Deus EX: Mankind Dividied system specs revealed

[Red Hawk]

TotalBiscuit's graphics options breakdown:

He mentions that he's actually not using the game ready drivers for Deus Ex, since they've been giving him BSODs. So his performance is not necessarily representative of the latest drivers from Nvidia.

 

[Bryf50]

Which makes threads like "1060: faster than 480" extremely misleading to uniformed consumers that come here looking for quick advice. Wonder if Sweepr will change the title?

Beyond that, excellent excellent showing for the Radeon 480 and Geforce 1080, pretty abysmal showing for the 1060.

That's been the title since before the first benchmark came out so I doubt it.

 

[Erenhardt]

This game is killing GPUs
1080p@60 for the elites only:

 

[dogen1]

This game is killing GPUs
1080p@60 for the elites only:

Yeah but very high and high run way faster and look almost identical.

 

[Carfax83]

This game is killing GPUs
1080p@60 for the elites only:

Dat AMD optimization. TPU used the scripted benchmark which gives inflated results to AMD hardware, unlike PCGH.de which used actual gameplay..

Played the game for about 40 minutes just now, and it seems to run smooth. Have all the options enabled and maxed out except for MSAA which is off, and CHS which I have set to on rather than ultra.

Haven't looked at framerate, but the actual gameplay is quite smooth at these settings at 1440p.

 

[escrow4]

The CPU scaling:

Is most impressive. Otherwise this 1070 was detected as "High" with Vysnc and Double Buffering. Upped the preset to Very High with Vsync and MSAA disabled and its around 50-60+ in terms of FPS. RIP old videos cards and old CPUs.

EDIT: This thread should be merged with the one in PC Gaming.

 

[Azix]

Dat AMD optimization. TPU used the scripted benchmark which gives inflated results to AMD hardware, unlike PCGH.de which used actual gameplay..

Played the game for about 40 minutes just now, and it seems to run smooth. Have all the options enabled and maxed out except for MSAA which is off, and CHS which I have set to on rather than ultra.

Haven't looked at framerate, but the actual gameplay is quite smooth at these settings at 1440p.

is it really inflated? I thought all cards lost perf with the benchmark. Looks really demanding with all the lights and objects in the scene. You can tell those lights aren't realtime. but I dont think most of the game will be that demanding. Supposed to be good for a benchmark to be near worst case

 

[Red Hawk]

EDIT: This thread should be merged with the one in PC Gaming.

No, not really. This thread is for purely discussing the technical aspects of the game, not gameplay, and without risk of spoilers and plot discussion. It should stay that way. Though it could be merged with the thread in the CPU section.

 

[escrow4]

Poking around Prague, this time with Textures on Ultra, the rest on the Very High preset, near 7GB(!) vRAM, 60FPS is easily attainable:

 

[RussianSensation]

Bit-Tech has their review.
http://www.bit-tech.net/hardware/graphics/2016/08/23/deus-ex-mankind-divided-benchmarked/3

R9 390 and RX 480 both beat the GTX980. For those who kept saying that GTX970 was as good or better than the RX 480 -- RX 480 is 35% faster under DX11. RX480 also beats its direct GTX1060 competitor by 24%. Looks like the "Keplerization" of GTX950/960 has now reached 970/980 and the only Maxwell card that looks good is the 980Ti (overpriced Titan X).

Looks like the editor of GameGPU was correct as even a 1.9Ghz 1080 cannot maintain 60 fps locked at 1080p with even TAA. At 1440p, things are looking really bad as 980Ti/1070/Fury X only manage 36-38 fps averages. If next gen PC games are this demanding, 1080Ti/Volta is needed asap.
http://www.bit-tech.net/hardware/graphics/2016/08/23/deus-ex-mankind-divided-benchmarked/4

Very poor performance/scaling from Fury/Fury X cards compared to R9 390/RX 480. Perhaps they'll flex their power under DX12?

--------
Old generation GPUs = all RIP. HD7970 can finally be put to pasture but next to it the $700 780Ti looks simply atrocious. Cut down Hawaii in R9 290/390 form is beating 980 but 780Ti is barely ahead of the R9 280X. Mind-blown. Kepler looks like it's really trying to join GeForce 5 and 7 as one the worst aged NV architectures of all time.

GPUs with < 4GB of VRAM are also DOA, which vindicates all those people who recently made fun of the GTX1050 (aka 1060 3GB) and told gamers to stay away from 2GB versions of 960/380.

Whatever magic AMD did with RX 480, the card is on fire trading blows with 64 ROP R9 390X.

