[Sweclockers]Total War: Atilla benchmarks.

[ShintaiDK]

Assuming that is true, then the only reason I can think of for the horrible stuttering that only the GTX 970 exhibits is that somehow the 0.5GB partition is being utilized despite <3.5GB of frame buffer usage. If anything, that's a far worse scenario because it could be indicative of a future scenario where Nvidia releases a new generation of cards and GTX 970 users are abandoned by the driver team. Do you have any alternative theories on why the GTX 970 is exhibiting unplayable levels of stutter while the 290X and GTX 980 are not?

Or an issue with their irregular benchmarking. Notice the huge difference between no AA and 4xAA.

I bet you I can get the same graph on my GTX980, simply due to a CPU bottleneck in certains area of a map.

 

[HurleyBird]

Or an issue with their irregular benchmarking. Notice the huge difference between no AA and 4xAA.

The drop seems to be consistent between architectures -- very small drop for Maxwell products, moderate drop for Kepler, big drop for GCN. Am I missing something?

I bet you I can get the same graph on my GTX980, simply due to a CPU bottleneck in certains area of a map.

I'm sure they put in a decent amount of effort to make each run as consistent as possible.

 

[ShintaiDK]

Another example, look at sweclockers.

Completely different to computerbase graphs for the same cards. Whats the explanation there? One of them must be wrong.

 

[ShintaiDK]

The drop seems to be consistent between architectures -- very small drop for Maxwell products, moderate drop for Kepler, big drop for GCN. Am I missing something?



I'm sure they put in a decent amount of effort to make each run as consistent as possible.

Does it?

4xAA vs 0xAA

 

[3DVagabond]

Another example, look at sweclockers.

Completely different to computerbase graphs for the same cards. Whats the explanation there? One of them must be wrong.

Comparing benchmarks from different sites doesn't work, I'm afraid.

 

[HurleyBird]

Does it?

Yes. All three architectures appear to behave differently from one another with added MSAA, but in a consistent fashion.

Also, I'm not sure why you posted two separate graphs since you can easily select and show both settings in a single image. You may want to edit your post in this way, as it makes the comparison a lot easier.

 

[ShintaiDK]

Comparing benchmarks from different sites doesn't work, I'm afraid.

But its the same game isnt it. And we talk about frame times.

Same site, same game.

http://www.computerbase.de/2015-02/...agramm-grafikkarten-benchmarks-in-1920-1080_2

And now we are to believe 980/290X is butter sweet while 970 is utter flux? Funny enough the GTX970 was the card with the least stutter in their most VRAM demanding test.

 

[ShintaiDK]

Yes. All three architectures appear to behave differently from one another with added MSAA, but in a consistent fashion.

Also, I'm not sure why you posted two separate graphs since you can easily select and show both settings in a single image. You may want to edit your post in this way, as it makes the comparison a lot easier.

The 960 sure stands out vs 970 and 980. And whats up with the irrgularity of the 290 vs 290X in minimum FPS? maybe computerbase method isnt exactly the best. Specially not when looking on the results of other benchmark sites.

 

[caswow]

But its the same game isnt it. And we talk about frame times.

thats because the test was done with reviewer version. this version had pathfinding bug that caused fps drops.

http://extreme.pcgameshardware.de/n...-im-technik-test-post7210428.html#post7210428

 

[ShintaiDK]

thats because the test was done with reviewer version. this version had pathfinding bug that caused fps drops.

http://extreme.pcgameshardware.de/n...-im-technik-test-post7210428.html#post7210428

The reviewers got the exact same version as you and me.

Also he dont want to post the frame graphs for other cards for some reason. And they run their own benchmark setup that shows irregularities. Not really the best case to make any conclusions.

 

[Silverforce11]

The reviewers got the exact same version as you and me.

Also he dont want to post the frame graphs for other cards for some reason. And they run their own benchmark setup that shows irregularities. Not really the best case to make any conclusions.

