Battlefield 1 Benchmarks (Gamegpu & the rest)

[moonbogg]

Nvidia gimping the 980ti to make us hungry for the soon to be released 1080ti?

 

[RussianSensation]

Nvidia gimping the 980ti to make us hungry for the soon to be released 1080ti?

If you read AT's technical overview of Pascal 1070/1080, the fundamental architecture and IPC is mostly equivalent to Maxwell. Remember NV didn't even have Pascal on the roadmap between Maxwell and Volta. This April 2014 article highlights that NV had Maxwell and Volta HBM, but the new roadmap squeezed a stop-gap Pascal. Brilliant move to sell Maxwell+ for 2 more years. Straight from AMD's "copy and paste GCN+" playbook
http://www.gamersnexus.net/guides/1383-nvidia-unified-virtual-memory-roadmap-tegra-future

In fact, all the key building blocks such as shaders, ROPs, TMUs, etc. appear to have identical IPC. The only key advantages are SMT (basically useless right now outside SW2), enlarged L2 cache, dynamic scheduler/pre-emption (this primarily helps under DX12 codepath) and 4th gen memory compression. Other than that, Pascal = Maxwell refresh with high GPU boost. This actually means should Async compute take off and DX12 becomes prevelant in 2018, NV will throw both Maxwell and Pascal under the bus once Volta comes out. That means 1080Ti/2080Ti GP102 has no future-proofing built into it. Paxwell smells like a stop-gap solution between DX11 and DX12. Luckily for NV, almost no DX12 games are Asynchronous heavy, nor are they way better optimized that DX11 codepath for the same games. This is why NV is going to get away with Maxwell and Pascal during they early DX12 era.

That means 980Ti should not be gimped unless it's a user testing error or NV is purposely gimping it in the drivers.

 

[Head1985]

Guru already have new results..Stock throttling 980TI=GTX1070 as i predicted.TITANX is even faster than 1070 at 1440p but at 1080p GTX1070 is faster.
GTX1070 is only 42% faster than GTX970.GTX1080 is 69% faster than GTX980.TITANXP is 55% faster than TITANXM.
It looks like GTX1070 again is worst of all pascal cards.

 

[Erenhardt]

Seen this already?

Holy moly that FX is no slack. It probably was singleplayer, but still, who would have thought\?
Someone should test this game with lower clocked CPUs, like fx8320e or smthn

 

[Udgnim]

It looks like GTX1070 again is worst of all pascal cards.

I don't get this line

isn't the biggest selling point of the GTX 1070 its price / performance?

 

[Head1985]

I don't get this line

isn't the biggest selling point of the GTX 1070 its price / performance?

When i buy GTX1070 it cost me 50% more than GTX970.
When i look at that i get 42% perf gain for +50% price.
GTX1080 users get +70% perf for 20% more price?(from GTX980)
TITANXP users get +55% perf for 20% more price(from TITAXM)

 

[Mondozei]

Here is a new review!

http://www.guru3d.com/articles-pages/battlefield-1-pc-graphics-benchmark-review,1.html

Interesting how well the Fury X does in DX11. Doing slightly better than a Maxwell Titan X at 1440p. Frostbite is not known as AMD-biased IIRC.

 

[dogen1]

Interesting how well the Fury X does in DX11. Doing slightly better than a Maxwell Titan X at 1440p. Frostbite is not known as AMD-biased IIRC.

I think it favored AMD slightly in battlefront.

But yeah the reason dx11 is fine here is that frostbite does a lot of culling on the GPU. So, it's able to solve(at least partially) AMDs problem of low geometry perf starving their CUs. That's what their "optimizing the graphics pipeline with compute" presentation was about.

 

[Bacon1]

Computerbase has some initial testing up:

https://www.computerbase.de/2016-10/battlefield-1-benchmark/3/

 

[Samwell]

Computerbase has some initial testing up:

https://www.computerbase.de/2016-10/battlefield-1-benchmark/3/

One more crappy DX12 port. At least from Dice i expected more. With weak Cpu it can help, but otherwise it's just bad. Frametimes in Multiplayer also terrible.
All Nvidia Gpus loose 15% with DX12 and AMD isn't gaining nearly nothing. Looks as if DX12 path is 10% slower, but AMD can at least use Async to get from the slower code path to ~1-4% gain.

 

[crisium]

Interesting. For AMD cards, DX12 best for single player and DX11 best for multiplayer?

edit: At 1080p. At 1440P multiplayer DX11 and DX12 are identical for AMD.

Looks like DX12 multiplayer best for weak CPUs only.

 

[antihelten]

When i buy GTX1070 it cost me 50% more than GTX970.
When i look at that i get 42% perf gain for +50% price.
GTX1080 users get +70% perf for 20% more price?(from GTX980)
TITANXP users get +55% perf for 20% more price(from TITAXM)

You paid $500 for a 1070 (50% more than the $330 of the 970)?