 

[SPBHM]

nice that they even tested with medium settings,

but you need a 980 ti/1070 to get 60FPS min at medium 1080!

all and all this looks like the ideal game to use a 30FPS cap unfortunately, or for variable refresh monitors (or tearing/stutters)

670 can't even get 30FPS at medium

 

[eddman]

Whatever magic AMD did with RX 480, the card is on fire trading blows with 64 ROP R9 390X.

I don't know why you are so fixated on ROP counts. The key is to have a balance between shaders, TMUs and ROPs. Jacking the number of one without taking into account the others won't help. This isn't the first time a GPU with less ROPs performs as good or better than a high ROP GPU.

Besides, games are not coded the same. One could rely more on ROPs, another might not.
----------------------------

As for the benchmark numbers; some games run better on Radeons. COD:BO3 is another example.

 

[RussianSensation]

I don't know why you are so fixated on ROP counts. The key is to have a balance between shaders, TMUs and ROPs. Jacking the number of one without taking into account the others won't help. This isn't the first time a GPU with less ROPs performs as good or better than a high ROP GPU.

Besides, games are not coded the same. One could rely more on ROPs, another might not.

Because there is no other card from AMD or NV from any generation that performs as well with only 32 ROPs other than Polaris. That's why this result stands out completely from the rest and historical examples. As the resolution increases, the load on ROPs goes up. Even when switching to 2560x1440 Ultra, 980Ti and GTX1070 only beat RX 480 by 12.5% and 15.6%, respectively. This means Vega 10 may be highly efficient with only 64 ROPs and it might not even need 96.
http://www.bit-tech.net/hardware/graphics/2016/08/23/deus-ex-mankind-divided-benchmarked/4

You state that games have different loads on various aspect of the GPU but that's nothing new. The average load on ROPs increases at 4K which is why AMD doubled the ROPs when moving from Tahiti to Hawaii. I've seen some comments on this forum that as resolution goes up, the load on ROPs barely increases, which is absolutely false. Moving from 2560x1600 to 3840x2160 is 2X the pixel workload for a GPU. And here is the kicker - on average (your metric), the 32 ROP RX 480 keeps up or is very close to 64 ROP R9 390/390X/980 and 48 ROP 1060 at 4K. That should not happen because on average there should be plenty of modern AAA games that bottleneck the 32 ROPs on Polaris 10 enough to ensure its performance would not be that close.

What makes RX 480 32 ROPs a huge deal is that AMD kept ROPs the same as on the R9 380 while NV increased them from 32 to 48 when moving from GTX960 to 1060. This suggests AMD is increasing its real-world ROP efficiency behind the scenes because next gen games are only going to get even more demanding and resolutions will keep going up as more gamers move to 2560x1440, 3440x1440 and 4K. By reducing the required number of ROPs, AMD can allocate the transistors towards more texture and shader units of the GPU. This also suggests AMD and NV have a different idea as to GPU demands for next gen titles. AMD is clearly banking on the fact that games will be even more shader and texture demanding.

 

[Yakk]

This also suggests AMD and NV have a different idea as to GPU demands for next gen titles. AMD is clearly banking on the fact that games will be even more shader and texture demanding.

We can already see AMD has been steering game engine developement through consoles to use more compute intensive effects. Nvidia was steering more towards rasterization based engine until AMD added improved primative discards to their cards. Now we don't really seem to hear much about it anymore from nvidia anymore.

 

[Carfax83]

is it really inflated? I thought all cards lost perf with the benchmark. Looks really demanding with all the lights and objects in the scene. You can tell those lights aren't realtime. but I dont think most of the game will be that demanding. Supposed to be good for a benchmark to be near worst case

I suppose it's a good tool to use as a quick judge of your hardware's capability to run the game. But AMD sponsored titles seem to have a history of doing this. Total Warhammer is the same, in the sense that the scripted benchmark returns much higher numbers for AMD hardware than the actual game itself. IHVs can easily tune their drivers for a specific benchmark in a relatively short order, but to do so for an entire game is a much longer process.

 

[Carfax83]

BTW, if anyone has CHS enabled, take a look at this:

CHS disabled

CHS Ultra enabled

As you can see with CHS enabled, the shadows in the distance are not being rendered, likely to mitigate up the horrible performance loss with it enabled. AMD's CHS is the worst contact hardening soft shadows I've ever seen in a game..

 

[eddman]

...

Maybe AMD went overboard with those 64 ROPs and they ended up being underutilized? If the rest of the pipeline are at their limit, more ROPs can't help.

480 vs. 1060; comparing ROP counts between completely different GPU architectures doesn't tell much.

