NVIDIA Kepler speculation thread

[Wynix]

I think the real value buy is the 4GB models, EVGA 4GB reference 760 is $280 on newegg.
256bit bus, 6hz memory, 7950 like performance, I'm being drawn to the green side, but I.. must.. hold out till 8870 release.

 

[Psycho]

Eh? We test in a closed case. A Thermaltake Spedo, to be precise.

Sorry, somehow failed to see that in the first place (but good to see that more are testing in realistic cases now the environment matters).

But it's still a remarkable ~100mhz difference on average, which can easily make or break the verdict - would be nice if more reviews were putting up those numbers.

 

[Kaotik]

It's not a theory it's a fact. BF3 and FC3 are contributing huge wins for Nvidia in this case (BF3 always does) - margins of victory that aren't generally being seen elsewhere.

Actually BF3 doesn't do that always, it all depends on the map reviewers wants to use (or at least that's the only sane explanation I can figure out why some reviews show so different results than some others on the same game, drivers and cards with similar rigs).

Couple quite different results as example, first off Muropaketti

Seems about right considering the cards average performance?

Same goes for TPU:

But then let's go Anand:

And it's suddenly quite a bit different

 

[jimbo75]

You're right Kaotik, BF3 can throw up really different results depending on map.

This can be the case legitimately or not. I'm not saying that the results on AT or anywhere else are false or fabricated, but they can be cherry picked. Sometimes you end up with a situation where you aren't getting the true story of how the game plays generally on a specific card.

BF3 is overperforming in one map for Nvidia, and it's the map of choice for AT and a few others. The logic behind it is reasonable (it's an easier map to benchmark), but that really doesn't matter when it's giving a false impression of the overall game performance.

I'm pretty sure Ryan is getting paid enough by Anand to get a representative benchmark of BF3 out. Yet over and over again, AT remains an outlier for this and other games, and continues to be an Nvidia outlier with every new card. Is it really so hard to benchmark the leading last gen title on every scene and get a general idea of performance?

For the last year Nvidia has had a 40%-20% lead in BF3 on Anandtech. It's just bullshit.

 

[Wynix]

You're right Kaotik, BF3 can throw up really different results depending on map.

This can be the case legitimately or not. I'm not saying that the results on AT or anywhere else are false or fabricated, but they can be cherry picked. Sometimes you end up with a situation where you aren't getting the true story of how the game plays generally on a specific card.

BF3 is overperforming in one map for Nvidia, and it's the map of choice for AT and a few others. The logic behind it is reasonable (it's an easier map to benchmark), but that really doesn't matter when it's giving a false impression of the overall game performance.

I'm pretty sure Ryan is getting paid enough by Anand to get a representative benchmark of BF3 out. Yet over and over again, AT remains an outlier for this and other games, and continues to be an Nvidia outlier with every new card. Is it really so hard to benchmark the leading last gen title on every scene and get a general idea of performance?

For the last year Nvidia has had a 40%-20% lead in BF3 on Anandtech. It's just bullshit.

I remember recently dave mentioned that Anand only test 1 map for BF3, let me dig that up.

Edit:

FYI - AnandTech happens to use one of the few maps that is a little more unfavorable to Radeon than the other maps. If you check other reviews I believe most will paint a different picture, certainly at the higher resolutions.

If Ryan is around I'm sure he can explain their procedure. IIRC our performance labs standard sweep uses around 4 or 5 maps in BF3 and we can replicate Anand's positioning with one of them, the rest shows positioning more akin to what you'll see at many other sites.

 

[Ryan Smith]

And my reply 4 pages ago:

Actually it's pretty high. We're using Thunder Run, the only significant ground based on-rails map in the game. I don't know what anyone else is running, but if they followed the same train of thought they'd end up on the same map.

And why on-rails? It's very easy to replicate (I'm not the only one that needs to be able to do this) and very consistent run-to-run, which makes it easy for other editors to repeat it while not having our results bounce all over the place for no good reason. Also, it's not CPU bottlenecked at the high end, which gives us room to grow for more exotic configurations like tri-Titan. I'd much rather have a proper recorded benchmark (and I'm annoyed to this day that DICE took that out of the game; it was in the betas) but in a pinch on-rails will have to do.

 

[gkar1]

Ah, so we can simply ignore your reviews and conclusions since you deem it better to put personal convenience above accurate reporting. Thanks for clearing that up.

 

[Wynix]

Ah, so we can simply ignore your reviews and conclusions since you deem it better to put personal convenience above accurate reporting. Thanks for clearing that up.

Ignoring their entire review seems like an overreaction as they generally have great reviews, perhaps just ignore the BF3 bench test.

 

[Deleted member 2197]

Ignoring their entire review seems like an overreaction as they generally have great reviews, perhaps just ignore the BF3 bench test.

Excellent point, and one I hope it is not lost on deaf ears.