It's also worth noting that the 970 had vastly better perf/$ compared to the 980 and the Titan X, so the 1070 seeing a smaller jump in perf/$ than the 1080 and Titan XP is in some ways just a normalization of the perf/$ progression.

Interesting. For AMD cards, DX12 best for single player and DX11 best for multiplayer?

edit: At 1080p. At 1440P multiplayer DX11 and DX12 are identical for AMD.

Looks like DX12 multiplayer best for weak CPUs only.

I would probably be a bit careful about saying that DX12 and DX11 are identical for AMD in 1440P multiplayer, since the average FPS might hide some rather iffy looking frametimes:

Could probably use a couple of more patches/driver updates.

 

[antihelten]

dp

 

[Headfoot]

When Mantle came out on BF4 it took a good while for them to iron out the issues. But it ended up working great. DICE has got to be one of, if not the best, graphics engine developer today so I have no doubts they will pull off one of the best DX12 implementations. The "when" is the question. It'll probably be another 6 months from now.

 

[Bacon1]

When Mantle came out on BF4 it took a good while for them to iron out the issues. But it ended up working great. DICE has got to be one of, if not the best, graphics engine developer today so I have no doubts they will pull off one of the best DX12 implementations. The "when" is the question. It'll probably be another 6 months from now.

Yep DICE are the ones who wanted the newer APIs in the first place and helped create Mantle.

 

[Head1985]

You paid $500 for a 1070 (50% more than the $330 of the 970)?

It's also worth noting that the 970 had vastly better perf/$ compared to the 980 and the Titan X, so the 1070 seeing a smaller jump in perf/$ than the 1080 and Titan XP is in some ways just a normalization of the perf/$ progression.

Yep..FE cost 450USD thats alone is 40% more than 330USD GTX970 and aftermarket cards cost even more in EU when i buy my card.SO yeah 50% more than GTX970.Pretty crazy.
Today its slighly better, but its already almost 6months since launch.Today my card cost 13300czk.When i buy one it cost 14000czk.(gainward 1070 GS)
GTX970(gigabyte G1) cost me 9500czk day1.

 

[AtenRa]

Yep..FE cost 450USD thats alone is 40% more than 330USD GTX970 and aftermarket cards cost even more in EU when i buy my card.SO yeah 50% more than GTX970.Pretty crazy.
Today its slighly better, but its already almost 6months since launch.Today my card cost 13300czk.When i buy one it cost 14000czk.(gainward 1070 GS)
GTX970(gigabyte G1) cost me 9500czk day1.

Well, thats what we get with lack of competition in the high-end segment and lack of competition in general.

 

[antihelten]

Yep..FE cost 450USD thats alone is 40% more than 330USD GTX970 and aftermarket cards cost even more in EU when i buy my card.SO yeah 50% more than GTX970.Pretty crazy.
Today its slighly better, but its already almost 6months since launch.Today my card cost 13300czk.When i buy one it cost 14000czk.(gainward 1070 GS)
GTX970(gigabyte G1) cost me 9500czk day1.

To my knowledge most aftermarket cards cost about the same $450 as the FE, which is 36% more, not 40%. There may have been a few extreme cards (like MSI lightning) that were more expensive, but the majority was in the aforementioned range.

Obviously the prices are higher in the EU due to VAT, but so were the prices for the 970, it's not like you could get a 970 for the equivalent of $330 in the EU when that card launched. So the relative price difference more or less stays at the above 36%.

The 1070 might not have been the same improvement in perf/$ over the 970 as the 1080 was over the 980, but given the large gap in perf/$ between the 970 and the 980, the 1070 still remains ahead of the 1080. Even in BF1 which admittedly has a larger gap (26%) between the 1080 and 1070 than usual (20% or so), the 1070 still remains ahead in perf/$ seeing as the 1080 is roughly 50-60% more expensive ($380 vs. $600).

So basically in BF1 the gap in perf/$ between the 1070 and the 1080 is about 25%, compared to 57% for the 970 and 980 (using launch prices). Is that worse? sure, but at the end of the day the 1070 is still ahead of the 1080 perf/$ wise, it just isn't as lopsided as the 970/980 situation.

 

[moonbogg]

When i buy GTX1070 it cost me 50% more than GTX970.
When i look at that i get 42% perf gain for +50% price.
GTX1080 users get +70% perf for 20% more price?(from GTX980)
TITANXP users get +55% perf for 20% more price(from TITAXM)

Nvidia products are the EpiPens of the gaming world. You need their products to survive, and they know it. So they jack up the price like crazy. Their philosophy is "Either pay us, or go die in a corner somewhere."

 

[AtenRa]

Nvidia products are the EpiPens of the gaming world. You need their products to survive, and they know it. So they jack up the price like crazy. Their philosophy is "Either pay us, or go die in a corner somewhere."

To my knowledge most aftermarket cards cost about the same $450 as the FE, which is 36% more, not 40%. There may have been a few extreme cards (like MSI lightning) that were more expensive, but the majority was in the aforementioned range.