The best direct comparison I could find is 380X vs. 470. Both have the same amount of shaders, TMUs and ROPs. In bit-tech's DE:MD test, 470 is quite faster, but wouldn't that be mostly because of its higher clockspeed and maybe also higher bandwidth?

Perhaps someone should take these cards, lock them to the same clockspeed and see how much GCN's architecture has improved, clock for clock.

I'm sure GCN 4th is more efficient overall, and its ROPs are probably more efficient too, but not 2 times more efficient.

 

[swilli89]

Maybe AMD went overboard with those 64 ROPs and they ended up being underutilized? If the rest of the pipeline are at their limit, more ROPs can't help.

480 vs. 1060; comparing ROP counts between completely different GPU architectures doesn't tell much.

The best direct comparison I could find is 380X vs. 470. Both have the same amount of shaders, TMUs and ROPs. In bit-tech's DE:MD test, 470 is quite faster, but wouldn't that be mostly because of its higher clockspeed and maybe also higher bandwidth?

Perhaps someone should take these cards, lock them to the same clockspeed and see how much GCN's architecture has improved, clock for clock.

I'm sure GCN 4th is more efficient overall, and its ROPs are probably more efficient too, but not 2 times more efficient.

Here you go:

https://www.computerbase.de/2016-08/amd-radeon-polaris-architektur-performance/2/

Polaris GCN 4.0 is roughly 16% faster than GCN 1.0 Tahiti, and 6% faster than GCN 3.0 Tonga per SM.

 

[2is]

I averaged 37.9 FPS with my sig machine using the in-game benchmarking tool. My actual in-game performance appears to be about 5 fps higher, that said, I have not gotten to that same scene the tool tests. 1440p, Ultra preset with MSAA disabled.

 

[Thinker_145]

lol at the all outrageous conclusions in this thread. It's an AMD sponsored title take it easy.

Talking about 1060 vs 480 the 1060 destroyed it in No Man's Sky so it really depends on the game.

Sent from my HTC One M9

 

[Maverick177]

lol at the all outrageous conclusions in this thread. It's an AMD sponsored title take it easy.

Talking about 1060 vs 480 the 1060 destroyed it in No Man's Sky so it really depends on the game.

Sent from my HTC One M9

Problem is NSM is garbage, and DEMD is a superb game.

 

[Thinker_145]

Problem is NSM is garbage, and DEMD is a superb game.

Irrelevant to the discussion of which card will have an advantage in future.

Sent from my HTC One M9

 

[Maverick177]

Irrelevant to the discussion of which card will have an advantage in future.

Sent from my HTC One M9

The 480 will get a upper hand for sure, and I don't see your point here. You said "it depends on the game". I doubt anyone who will use NMS to benchmark graphics cards.

Awful game? Check
Awful performance? Check
Awful graphics? Check
Players base dropped like flies? Check.

Meanwhile:

Deus Ex Mankind Divided

Great game? Check
Good performance? Check
Good graphics? Somewhat
Players base? Gigantic.

 

[Enigmoid]

Here you go:

https://www.computerbase.de/2016-08/amd-radeon-polaris-architektur-performance/2/

Polaris GCN 4.0 is roughly 16% faster than GCN 1.0 Tahiti, and 6% faster than GCN 3.0 Tonga per SM.

Noting that Tahiti was an incredibly unbalanced design (compare it to Pitcarin or Hawaii) and Tonga had many other problems.

 

[2is]

--------
Old generation GPUs = all RIP. HD7970 can finally be put to pasture but next to it the $700 780Ti looks simply atrocious. Cut down Hawaii in R9 290/390 form is beating 980 but 780Ti is barely ahead of the R9 280X. Mind-blown. Kepler looks like it's really trying to join GeForce 5 and 7 as one the worst aged NV architectures of all time.

GPUs with < 4GB of VRAM are also DOA, which vindicates all those people who recently made fun of the GTX1050 (aka 1060 3GB) and told gamers to stay away from 2GB versions of 960/380.

Whatever magic AMD did with RX 480, the card is on fire trading blows with 64 ROP R9 390X.

I'm not sure if I'd describe it quite that dramatically. If you're trying to do Ultra, then yeah, but you can still get decent performance and respectable IQ with older hardware. My 7970 @ 1080p, high preset is pulling virtually identical performance as my 980Ti @ 1440p, ultra preset. (right around 38 fps) which seems perfectly ok for this style of game, and you're already a few FPS up on what a PS4/XB1 is going to do for you. I still have a 2GB 680 that I'd like to test but that's in a Sandy Bridge i3 box which wouldn't be a very fair comparison.