[benchmarks] The Division - Steam Release

[raghu78]

They tested in tunnel.Like really WTF they are doing?
CB 1080p benchmarks

Just because 980 Ti does not come on top you cannot say the test is invalid. Anyway we will see more results from other sites.

 

[Erenhardt]

They(Gamegpu) tested in tunnel.Like really WTF they are doing?
CB 1080p benchmarks

http://www.computerbase.de/2016-03/...abschnitt_benchmarks_von_full_hd_bis_ultra_hd

I wouldn't read much into it, but their GPU graphs do not match with CPU graphs.
FuryX with 4.5GHz i7 does 64fps not 55. Explains why it doesn't slow down when you go from 1080p to 1440p. It may be just an error on the graphs.

 

[Krteq]

They(Gamegpu) tested in tunnel.Like really WTF they are doing?

Some explanation from GameGPU site

Первоначально планировалось тестировать в игровом бенчмарке, но позже выяснилось что иногда он неадекватно выдаёт итоговый результат, так что от его использования в конечном итоге отказались. Так же по поводу тестового отрезка видеокарт- погода и время суток в игре динамичные, что искажает итоговый результат(добиться идентичных условий практически невозможн&#1086, так что пришлось выбрать место, где окружающая среда не влияет на FPS.

Translation

Well they only benched it underground. Reading further explains the main reason behind doing so. Well the built-in benchmark may not display the appropriate fps avg for an unknown reason. Further more, due to the ever changing weather condition, it isn't to replicate the setting resulting on varied results on different setup

 

[Udgnim]

more benchmarks

http://www.sweclockers.com/test/21839-snabbtest-grafikprestanda-i-tom-clancys-the-division/3#content

 

[raghu78]

I wouldn't read much into it, but their GPU graphs do not match with CPU graphs.
FuryX with 4.5GHz i7 does 64fps not 55. Explains why it doesn't slow down when you go from 1080p to 1440p. It may be just an error on the graphs.

Different settings for 1080p (Ultra), 1440p (High) and 4k(medium) give different results.

Some explanation from GameGPU site

very good reasoning by gamegpu to take out day night cycle effect on benchmark.

 

[Silverforce11]

Basically benchmarks that run default Ultra with HBAO+ and PCSS+ shows GCN tanking a bit, though still doing well. Benchmarks that disable those GameWork features show GCN flying in front.

Looks like AMD will need their "Game Ready" drivers for these features.

http://www.computerbase.de/2016-03/...schnitt_gameworks_mit_komplett_neuen_schatten

 

[crisium]

380X looking like a winner here. Even with sites that are showing an overall more favorable result with the 970, like PCgameshardware, show the 380X demolishing and obliterating the 960, pulling a decent lead over the 280X, and walking awfully close to the 290.

It's always interesting to observe Tonga. Sometimes you see the 280X as decidely the faster card, then you have scenarios such as here.

 

[caswow]

new amd driver tomorrow :thumbsup:

 

[PhonakV30]

hmm? Source?

 

[Game_dev]

970 is still the best mid range card available. Amazing considering how much less power it uses compared to the competition.

 

[crisium]

I'm starting to think it's a bot. How off topic can a thing be?

 

[raghu78]

hmm? Source?

https://twitter.com/CatalystMaker/status/707306465112006658

"Just saw the early release notes for tomorrow's Radeon Software driver update coming. I like what I see!"

- Terry Makedon

 

[maddie]

I'm starting to think it's a bot. How off topic can a thing be?

You think it's an offshoot of Nvidia's deep learning work. If so, we're doomed.

 

[YBS1]

It sure would be nice if these reviewers would use the built in benchmarks when a game provides them rather than their own homebrew, it may or may not be repeatable, no one else can duplicate to compare (or verify) benches.

If they are concerned there may be vendor specific voodoo black magic driver optimizations going on with a prescripted bench, then run a homebrew bench to see if it matches up.

 

[Zodiark1593]

wow, 380x beating 970. hmm, I guess maxwell tanking performance is no longer my own prediction.

In polygon pushing and fill rates, Maxwell stomps anything AMD has to offer, which is why indy games tend to run better on Nvidia as the architecture is simply more forgiving to things like overdraw. GameWorks features also tend to heavily utilize these advantages.

However, AMD's GCN is a very compute focused design by comparison, with 7970 theoreticals exceeding the 970. Achieving peak thoroughput was probably a royal pain for engine coders, though between it's use in consoles, and the flexibility of GCN's compute units, there's real motive to do so, and the results show.

Pascal will most probably be a compute focused design as well, and thus will probably see very substantial gains not even counting the new node.

Nvidia is caught between a rock and a hard place as far as PR goes. They need a new architecture to maintain parity with AMD due to changes being done in how engines are coded. However, to introduce a similar architecture to GCN and obtain very high performance increases over Maxwell may risk alienating their customers again. Over the short term, a moment or two of bad PR is negligible, though as we transition to the long term, Nvidia could be harmed greviously, both by it's own doing (GameWorks, 970 snafu) and factors outside it's control (both console wins).

