad idea). It would be interesting to know what's the cause. As AoTS had bad perf issues with non-temporal writes, I would guess that write combining in general could be an issue for Ryzen. All games write their constant buffers to write combined memory (update object matrices etc for each draw call). It could be that Ryzen is more picky than Intel in the way write combining must be used. SMT and write combining is a interesting combination to get right.
Well it was what I saw in many reviews.
Similar threads
