GeForce Titan coming end of February

Page 24 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
Status
Not open for further replies.

sontin

Diamond Member
Sep 12, 2011
3,273
149
106
You are hopeless. You have no clue as to the difference between Single Precision Compute and Direct Compute / Compute shaders.

That's not a counter saying the same think to me.
BTW: Using DirectCompute to accelerate effects is as good as using it to calculate 100 hundreds of lights.

So now you are debating the efficiency of games vs. their graphics? What DirectCompute features does Crysis 2 have? I am all ears. Global illumination, contact hardening shadows? Let's go, name them.

"DirectCompute features"? :$
Crysis 2 is a DX11 game. There is no reason to use DirectCompute when it's not better or faster. So where is all the compute performance of GCN?

Perfomance is accelerated because AMD leveraged Compute Shaders in the game engine. :rolleyes:

Yeah, i guess you have still no clue. Sleeping Dogs used a internal Downsampling filter. No ComputeShader. There are DX9 games like The Witcher 2, Trine 2, SeSam3 with have the same option.

Xbitlabs HD7870 beating GTX670 no problem in Dirt Showdown, a heavy Compute title.

Repeating the same stuff over and over and over makes is not right. It's Showdown. Why don't you use Sleeping Dogs? Or Hitman? Or Farcry 3? Or Secret World? Or Metro2033? Or Lost Planet 2?

Irrelevant to the discussion of DirectCompute. That's like me downplaying Tessellation advantage of GTX480's by focusing on its 6 months later launch, or its price against the 5870. What's that have to do with Fermi's tessellation advantage over Cypress/Cayman? Nothing. Dropping a red herring I see.

What? You comparing only the 7970GHz to the GTX680 ignoring that the later came 5 months earlier to the market. :rolleyes:
BTW: A GTX680 is faster in Farcry3 than the 7970. And that game uses DirectCompute in the exact same way you promoted it. :awe:
 

f1sherman

Platinum Member
Apr 5, 2011
2,243
1
0
I am going to take a wild stab and say the Titan has 905-925mhz GPU clocks with 235W power consumption in the same game. But then I can't reconcile how a 550mm2 chip with 6GB of VRAM can hit 925mhz on 28nm node when a 365mm2 925mhz 7970 used 189W with 3GB of GDDR5. Could the 28nm node have matured that much?


Don't go Charlie on us! You can't have it both ways, and afterwards go "told ya" :colbert:

If you are really taking chance with 925MHz / 235W forecast, you don't get to have safety net, and in the very next sentence proclaim "But then I can't reconcile how..."
 

RussianSensation

Elite Member
Sep 5, 2003
19,458
765
126
BTW: A GTX680 is faster in Farcry3 than the 7970. And that game uses DirectCompute in the exact same way you promoted it. :awe:

Keep spinning facts. It doesn't use it to the same extent. Go read AMD's blogs. Also, 925mhz 7970 is not a competitor to the 680, Ghz edition is.

Far-Cry-3-Test-MGPU-1080p.png


"With the Radeon HD 7970 GHz Edition we were able to have HDAO enabled, but still had to disable AA. Therefore, the HD 7970 GHz Edition was a bit faster in Far Cry 3 compared to the GTX 680 by allowing the higher Ambient Occlusion setting at 2560x1600." ~ Source

FC3 doesn't use DirectCompute to the extent the other 4 games I listed do. That's why I specifically stated "compute heavy" titles. FC3 is not one of those. It's not heavy on compute. I don't even know why you brought up Metro 2033, Planetside 2, etc. None of those games takes advantage of GCN's DirectCompute extensively. The reason Metro 2033 performs faster on AMD cards is not because of Compute Shaders. Why you are throwing random games into the same pile?

Here clear as mud the 3 compute heavy titles:

11_797vs68_big.png
 
Last edited:

RussianSensation

Elite Member
Sep 5, 2003
19,458
765
126
Don't go Charlie on us! You can't have it both ways, and afterwards go "told ya" :colbert:

If you are really taking chance with 925MHz / 235W forecast, you don't get to have safety net, and in the very next sentence proclaim "But then I can't reconcile how..."

I don't get paid for throwing wild guesses on AT forum. What do I need a safety net for? I said it's a wild guess, even though I logically can't reconcile it. In other words I am perfectly fine being incorrect on that guess. This is not the Bar Exam or a Medical License Exam. My world won't change upside down if I guess wrong. It's a lot more conservative than other people's guesses of 1Ghz 2880SP or 1.13ghz 2688 SP parts. ^_^
 

BallaTheFeared

Diamond Member
Nov 15, 2010
8,115
0
71
Isn't DirectCompute woven into DX11, I would assume most games that use DX11 use DC in some form or fashion, is that not correct?