I think that was meant to say PCgameshardware.de had used the Steam version recently, with bugs fixed while Sweclockers (they were comparing differences in that thread) used an earlier reviewer's build which had a bug causing stutter.

Thats why they did not see the stuttering for the 980, R290/X when they tested. The reviewer posted specifically he saw gameplay stutters on 3gb cards and less in that post. It just so happens the 970 suffers the same fate.

 

[RussianSensation]

But its the same game isnt it. And we talk about frame times.

http://www.computerbase.de/2015-02/...agramm-grafikkarten-benchmarks-in-1920-1080_2

And now we are to believe 980/290X is butter sweet while 970 is utter flux? Funny enough the GTX970 was the card with the least stutter in their most VRAM demanding test.

The graphs you linked show the opposite. 970 shows the worst stuttering in terms of magnitude in the ms spikes at 1080P under HQ and Extreme tests in the very graphs you linked. Also, you linked data showing 980 having superior ms vs. 290X but that's not surprising since it's the faster card in this game. That's actually consistent.

Jay already showed that 970 bombsin SoM at 1080P. If anyone just linked screenshots of SoM performance they would see 50-80 fps which would ultimately result in excellent averages. However, real world gameplay shows the game completely stutters with freezing which makes it totally unplayable.
http://www.youtube.com/watch?v=v6k55epUBCE

Clearly in Total War both ComputerBase and PCgameshardware show that 970 has much more higher spikes in terms of measured ms compared to a 290X/980. PCGameshardware also shows that 970 has inferior minimums to a 290X/980 as well.

People keep repeating findings based on old data about 970 stating that it trades blows with a 290X but it's not true anymore. That was only true around launch. Today just 1% separates 290X and 970 at 1080p, 290X is 4% faster at 1440P and 7% faster at 4K.

Today the 290X is an overall faster videocard high-end gaming and it happens to have the entire 4GB VRAM and 64 ROPs unlike the 970. Since most sites don't investigate frame times anymore as was the case 2-3 years ago, not enough testing was performed on the 970 around launch reviews to reveal its flaw/stuttering. Now we are starting to see this happen because reviewers/testers are more aware of how the 970 functions. It's not surprising then that 970 will start to show major issues in performance because it needs a special driver from NV to optimize VRAM/ROP/L2 cache for every single title. That's an insane amount of work to do for NV, which means there is no way they are going to be able to optimize for a 970 for every title and for sure once they release newer gen cards, 970's support will go the way of the dinosaurs. We already have NV's Kepler support as an example of how NV prioritizes optimization for modern titles.

 

[ShintaiDK]

You are either colour blind or cannot read graphs. 970 shows the worst stuttering in terms of spikes at 1080P under HQ and Extreme tests in the very graphs you linked.

Did you just ignore the 1440P result?

 

[ShintaiDK]

I think that was meant to say PCgameshardware.de had used the Steam version recently, with bugs fixed while Sweclockers (they were comparing differences in that thread) used an earlier reviewer's build which had a bug causing stutter.

Thats why they did not see the stuttering for the 980, R290/X when they tested. The reviewer posted specifically he saw gameplay stutters on 3gb cards and less in that post. It just so happens the 970 suffers the same fate.

First of all there isnt any special version. Secondly even reviews posted after release shows the same.
http://www.computerbase.de/2015-02/...agramm-grafikkarten-benchmarks-in-1920-1080_2

When one benchmark is out of the ordinary. Specially one that uses an irregular method. Then its time to ignore that one.

 

[Insert_Nickname]

However, real world gameplay shows the game completely stutters with freezing which makes it totally unplayable.

Unplayable? I'm pretty sure I'd have noticed by now, been playing since release @ 1440p/max. settings.

Anyway, as have already been pointed out in this thead Attila/Rome2 can't see/use more then 3GB VRAM anyway.