Obviously the prices are higher in the EU due to VAT, but so were the prices for the 970, it's not like you could get a 970 for the equivalent of $330 in the EU when that card launched. So the relative price difference more or less stays at the above 36%.

The 1070 might not have been the same improvement in perf/$ over the 970 as the 1080 was over the 980, but given the large gap in perf/$ between the 970 and the 980, the 1070 still remains ahead of the 1080. Even in BF1 which admittedly has a larger gap (26%) between the 1080 and 1070 than usual (20% or so), the 1070 still remains ahead in perf/$ seeing as the 1080 is roughly 50-60% more expensive ($380 vs. $600).

So basically in BF1 the gap in perf/$ between the 1070 and the 1080 is about 25%, compared to 57% for the 970 and 980 (using launch prices). Is that worse? sure, but at the end of the day the 1070 is still ahead of the 1080 perf/$ wise, it just isn't as lopsided as the 970/980 situation.

If im not mistaken he wasnt comparing the GTX 1070 vs GTX1080, he was comparing the GTX 970 to GTX 1070 and GTX 980 to GTX 1080.

He said that going from GTX 970 to GTX1070 is worse than going from GTX980 to GTX 1080.

 

[Head1985]

To my knowledge most aftermarket cards cost about the same $450 as the FE, which is 36% more, not 40%. There may have been a few extreme cards (like MSI lightning) that were more expensive, but the majority was in the aforementioned range.

Obviously the prices are higher in the EU due to VAT, but so were the prices for the 970, it's not like you could get a 970 for the equivalent of $330 in the EU when that card launched.

As i said i buy GTX970 Gigabyte G1 gaming day1 for 9500czk.Gainward GTX1070 GS cost me 14000czk thats 47% more.
FE edition cost here 13800czk so gainward 1070 GS cost more than FE(time when i buy 1070)

If im not mistaken he wasnt comparing the GTX 1070 vs GTX1080, he was comparing the GTX 970 to GTX 1070 and GTX 980 to GTX 1080.

He said that going from GTX 970 to GTX1070 is worse than going from GTX980 to GTX 1080.

Yeah..
70% performance jump in GTX980 vs GTX1080 for 20-25% cost increase.
42% performance jump in GTX970 vs GTX1070 for 50% price increase(my case)

 

[96Firebird]

So you overpaid for the 1070, and are now complaining about it.

Gotcha.

 

[antihelten]

If im not mistaken he wasnt comparing the GTX 1070 vs GTX1080, he was comparing the GTX 970 to GTX 1070 and GTX 980 to GTX 1080.

He said that going from GTX 970 to GTX1070 is worse than going from GTX980 to GTX 1080.

Yes he was comparing the gains of the 1070 vs 970 and the gains of 1080 vs 980, but then he compared those gains to each, and thereby in effect comparing the two cards to each other. The whole discussion is about how the 1070 provides a smaller gain in perf/$ over it's predecessor than the 1080 does, which it definitely does, but my point is simply that this isn't a result of the 1080 having a great price relative to the 1070, but rather a result of the 980 having a horrible price relative to the 970.

As i said i buy GTX970 Gigabyte G1 gaming day1 for 9500czk.Gainward GTX1070 GS cost me 14000czk thats 47% more.
FE edition cost here 13800czk so gainward 1070 GS cost more than FE(time when i buy 1070)

Yeah..
70% performance jump in GTX980 vs GTX1080 for 20-25% cost increase.
42% performance jump in GTX970 vs GTX1070 for 50% price increase(my case)

When the 970 launched the USD/CZK exchange rate was 21.44, when the 1070 launched it was 23.88, a jump of 11.4%. If you account for this then the increase in price is only 32%

So unless you want to blame Nvidia for changes in international currency exchange rates, the real price increase is only 32%.

Also a quick search seems to indicate that the 980 was sold for roughly 15500 CZK at launch and the 1080 for 21400 CZK, so an increase in cost of 38% (24% when accounting for the increase in the exchange rate).

 

[krumme]

When Mantle came out on BF4 it took a good while for them to iron out the issues. But it ended up working great. DICE has got to be one of, if not the best, graphics engine developer today so I have no doubts they will pull off one of the best DX12 implementations. The "when" is the question. It'll probably be another 6 months from now.

Imo its a big disappointment if they need more than a month or two this time to iron out the bugs. Its nearly 3 years ago so they should have learned. As it is i am dissapointed it isnt more clean dx12 game like doom and at least work more polished out the gate with such a giant budget.

Still looks pretty darn cool especially art and surely i will play this stuff instead of this tf2 i am playing now....at least for a while. Lol.

 

[Bacon1]

They've updated their results today:

http://www.gamegpu.com/action-/-fps-/-tps/battlefield-1-test-gpu

Looks to basically just have CFX support, don't think they retested anything else (all others appear to match original tests).