All this arguing, there are benches for DirectCompute guys...

http://www.videocardbenchmark.net/directCompute.html

45163.png


45165.png



There are more, one called DirectCompute & OpenCL Benchmark... A 5870 is twice as fast at stock as my 470 is @ 900MHz in that bench. *waffles*
 

f1sherman

Platinum Member
Apr 5, 2011
2,243
1
0
I don't get paid for throwing wild guesses on AT forum. What do I need a safety net for? I said it's a wild guess, even though I logically can't reconcile it. In other words I am perfectly fine being incorrect on that guess. This is not the Bar Exam or a Medical License Exam. My world won't change upside down if I guess wrong.

no doubt, no doubt...

but what is the meaning of logically irreconcilable forecast :confused:

looking at the logic of it, is that even a stab/forecast?
 

sontin

Diamond Member
Sep 12, 2011
3,273
149
106
FC3 doesn't use DirectCompute to the extent the other 4 games I listed do. That's why I specifically stated "compute heavy" titles. FC3 is not one of those. It's not heavy on compute. I don't even know why you brought up Metro 2033, Planetside 2, etc. None of those games takes advantage of GCN's DirectCompute extensively. The reason Metro 2033 performs faster on AMD cards is not because of Compute Shaders. Why you are throwing random games into the same pile?

So, now Metro2033 is not a "compute heavy game" even it's using DirectCompute for the DoF and looks miles better than Sniper Elite v2?
And Farcry 3 is not using DirectCompute like Sleeping Dogs? Yeah. :rolleyes:

Here clear as mud the 3 compute heavy titles:

11_797vs68_big.png


Ah, i get it: Compute heavy game = Game in which my company has the faster card.
So - Sniper Elite v2 is a compute heavy game but Farcry 3 is not.
Sleeping Dogs is a compute heavy game but Crysis 2 is not.
Nether Metro2033 nor Hitman is a compute heavy game because Tahiti is only faster thx to the higher bandwidth.

:awe:
 

SirPauly

Diamond Member
Apr 28, 2009
5,187
1
0
Apparently HD7850-7870's GLOPs, texture fill-rate, pixel fill-rate, VRAM, GPU clock speed, tessellation, memory bandwidth advantages over the GTX680........oh wait. [Clue: contact hardening shadows, global illumination, HDAO/AO are accelerated using DirectCompute in that game]

My point was more-so about the huge discrepancy with sleeping dogs like where a 7850 was faster than a GTX 690.
 

tviceman

Diamond Member
Mar 25, 2008
6,734
514
126
www.facebook.com
AA and Compute are related in those games. Some of the AA is calculated using Compute Shaders. Same story with Dirt Showdown forward + AA. That's the whole point. AMD is using Compute Shaders to perform certain graphical functions faster. SSAA in Sleeping Dogs is one of those. It says so on the AMD Blog.

It's not memory bandwidth since HD7850 can outperform GTX680 in Sleeping Dogs.

Hmmmm that's interesting. Sounds like AMD is optimizing games to run exceptionally well on their hadware and exceptionally poorly on Nvidia hardware. Why move AA to a compute-shader function when it wasn't needed to be moved there to begin with? Sorta like why are concrete bricks being overly tessellated. Both companies play dirty.

Anyways, missing out on something like SSAA isn't the same to me as missing out on physx when available. When playing a game that has physx, I can tell when it is or isn't enabled. When playing a game with SSAA, I have to stop and study the differences to be able to tell.
 

SirPauly

Diamond Member
Apr 28, 2009
5,187
1
0
Personally welcomed the in-game SSAA though in SLeeping Dogs, there was a lot of texture and shader aliasing with out the ability.
 

2is

Diamond Member
Apr 8, 2012
4,281
131
106
Some games work better on nVidia others work better on AMD, this has always been the case with the exception of a couple slip ups by both companies. Like the 2900 series from AMD/ATI that couldn't compete and the FX ilne from nVidia. Short of those generational anomalies, they've always traded blows depending on the game.
 

The Alias

Senior member
Aug 22, 2012
646
58
91
Hmmmm that's interesting. Sounds like AMD is optimizing games to run exceptionally well on their hadware and exceptionally poorly on Nvidia hardware. Why move AA to a compute-shader function when it wasn't needed to be moved there to begin with? Sorta like why are concrete bricks being overly tessellated. Both companies play dirty.

Anyways, missing out on something like SSAA isn't the same to me as missing out on physx when available. When playing a game that has physx, I can tell when it is or isn't enabled. When playing a game with SSAA, I have to stop and study the differences to be able to tell.

doesn't forward+ rendering reduce the perf overhead when being used compared to d.r ? that's not playing dirty that's moving forward

and yes the hell you can bro
 

f1sherman

Platinum Member
Apr 5, 2011
2,243
1
0
SSAA out of the box in recent AMD titles is freaking great!