 

[Silverforce11]

Looks like NV needs an optimized driver for this game, the amount of stutter should be unacceptable.

http://www.computerbase.de/2015-02/...chmark/#diagramm-frametimes-1920-1080-quality

"In 1920 × 1080 and the Quality Preset the AMD Radeon R9 290X shows you the best frametimes."

"Far more varied, it is in the Extreme preset at the same resolution. The activated MSAA does the Radeon card, although more power than the GeForce counterparts, but the frametimes the R9 290X are very good. The two Nvidia graphics cards, however, show a constant up and down."

"In the higher resolution of 2,560 × 1,600 AMD goes again win the match. The GeForce GTX 970 cuts also better than the bigger brother - obviously the drivers are not optimized correctly. Also noticeable but this is not because the title is jerky in all three 3D accelerators due to the low base power.

Unit even at the highest elected by settings the GeForce GTX 970 is not in trouble with the memory. The memory consumption is on the graphics card at 3,430 megabytes, while the GeForce GTX 980 approved 3,434 megabytes. The "Leader" is the Radeon R9 290X with 3,725 megabytes."

Could explain the strange result for the 980, its limited to similar to the 970's 3.5gb cap? Would explain the major frame spikes at 1600p.

 

[seitur]

That sounds pretty disappointing. I'm glad I never picked it up... I had it on my Steam Wishlist for quite some time but then decided it would probably not be worth it after my lousy experience with Shogun 2

People are buying Total War games in droves, so SEGA has no reason to make better games.

 

[xthetenth]

At least Attila's pretty good already, compared to Rome, which wasn't built in a day if you get my drift.

 

[96Firebird]

Could explain the strange result for the 980, its limited to similar to the 970's 3.5gb cap? Would explain the major frame spikes at 1600p.

AMD and Nvidia GPUs typically never use the same amount of VRAM, for whatever reason.

 

[Red Hawk]

AMD and Nvidia GPUs typically never use the same amount of VRAM, for whatever reason.

You know, I'm curious. How do they measure video RAM usage? I'm curious to actually measure and see if I use over 2 GB in Dragon Age Inquisition at 1080p with my 270X.

 

[AtenRa]

You know, I'm curious. How do they measure video RAM usage? I'm curious to actually measure and see if I use over 2 GB in Dragon Age Inquisition at 1080p with my 270X.

For DA:I you can open a console (push the ~ key) and type "Render.DrawScreenInfo" (without the ""). In will display one/two lines of system information at the very top of the screen.

For every other game use the MSI afterburner utility.

 

[96Firebird]

You know, I'm curious. How do they measure video RAM usage? I'm curious to actually measure and see if I use over 2 GB in Dragon Age Inquisition at 1080p with my 270X.

I use Afterburner to monitor my VRAM usage. They probably log the VRAM usage and find the max.

 

[Udgnim]

You know, I'm curious. How do they measure video RAM usage? I'm curious to actually measure and see if I use over 2 GB in Dragon Age Inquisition at 1080p with my 270X.

I use GPUz

http://www.techpowerup.com/gpuz/

 

[Red Hawk]

For DA:I you can open a console (push the ~ key) and type "Render.DrawScreenInfo" (without the ""). In will display one/two lines of system information at the very top of the screen.

For every other game use the MSI afterburner utility.

Ah! I like that you can use a direct method in the game console rather than a utility. There's also a DrawFPS console command to bring up a framerate counter in DAI, and utilities like Fraps and Afterburner have been shown to be inconsistent with it.

Edit: Hmm. Doesn't seem to work. I type in "Render.DrawScreenInfo 1" (nothing happens without the 1, and typing 0 instead dismisses the lines of information). Those two lines come up, giving system information, and there's one segment that says something like "Memory 0/2027". But the 0 never changes.

Edit 2: Ah, well, I was able to make some good findings with Afterburner anyways. To the DAI thread with my report!