• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

Hitman DirectX-12 BenchmarksupdateComputerbase

Page 10 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
These are some impressive numbers put up by Hawaii. May have to switch back to AMD next gen, especially with all the Freesync monitors available.

Again, these titles are all early DX12 implementations. Hardly a K.O. for Nvidia cards. Hitman in particular won't be fully released until 2017. RIP Hitman series.
 
Something just doesn't seem right. I know that consoles are GCN based but those GTX 980TI numbers are horrible. I don't understand how it can lose to a 390x. On paper, the GTX 980TI is much stronger. DX12 gave Maxwell negative scaling. Something is off. I think some kind of optimization is needed for Maxwell. This reminds me of Project Car; but the other way around.
 
iiiankiii said:
Something just doesn't seem right. I know that consoles are GCN based but those GTX 980TI numbers are horrible. I don't understand how it can lose to a 390x. On paper, the GTX 980TI is much stronger. DX12 gave Maxwell negative scaling. Something is off. I think some kind of optimization is needed for Maxwell. This reminds me of Project Car; but the other way around.
Click to expand...

going off of Wiki #s

there's a lot more to GPU performance than just flop #s, but viewing FLOPS as a general horsepower #

390X (1050 / 6000) has 5913.6 / 739.2, single / double precision GFLOPs

980Ti (1000 / 1076 / 7010) has 6144 / 336, single / double precision GFLOPs

throw in async compute being used in DX12 and if it can't be disabled, then it's going to degrade 980Ti performance in DX12
 
Udgnim said:
going off of Wiki #s

there's a lot more to GPU performance than just flop #s, but viewing FLOPS as a general horsepower #

390X (1050 / 6000) has 5913.6 / 739.2, single / double precision GFLOPs

980Ti (1000 / 1076 / 7010) has 6144 / 336, single / double precision GFLOPs

throw in async compute being used in DX12 and if it can't be disabled, then it's going to degrade 980Ti performance in DX12
Click to expand...

The only performance degradation with Asynchronous compute turned on would be in terms of the 1-5% cost associated with the fence which GCN more than makes up for due to the performance boost that Asynchronous compute + graphics gives.

As for why a GTX 980 Ti doesn't perform as well as you'd expect it too...well that's due to GPU utilization. Since the GTX 980 Ti doesn't make use of Asynchronous compute + graphics then it cannot make use of untapped compute resources while the GCN cards can. That's not performance degradation at all.

Hitman also does not cause too much API overhead, in DX11, for GCN to begin with so GCN performs as it should. You've likely become accustomed to seeing the GTX 980 Ti dominate under DX11 titles and believe it is due to hardware supremacy on NVIDIA's part when it's really that GCN was CPU bottlenecked due to an API overhead the whole time.
 
airfathaaaaa said:
and no one talks about the elephant on the room (780ti)
http://www.computerbase.de/2016-02/ashes-of-the-singularity-directx-12-amd-nvidia/3/
Click to expand...

Oh neat, it's doing well there. Goes to show how stripped down Maxwell is. It makes sense because it looks like NV was taking out loans against their chips' future to get more of what games needed right then. Kepler might not be a legend like Hawaii, but I'm kind of expecting to see it gain ground back from Maxwell.

I'm actually starting to wonder if Pascal is going to be all that it's cracked up to be in terms of raw power relative to Maxwell because either they're going to have to add all that back in or they're going to have to eat the perfomance hit in current games. I think the former's more likely because NV has done a good job of delivering the right architecture for the market as it is when the card drops, but that's been a one way trend and the games are going another.
 
Last edited:
xthetenth said:
they're going to have to add all that back in or they're going to have to eat the perfomance hit in current games. I think the former's more likely because NV has done a good job of delivering the right architecture for the market as it is when the card drops, but that's been a one way trend and the games are going another.
Click to expand...


They're the ones that drove the give them less for more tact. Now that DX12 is pushing to utilize the full capabilities of the gpu, that less for more tact is biting them in the ass. I suspect they will have to add that all back in or else they will lose on the compute side. And frankly, the less for more aspect is a lose lose for all of us in the longterm. Imagine applying that to other products?
 
3DVagabond said:
With trending performance I wouldn't be recommending any nVidia. Especially at that high of a price point.
Click to expand...

980 Ti is still an excellent card for people with big wallets. Smart buy right now is definitely the 390 8GB, though.

Anyway, I'm getting neither. Waiting for the big guns to drop, then I'm building a new rig.
 
Fiji, pls. I'm glad its still technically faster than Hawaii, but come on. There's no reason to see single digit % difference between Fiji and Hawaii yet we consistently see this.
 
Yeah I do find Fiji somewhat disappointing. Hopefully they can unlock more performance down the road at least. Glad I've been "stuck" on 290s for a while lol...I'm regretting selling the 3rd 290 I had haha 😀
 
You have to worry a little for Fiji long term. It (and even GCN 1.2) are only in a few cards selling small numbers for a fairly short time period and seem prone to be forgotten at release already 🙁

Shame because Fury is definitely > then the 390, but they didn't have the production quantity (or the VRAM I guess) to do a total replacement it when it released.

It is nice to see that the 780ti can do fine if you put a decent effort in 🙂 It has seemingly struggled on some of these things, but no real reason for it to.

Not a negative on the 970 this - mustn't confuse the pricing structure with card size etc.

Rather like Udgnim pointed out a little above. The 780ti and 2/390(x) are both sensibly designed, absolutely top end cards designed on the same node as Maxwell. The 970 has no real right to be keeping up with them half as well as it has.
 
Jaydip said:
Just because in one game it is a bit slower doesn't mean suddenly 980 Ti isn't the top dog, it still is the king of the hill.
Click to expand...

It's actually only king of the hill (in single GPU config) due to it's OC ability vs Fury X.

In the stock form, Fury X has actually got it beat for awhile now, when you factor a lot of games (TPU, Swedeclockers etc).

Add these new games like The Division, Far Cry Primal, Hitman and soon Ashes to the list, Fury X's lead will grow.

As modern games in 2016 replace older games in benchmarks, it will be a clear separation, and OC 980Ti will be needed to maintain parity.

This is the EXACT same scenario as the R290X vs 780Ti in 2014, where the 780Ti had the lead and over time we see the R290X catch up, beat it and now in new games, just so far ahead.
 
Qwertilot said:
You have to worry a little for Fiji long term. It (and even GCN 1.2) are only in a few cards selling small numbers for a fairly short time period and seem prone to be forgotten at release already 🙁
Click to expand...

They've been faster than in the past I have to admit, they release optimized drivers for all the new titles within a few days or a week. RTG is producing tangible results for them.

Especially pcgameshardware.de, they told AMD of the DX12 60fps cap bug and half a day later, they get a beta driver that fixes it. Impressive when compared to the past.
 
Async compute is useful for many things as well. I mean you can have the copy queue streaming in resources while your compute and graphics pipelines are busy with other work so that when those resources (ex textures) are needed, there's no stall/wait to stream/copy them over, as they're ready for use. People who reject Async compute as some AMD conspiracy don't seem to realize just how useful and beneficial the feature is.
 
Wow those latest Fury #'s.

Fury AIR @ stock is same FPS as a 28% OC'd 980 TI (1380mhz)! HOLY COW

Will be interesting to see what Fury Nano and Fury X get. Will show how clocks / shaders matter in this title.
 
You must log in or register to reply here.

TRENDING THREADS

Back
Top