As much as I don't want an Nvidia monopoly, I also don't want an AMD monopoly either.

 

[Bacon1]

It sure would be nice if these reviewers would use the built in benchmarks when a game provides them rather than their own homebrew, it may or may not be repeatable, no one else can duplicate to compare (or verify) benches.

If they are concerned there may be vendor specific voodoo black magic driver optimizations going on with a prescripted bench, then run a homebrew bench to see if it matches up.

One of the sites was saying that the in game was giving the wrong average #s so they tested it using a tunnel (to prevent weather / lighting differences).

Looks like PCSS causes ~15-20% hit on AMD cards, ~5-10% more than Nvidia. HBAO+ takes little from either from the ultra SSAO.

 

[Erenhardt]

Different settings for 1080p (Ultra), 1440p (High) and 4k(medium) give different results.

That is not what I wanted to say.
If you look at GPU graphs at 1080p and CPU graphs (at 1080p aswell), 980ti perform the same with the same CPU (4.5Ghz i7), while FuryX is a lot faster on the CPU graph.

 

[Silverforce11]

That is not what I wanted to say.
If you look at GPU graphs at 1080p and CPU graphs (at 1080p aswell), 980ti perform the same with the same CPU (4.5Ghz i7), while FuryX is a lot faster on the CPU graph.

Computerbase?

That's because the CPU bench looks like they ran without any GameWorks, if you compare it to their GW specific bench. Fury X gets 63/64 fps in those tests (above the 980Ti). Then in the main charts, it tanks down to 55 fps, they say only HBAO+ is active, so different scenes could have different HBAO+ penalties.

Here's NVIDIA's own testing results:

http://www.geforce.com/whats-new/gu...de#tom-clancys-the-division-ambient-occlusion

Note it differs to the results Computerbase.de has with HBAO+ testing. There is actually a clear perf hit, around 10%.

 

[Flapdrol1337]

Nvidia is caught between a rock and a hard place as far as PR goes. They need a new architecture to maintain parity with AMD due to changes being done in how engines are coded. However, to introduce a similar architecture to GCN and obtain very high performance increases over Maxwell may risk alienating their customers again. Over the short term, a moment or two of bad PR is negligible, though as we transition to the long term, Nvidia could be harmed greviously, both by it's own doing (GameWorks, 970 snafu) and factors outside it's control (both console wins).

Nvidia isn't in a bad position at all, look at the chip size, memory bus, power consumption. A 980 chip is 10% bigger than a 380x, same amount of memory, same bus width, but even the cut 970 sells for ~40% more.

Poor nvidia may have to lower prices a little bit.

 

[Silverforce11]

Nvidia isn't in a bad position at all, look at the chip size, memory bus, power consumption. A 980 chip is 10% bigger than a 380x, same amount of memory, same bus width, but even the cut 970 sells for ~40% more.

Poor nvidia may have to lower prices a little bit.

I don't think it really matters though since next-gen is arriving soon. Current stuff will have to get price drops to clear inventory.

 

[Qwertilot]

That depends how much built up inventory there is of course

Basically NV have been able to sell their cards up a size point in price terms this generation. That's good for their margins on those cards but the consequent lack of real competition at the bottom and top of the market has been worth even more.
(mobile and ultra high end so both very lucrative.).

Has to change if AMD are going to get decent profits out of the GPU market and hopefully the next gen will be much closer. No guarantees of course.

 

[Silverforce11]

Oh, what's an Ubisoft game without this honest to goodness SEAL OF QUALITY!

And some genius developer decided to lock the terminal to one interaction at a time...

We cannot get past the welcome mission. Hitting "F" to activate the laptop does not work. We assumed it was due to all the players in the safe house trying at once so we formed a line.....

 

[caswow]

That depends how much built up inventory there is of course

Basically NV have been able to sell their cards up a size point in price terms this generation. That's good for their margins on those cards but the consequent lack of real competition at the bottom and top of the market has been worth even more.
(mobile and ultra high end so both very lucrative.).

Has to change if AMD are going to get decent profits out of the GPU market and hopefully the next gen will be much closer. No guarantees of course.

we actually dont know what it cost them to make maxwell. amd made every gcn just once while nvidia had to make kepler and design maxwell too. could be lower millions could be >xx millions. you need to break even which amd has certainly long ago and nvidia too ofc but first they had to pour millions in.

 

[csbin]

gamegpu has their results out. 32 sec fraps run of actual gameplay. Fury X is impressive. 15% faster than 980 Ti at 1440p Ultra. would need a 980 Ti OC to match Fury X. 970 matched by 280x at 1440p Ultra. :thumbsup:

http://gamegpu.com/mmorpg-/-onlayn-igry/tom-clancy-s-the-division-test-gpu.html

1080P Fury X faster than 980 Ti

 

[coercitiv]

We cannot get past the welcome mission. Hitting "F" to activate the laptop does not work. We assumed it was due to all the players in the safe house trying at once so we formed a line.....

If true, this is legendary.