[GPU.RU]The Bureau XCOM Declassified GPU Bench

[Jaydip]

http://gamegpu.ru/action-/-fps-/-tps/the-bureau-xcom-declassified-test-gpu.html

Couple of interesting points

1.780 is almost as fast as 690

2.79XX series have a very strong showing here, even the recently launched 770 can't match a 7970 GHz.Also remember this is a TWIMTBP title.

3.This game is very sensitive to memory bandwidth.Even a 580 is faster than 670 and 660Ti edit: actually I was mistaken, 670 has ~ same memory bandwidth of a 580.This is really weird.

 

[Tempered81]

In these newer games, the 256bit cards get slower & slower

 

[Sable]

Errr. Second chart. 680 has same min FPS and higher average FPS than the 7950 yet is below it on the chart. Wat?!

Serious spanking of the 670/680 though.

edit:

680 barely faster than a 7870?! Unpossible!

edit2:

Can anyone tell if Physx was on or off for the nvidia runs? My russian is a little rusty.

 

[Jaydip]

@ Sable, yeah 680 should be above 7950

 

[RussianSensation]

DX9 game based on UE3 that gets 33 fps at 1600P on a 780? Who coded this game, monkeys? The level of optimization is very poor.

@ Sable, yeah 680 should be above 7950

They sort it by minimums - 680 is slower. That's an 800mhz 7950 they keep using too. Lame.

3.This game is very sensitive to memory bandwidth.Even a 580 is faster than 670 and 660Ti

There is more to it.

GTX480 has 177.4GB/sec or 17% more than 570 but they have identical performance at 1080P.
GTX580 has only 8% more memory bandwidth vs. the 480 but is 21% faster.
GTX670 and 680 have identical memory bandwidth but the 680 is 15% faster.
HD7870 only has 154GB/sec bandwidth but is very close in performance to the 192GB/sec 580.

Something else is killing performance in this game and but GTX780/Titan perform so much better than everything else.

 

[tviceman]

The game does some GPU-accelerated effects too. Reviews of the game, though, are saying it's not particularly good.

 

[Sable]

They sort it by minimums - 680 is slower. That's an 800mhz 7950 they keep using too. Lame.

I'm looking at the second graph, the minimums are the same. So surely you'd then look to the average to rank the card on the chart.
I'm not complaining, the 680 is getting seriously pounded, I just thought it was a bit illogical.

 

[SlowSpyder]

Another new game where the 5870 is faster than the GTX4xx cards. Impressive.

 

[RussianSensation]

I'm not complaining, the 680 is getting seriously pounded, I just thought it was a bit illogical.

I got ya. I thought you guys were talking about the 1st graph. I wouldn't take this review that seriously. The game has horrible optimization. Let's wait for some patches, new drivers and other sites to do the review. When you can't even get 35 fps in a DX9 game at 1600P on a 780, and your game doesn't blow every DX9 game out there out of the water (Witcher 2), you have a problem.

 

[wilds]

I'm surprised at how poorly the GTX 650 performs. Its a pretty capable card...

 

[Subyman]

I got ya. I thought you guys were talking about the 1st graph. I wouldn't take this review that seriously. The game has horrible optimization. Let's wait for some patches, new drivers and other sites to do the review. When you can't even get 35 fps in a DX9 game at 1600P on a 780, and your game doesn't blow every DX9 game out there out of the water (Witcher 2), you have a problem.

XCOM has DX11 with full tessellation and such. There are two versions that can be launched. I'm not sure what version they tested, I assume the DX11 version since they say those tests were done at "max" settings.

I'm not sure why people are saying its a terribly optimized game when all they did in this test was set everything to max. If the game uses FSAA, then that would be a big problem especially since the settings are qualitatively represented by "high", "medium", "low." If high is 8x FSAA, there's your problem. Max settings also brings with it "high" tessellation and full Physx.

Perhaps tweaking the settings brings the performance way up, we don't know, they didn't test tweaking. They set it to max and hit play.

 

[Red Hawk]

I got ya. I thought you guys were talking about the 1st graph. I wouldn't take this review that seriously. The game has horrible optimization. Let's wait for some patches, new drivers and other sites to do the review. When you can't even get 35 fps in a DX9 game at 1600P on a 780, and your game doesn't blow every DX9 game out there out of the water (Witcher 2), you have a problem.

The Bureau is a DirectX 11 game. This Joystiq review references that and the tessellation feature at the bottom (recommending that it be turned off to help performance).

That said, yowza. This is a TWIMTBP game? Then why the heck is a 680 sitting down with the 7950 and 7870? Has the TWIMTBP department at Nvidia just forgotten about the existence of the 600 series now that the 700 series has been released? It reminds me of the situation when Tomb Raider first came out, except that was a Gaming Evolved title where Nvidia was able to make up for a lot of performance post-release through driver updates and game patches. If the game is TWIMTBP I would assume (or hope) that the biggest optimizations have been done leading up to the release.