Global Ilumination Forward+ style is an utter garbage(don't make me post Dirt Showdown pics), rivaled only with Metro 2033 DOF.
Crysis 2 tessellation a whole another lvl altogether. Because it actually looks great, and it's the most impressive implementation of said tech. Overdone bricks notwithstanding.

Tessellation2.gif
 
Last edited:

boxleitnerb

Platinum Member
Nov 1, 2011
2,605
6
81
I thought we had concluded in my other thread that it wasn't DirectCompute in those games (Sleeping Dogs etc) that was the issue since Tahiti LE isn't doing so well. I don't get it...

Btw as for texel and pixel fillrate:
They requirements should increase with resolution and especially with OGSSAA like in Sleeping Dogs. According to hardware.fr, the GTX680 and the HD7970 have very similar fillrates, yet the 7970 (GE) can be significantly faster in the aforementioned titles. I still believe it is SP GFLOPs and memory bandwidth that plays a critical part here.
 

RussianSensation

Elite Member
Sep 5, 2003
19,458
765
126
Global Ilumination Forward+ style is an utter garbage(don't make me post Dirt Showdown pics), rivaled only with Metro 2033 DOF.

What you are saying is the performance hit with Global Illumination ON is not worth the performance penalty. A lot of people would agree with you. However, what most people don't realize is how extremely intensive global illumination is. If you don't want this feature in next gen games, no problem. Say you do?

"Thankfully, we were able to mitigate the performance penalty by migrating the lighting computation to DirectCompute, giving the breathtaking compute capabilities of the GCN Architecture a runway to strut its stuff. Graphics Core Next’s compute resources are actually so powerful that we were able to take this lighting one step further with a separable bilateral filter, which further improves the performance of the simulation and scrubs the scene of potential artifacts in the shadowing."

In other words, the performance hit without DirectCompute would have been way worse (like it is on HD6970 or GTX680). I don't disagree with you that the Global Illumination is way too advanced for modern GPUs today given the huge performance hit, even with DirectCompute path. However, in the future as GPUs get 4-5x more powerful, wouldn't you rather reduce the performance hit by 70-80% if you could do so using DirectCompute? See what you guys aren't reading is DirectCompute is used in cases where GPU works faster, not slower to perform the same calculations. This is not at all like tessellation. The downside is that some of these more advanced graphical features aren't worth their cost on most GPUs today because the existing architectures aren't truly advanced enough for games to start using way more compute shaders. At least in F+ model the MSAA works awesome in Dirt Showdown while games like BF3/Crysis 3 double the MSAA performance hit under the deferred lighting path.

Both companies play dirty.

From AMD blog on Sleeping Dogs:

"You, as the user, have configured your game to run at 1920×1080, and you’ve selected 4xSSAA as your anti-aliasing method. These settings tell the graphics card to render the game’s content at a 4x larger resolution of 3840×2160 (ultra-high definition), then resize that frame back down to 1920×1080 before display on the monitor. At 3840×2160, the game might have a hard edge with 16 pixels that are obviously jagged. After the resize to 1920x1080p, however, these pixels are reduced by our SSAA factor (4x in our example), leaving you with a considerably smaller jagged edge of just four pixels. This effect is applied across the entire scene, hiding visual artifacts in the same way shrinking a picture in an image editor can hide flaws in a photo. As we did with HDAO, however, we take AA one step further in Sleeping Dogs. The “Extreme” anti-aliasing setting uses the compute horsepower of Graphics Core Next to do another anti-aliasing pass on the final frame, which will smooth out those last four pixels of aliasing we described in the example above. The resources required to drive the extreme setting are quite intense." ~ Source


Remember how I said as a joke that once AMD starts throwing $ at developers with its AMD Gaming Evolved, we'd need an AMD and NV GPU for AMD GE titles and NV's TWIMTPB titles? :p

Anyway, most of us here already bought HD7000/GTX600 series. What's even the point of arguing so intensely since it's not as if anyone of us will switch back to the other card. No, we'll be upgrading to much faster cards.

Supposedly a leaked image of the Titan = 8+6 pin. The card has 12 memory chips on the front, 12 on the back are expected to have a backplate. The PCB holds a total 8-phase VRM, which is situated at the backside of the PCB along with ICs and MOSFETs. Additional two VRM phases are located near the SLI connectors that power the memory. There are actually 2 SLI fingers on the GPU which allows Quad-Way SLI Support.

NVIDIA-GeForce-GTX-Titan-PCB-Picture.jpg


Here is GTX480 for comparison.

gtx480_pcb.jpg
 
Last edited:

Pottuvoi

Senior member
Apr 16, 2012
416
2
81
Isn't DirectCompute woven into DX11, I would assume most games that use DX11 use DC in some form or fashion, is that not correct?
You must write your program specifically to use compute shaders.
Also using compute shaders may be slower than just using pixel shaders to do the same thing. (they are easier to break.)
 
Status
Not open for further replies.