I'm glad some reviews do not use the same maps, because there really wouldn't be the need to have any more than one review. Using different maps may show that a particular video card may not have consistent performance throughout a particular game. As a consumer I'm interested if I play this game alot, and to know which cards give me the best performance in different game scenarios (maps) is vital. It can go either way, but in this case the results from different BF3 maps shows me the gaming results may not be consistent and some cards may perform better in throughout the entire game experience. While it's almost impossible to tell performance for the whole game experience with just 2 maps, it's definitely is a lot better than just having 1 map to base my decision .

 

[Homeles]

Ah, so we can simply ignore your reviews and conclusions since you deem it better to put personal convenience above accurate reporting. Thanks for clearing that up.

Ah, so we can simply ignore your opinions, since you cannot seem to understand the value of having consistent, reliable benchmarks. Thanks for clearing that up.

 

[Kaotik]

Ah, so we can simply ignore your opinions, since you cannot seem to understand the value of having consistent, reliable benchmarks. Thanks for clearing that up.

I wouldn't go to point where you have to have "on rails map" to get consistent and reliable benchmarks, as many if not most do several (3+) runs on each card using the same predetermined route (who knows, they could even use program to use exact same inputs every single run) to see the average performance on the said test scenario.

 

[lanek]

after watch a bit more of the results, i still dont know what to think about this card.

The card position itself between the 660TI and 670 ( a bit slower of the 670 overall ).. this put the card at 18-20% under the 770.. nothind dramatic, but they could have adjust a bit more the performance by keeping more shaders and come with 1344SP instead of 1152 .. specially with the 770 over who is basically a 680 with 1536SP..

 

[Wynix]

after watch a bit more of the results, i still dont know what to think about this card.

The card position itself between the 660TI and 670 ( a bit slower of the 670 overall ).. this put the card at 18-20% under the 770.. nothind dramatic, but they could have adjust a bit more the performance by keeping more shaders and come with 1344SP instead of 1152 .. specially with the 770 over who is basically a 680 with 1536SP..

No doubt they will release a 760Ti at some point(January?) and i imagine it will be a tweaked 670, the gap between the 760 and 770 is too large for me to believe otherwise.

 

[lanek]

No doubt they will release a 760Ti at some point(January?) and i imagine it will be a tweaked 670, the gap between the 760 and 770 is too large for me to believe otherwise.

yes ofc, and you still have OC models for fill the gap without attacking the 770 performance. And lets be honest, it will not cost much time and developpement to Nvidia for react quickly if needed ( on the end of year ) by reuse one more time the GK104 or directly a 770 with 1 SMx disabled.

 

[DSC]

MS did a demo of Tiled Resources running on GTX 770 on Win 8.1 Preview at BUILD.

http://msdn.microsoft.com/en-us/library/windows/apps/bg182880.aspx

http://code.msdn.microsoft.com/Direct3D-Tiled-Resources-80ee7a6e#content

 

[DavidGraham]

I believe that was DX11.2 .. but I thought NVIDIA only supports DX11.0 , they can't do DX11.1 let alone DX11.2 !
http://www.shacknews.com/article/79932/directx-112-windows-81-detailed

 

[Dave Baumann]

Tiled resources has two feature levels - the first requires DX11_0 hardware feature level as the base while the second tier requires DX11_1 feature level as the base.

 

[Homeles]

My memory of DX 11.1 is a bit stale, but to me it sounds like DX 11.2 is a bigger deal and/or more interesting to the end user.

 

[Deleted member 2197]

Lars (Nvidia Tech Marketing Mgr) responded when asked last Dec:

The GTX 680 supports DirectX 11.1 with hardware feature level 11_0, including all optional features.This includes a number of features useful for game developers such as:

• Partial constant buffer updates
• Logic operations in the Output Merger
• 16bpp rendering
• UAV-only rendering
• Partial clears
• Large constant buffers

We did not enable four non-gaming features in Hardware in Kepler (for 11_1):

• Target-Independent Rasterization (2D rendering only)
• 16xMSAA Rasterization (2D rendering only)
• Orthogonal Line Rendering Mode
• UAV in non-pixel-shader stages

So basically, we do support 11.1 features with 11_0 feature level through the DirectX 11.1 API. We do not support feature level 11_1. This is a bit confusing, due to Microsoft naming. So we do support 11.1 from a feature level for gaming related features."

The article goes on to state while Southern Islands series is DirectX 11.1 compliant,

" some AMD hardware from the 7000 series come only with the hardware feature level 11_0, as they are rebranded products from 6000 series of products (Northern Islands GPU architecture). It is rumored that the upcoming Kepler refresh (rumored GK112, GK114 and 117) will feature full hardware feature level 11_1 as well."


http://www.brightsideofnews.com/new...directx-111-with-kepler-gpus2c-bute280a6.aspx

 

[Dave Baumann]

" some AMD hardware from the 7000 series come only with the hardware feature level 11_0, as they are rebranded products from 6000 series of products (Northern Islands GPU architecture).

That is not accurate of channel discrete GPU's.