I have no idea if the graphics of the game justify the performance. My normal sources for graphics analysis, Eurogamer and HardOCP, have yet to post articles on the matter.

In these newer games, the 256bit cards get slower & slower

 

[KompuKare]

I'm looking at the second graph, the minimums are the same. So surely you'd then look to the average to rank the card on the chart.
I'm not complaining, the 680 is getting seriously pounded, I just thought it was a bit illogical.

I'd say there's a decimal place we're not seeing here. Look at 6970/7850/5870 on the first graph. Sorting my mins and having a decimal point we can't see is an explanation which fits.

 

[SlowSpyder]

I got ya. I thought you guys were talking about the 1st graph. I wouldn't take this review that seriously. The game has horrible optimization. Let's wait for some patches, new drivers and other sites to do the review. When you can't even get 35 fps in a DX9 game at 1600P on a 780, and your game doesn't blow every DX9 game out there out of the water (Witcher 2), you have a problem.

The performance for how much video memory is being used really surprised me. Usually you see more than ~1GB used for a game that taxes the GPU so much. Like you said, I'm not sure how much thought I'd put into these benches just yet.

 

[SirPauly]

Nice to see another title with PhysX!

Here is an interview with the developer:

http://www.pcgameshardware.de/The-B...-Declassified-GPU-Physx-PC-Interview-1082482/

We have included a selection of PhysX and APEX features, including APEX Turbulence , GPU particles and breakable materials (for which both APEX and have PhysX Cloth used for various areas of application).

 

[Grooveriding]

I wouldn't worry about performance in this game, 20 fps or 100. It's absolutely terrible no matter what framerate it's running at

 

[Carfax83]

The comparatively lower scores for NVidia can be explained by the game's hardware accelerated PhysX option being turned on presumably.

And yeah, from the reviews, this game isn't worth much attention at all.. A shame really, as the concept was pretty interesting.

 

[Red Hawk]

The comparatively lower scores for NVidia can be explained by the game's hardware accelerated PhysX option being turned on presumably.

And yeah, from the reviews, this game isn't worth much attention at all.. A shame really, as the concept was pretty interesting.

I don't think that is the case. If it were so, I would expect the Titan and 780 to have a similar drop in performance, but also for AMD cards to go through the floor if PhysX was on.

 

[BallaTheFeared]

I don't think that is the case. If it were so, I would expect the Titan and 780 to have a similar drop in performance, but also for AMD cards to go through the floor if PhysX was on.

Why?

AMD would be running cpu only PhysX, even if you managed to lock in High on the CPU AMD would get like 10 fps, which clearly isn't the case.

 

[Carfax83]

I don't think that is the case. If it were so, I would expect the Titan and 780 to have a similar drop in performance, but also for AMD cards to go through the floor if PhysX was on.

The Titan and the 780 have higher core counts (which PhysX relies on), and so are able to absorb the impact of running PhysX much better than the 680/770.

As for why the Radeons didn't have a catastrophic performance hit, PhysX must have been disabled to make it fair...

 

[toyota]

wow the game looks like outdated ass so its hard to comprehend it being that demanding. Dishonored looks better to me and its pegged at the games 130 fps cap the whole time on max settings with my modest gtx660ti. and even at the capped 130 fps its not even remotely close to pushing my gpu hard.

 

[Red Hawk]

Why?

AMD would be running cpu only PhysX, even if you managed to lock in High on the CPU AMD would get like 10 fps, which clearly isn't the case.

Um, exactly? It's clearly not the case; if AMD cards are not PhysX at GPU-accelerated levels, why would the Nvidia cards on the same chart be?

The Titan and the 780 have higher core counts (which PhysX relies on), and so are able to absorb the impact of running PhysX much better than the 680/770.

As for why the Radeons didn't have a catastrophic performance hit, PhysX must have been disabled to make it fair...

Except that wouldn't make it fair, because the Nvidia cards would be running an extra taxing effect while the AMD cards don't. It's not logical to assume that the cards on the chart were running effectively different tests.

 

[amenx]

New game not optimized for Kepler based cards. The 580 beating a 670 is an indication of that. As such, useless bench (for the moment). In another month with new drivers I think the results will be different.

 

[BallaTheFeared]

Um, exactly? It's clearly not the case; if AMD cards are not PhysX at GPU-accelerated levels, why would the Nvidia cards on the same chart be?

Reviewer fail?

 

[Red Hawk]

Reviewer fail?

And we should assume reviewer fail rather than accepting that the benchmark was done with all cards at the same settings as is usual because...?