Soldato
2560x1440 on my 390X has gone from ~70fps average on ultra to ~85fps average on nightmare.
Not bad at all!
Not bad at all!
-
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
Doom Vulkan vs Open GL performance
- Thread starter Doom112
- Start date
Tried stock RX480 on latest 16.7.2 drivers on Windows 10 with 6700K @ 4.5GHz on my LG 34UM95P @ 3440x1440 in ultra - I tried looking for some way to run a benchmark but couldn't find one.
So I picked a spot to the left of the stairs in the room after getting the shotgun and get rid of all the zombies and focus on the ball of fire.
OpenGL got 42
Vulkan got 59
It was pretty consistent in most of the other areas I made some notes about too - so 40%
Damn 60fps at that resolution of a £200 card, great value.
Numbers aside does it actually feel smoother and more fluid on Vulkan?
I keep reading that on AMD cards it is smoother, especially for frametimes. So add in a FreeSync monitor and you're in for a great time.
I hope my 390X RMA goes through quickly now; and those PowerColor Devil 480s come in soon. Would love to get one of those or the Nitro as a replacement.
The answer is that AMD are obviously seeing far bigger gains, and will do so across almost all DX12 and Vulkan (Mantle) titles.
However, you're going to have to wait for the dust to settle with DOOM.
They haven't enabled a-synch compute on AMD or NVIDIA (I expect it will be default disabled on NVIDIA once it is released), and AFAIK resource heap isn't active yet either.
When those two are added, there should be another big jump for AMD hardware. iD haven't given a timetable for when it'll happen, but I assume relatively soon now that the Vulkan release has been done.
Once stock settles I am gonna do some RX-480 and Freesync monitor bundles, be good to hit £299 for an RX 480 and 22" Freesync 1080P monitor as for most that would be a killer setup that does not destroy the wallet.
Will also look into bigger screens and higher resolution Freesync options in coming months.
Sounds great! I have a friend that still has a GTX 470 as well, would be a great upgrade for them.
Got any hints as to when the Powercolor or Sapphire Nitro arrives? Or is the 22nd still the only date? I feel like I might get my RMA changed from "Replace" to refund to order one of those.
Oh and another thing I forgot to say ... I'll be very interested to see what Fiji & Polaris do relative to the other GCN architectures, once a-synch and resource heap are put in .....
Word is that the latest AMD drivers have converted 4 of every 8 ACEs in Fiji to 2x HWS (Hardware Schedulers), to mimic the setup in Polaris and Vega.
For those that don't know, each shader unit in Fiji has 8 ACEs, but AMD have since said that 2 ACEs = 1 HWS via micro-code update. By default there is no dedicated HWS attached to each shader unit in Fiji (or prior GCN architectures), but Fiji is overstuffed with ACEs.
Now some people are saying that Fiji has had a micro-code update and now has 4x ACEs and 2x HWS in each shader unit ... mimicking Polaris.
In a lot of games, particularly unoptimised ones, this could see very major gains.
Word is that the latest AMD drivers have converted 4 of every 8 ACEs in Fiji to 2x HWS (Hardware Schedulers), to mimic the setup in Polaris and Vega.
For those that don't know, each shader unit in Fiji has 8 ACEs, but AMD have since said that 2 ACEs = 1 HWS via micro-code update. By default there is no dedicated HWS attached to each shader unit in Fiji (or prior GCN architectures), but Fiji is overstuffed with ACEs.
Now some people are saying that Fiji has had a micro-code update and now has 4x ACEs and 2x HWS in each shader unit ... mimicking Polaris.
In a lot of games, particularly unoptimised ones, this could see very major gains.
probably because opengl is bad, would be great if we can have a dx11/vulkan game though, ther than the wraper of talos, but a correct implementation of the api.
dota2 isn't really a taxing game, you would be comparing like what, 300-500fps ? it must be cpu bound in all resolutions
A-sync compute is enabled on amd if you use tsaaa or no aa apparently.
https://twitter.com/idSoftwareTiago/status/752590016988082180
Yes a little but only tried it for a short time - will try more tomorrow night.
i doubt older games would see any gains if thats true, the functions HWS allows, must be coded in the game, like the quick response queue, driver side wont change much imo.
Interesting. Strange that they've done it like that, and not made clear that's what theyr'e doing.
Can anyone with DOOM and recent AMD cards try with TSAA / no AA, and another comparable type of AA (or very low quality AA to replicate no AA) to test with and without a-synch?
Soldato
probably because opengl is bad, would be great if we can have a dx11/vulkan game though, ther than the wraper of talos, but a correct implementation of the api.
dota2 isn't really a taxing game, you would be comparing like what, 300-500fps ? it must be cpu bound in all resolutions
It's not the most taxing game it's true. It's far more popular than DOOM though, so arguably more relevant.
Also, last time I tried OpenGL framerates were not great.
Also, based on ID Software's tweet it seems AMD helpped to implement Vulkan in DOOM, so that may also explain AMD's higher performance gains.
Actually, the games that would stand to benefit the most are old DX11 games with very little specific optimisation (driver or engine side) for AMD's ACEs (or specific layout in Fiji). The HWS would keep the now more managable number of ACEs per unit much, much better fed. Feeding the ACEs and bottlenecks thus are why Fiji has pretty disappointing performance in some titles, relative to its gigantic ACE count.
I'm sure they helped, as obviously they know Mantle inside out, which the majority of Vulkan is pretty much a cut/paste job of, but I think the gratitude shown is mainly for AMD donating their code to Khronos and making Vulkan happen in the first place.
Last edited:
Permabanned
I can confirm - SLI still broken 
Soldato
Vulkan Game feels so much better on Ultra settings now, before at 1440p I would need to drop down to High settings. In Hell on Ultra OpenGL I would get 40s in the tense fire fights now am getting 60/70 fps WOW!
1440p Ultra OpenGL - 68FPS
1440p Ultra Vulkan - Had to use Phone for taking image cant work out how to capture Vulkan and Steam screenshot would crash my PC. 80FPS
Game feels so much better on Ultra settings now, before at 1440p I would need to drop down to High settings. In Hell on Ultra OpenGL I would get 40s in the tense fire fights now am getting 60/70 fps WOW!1440p Ultra OpenGL - 68FPS
1440p Ultra Vulkan - Had to use Phone for taking image cant work out how to capture Vulkan and Steam screenshot would crash my PC. 80FPS
Soldato
The amount of Mantle in the released version of Vulkan is speculation as far as I'm aware. I don't think (m)any are arguing that it originated from Mantle but if the code is 99.99999% the same as Mantle why did it take so long to release?
I don't know if Vulkan is 99.9% mantle, it is very, very similar though. The delay in release? Well, Mantle was still in development when AMD handed it over. Getting it working with other cards and finishing it off would have taken some time.
You're making me regret selling my 290s for 980ti now... Or at least not keeping one of them. Edit Fiji not Hawaii. That's ok then
I did a quick check tonight and it did look like I was getting a boost on my 980ti, though I couldn't comment on lag
Last edited:
Soldato
Nice to see the difference between Async on vs off
Okay, on our first round with the RX 480 and the DOOM Vulkan patch we were informed that SMAA doesn't allow for Async Compute to function, so we set it to the suggested TSSAA, now for the comparison to include Async Compute.
Okay, on our first round with the RX 480 and the DOOM Vulkan patch we were informed that SMAA doesn't allow for Async Compute to function, so we set it to the suggested TSSAA, now for the comparison to include Async Compute.