[Videocardz]First GTX Titan X 3DMark benchmarks!

[Elfear]

Here is my guess:


So I think it will all come down to $650ish R9 390X that is 80% of the Titan X in performance. Those who have the money and want the best go for single Titan X or Titan X SLI. Those who cherish performance/$ will go for R9 390X CF over single Titan X.
I can see both sides here.

The 390X would have to be a pretty poor improvement over the 290X to only be 80% of the Titan X based off the rumored performance. According to TPU the 290X is 8-15% slower than the 980 depending on resolution. If the Titan X is 36% faster than the 980 than you're expecting the 390X to be 9% faster than the 980 and ~18-27% faster than the 290X? I'd really hope AMD can do better than that.

 

[xthetenth]

At 60% performance above 290X, it will be 10% below Titan. That is if both 290X and 390X is clocked the same. You have 290X which is at 290W. 45% more cores, you are looking at what TDP? Is 380W plausible on a single chip? What if the 390X runs on a slower clock?

They're likely covering the power draw of the cores with power savings on the memory and from architectural improvements. And even if they were trying to deal with 380W, they're trying to do it with a solution they were using to deal with 500W. They'd better be able to do it without having to lower clocks. It looks good on the face of it for both performance and having the cooling to let that performance go.

 

[Paul98]

At 1080p, 2x960 = 980.
http://lmgtfy.com/?q=960+sli+vs+980

It's a moot point anyway, the 960 can only do 2-way SLI apparently. The 660 and 760 could do 3-way. I wonder why they changed that?



Yeah, I just figured that out. Super lame.

So you didn't even bother to look at benchmarks, got it

I am not talking about 960 SLI vs 980, that has to do with SLI scaling and other factors.

I am talking about actual FPS of a single 960 vs 980. You can look at those benchmarks and see what sort of scaling you should see going forward.

 

[Pandamonia]

So does my 780 Lightning SLI perform as well as this single Titan X?

 

[Cloudfire777]

They're likely covering the power draw of the cores with power savings on the memory and from architectural improvements. And even if they were trying to deal with 380W, they're trying to do it with a solution they were using to deal with 500W. They'd better be able to do it without having to lower clocks. It looks good on the face of it for both performance and having the cooling to let that performance go.

I have read slides from Samsung stating around 0.8W @ 7Gbps per DRAM. Multiply that with 16 you get 13W (290x). I dont think HBM will make much difference here. The shaders is the "problem" here.

Remember that 295X2 had 2 x 2816 shaders with 20Mhz higher clock than 290X (1 x 2816). Now they are dealing with 1 x 4096 shaders on a single die.

We will see. I could be wrong here and they can clock the 390X the same. Cooler Master might make a good cooling solution for it. It will undoubtly be one massive die. thats for sure. But so is Titan X

The 390X would have to be a pretty poor improvement over the 290X to only be 80% of the Titan X based off the rumored performance. According to TPU the 290X is 8-15% slower than the 980 depending on resolution. If the Titan X is 36% faster than the 980 than you're expecting the 390X to be 9% faster than the 980 and ~18-27% faster than the 290X? I'd really hope AMD can do better than that.

Well, ok maybe 15% slower then.
0.85 (15%) of GTX Titan X puts it at 6312 on 3DMark11 X.
R9 290X is at 3900 (+50%).
GTX 980 is at 5309 (+20%)

 

[therealnickdanger]

You may be seeing an argument where there is none. For reference, this is what I was responding to:

The 980 is basically double the specs of the 960, then the Titan X is triple the specs of the 960, and 50% more than the 980.

Might be able to get get about where actual performance will land by looking at the performance difference between the 980 and 960.

You stated that 980 ~ double 960 specs and TitanX ~ triple 960 specs. I agree with this, so I took it a step further show the effect via benchmarks.

So you didn't even bother to look at benchmarks, got it

I am not talking about 960 SLI vs 980, that has to do with SLI scaling and other factors.

I am talking about actual FPS of a single 960 vs 980. You can look at those benchmarks and see what sort of scaling you should see going forward.

The benchmarks I posted show the practical evidence of your initial argument: a single 960 performs about half of a 980 while two 960s perform about the same as a 980 (give or take 15%). Double the architecture ~ double the performance. If it were possible to do 3-way 960, we could get a preview of TitanX performance. The obvious issue with that is the lack of VRAM on the 960, but the 4GB model is coming soon that should help close the gap.

TL;DR: I agree with you and thought some benchmarks would help validate your statement. Sorry for helping?

 

[Silverforce11]

These days R290X is ~10% behind a 980.

It's not like at Maxwell's launch where the gap was bigger and the R290X lagged behind the 780ti, now its faster (or Kepler tanked, whichever you prefer).

R390X from the same leaks is ~60% faster than R290X.

To put that into perspective, if Titan X is only ~35% faster than 980, it will lose. Titan X needs to be ~50% faster than 980 to secure a strong win.

Edit: If NV is targeting it around 225W, it should be around 35% faster than 980, due to the 980 being a 160-180W part, keeping efficiency the same, 35% extra puts it right into that mark. Pushing higher clocks (OC) should get 50% faster but obviously at increased TDP, well beyond what the Titan reference blower is capable of, and it'll get hot & noisy (which is not something NV wants!).

 

[tviceman]

These days R290X is ~10% behind a 980.

You keep saying this but it's such a crap comparison. The 980 has way more TDP and core clock headroom. Hawaii in it's current form is being basically pushed to it's near max out of the gate, while GM204 is well below it's limits. Most 980 cards come with decent overclocks to begin with and are ~20% faster out of the box than a max user OC'd 290x.

To put that into perspective, if Titan X is only ~35% faster than 980, it will lose. Titan X needs to be ~50% faster than 980 to secure a strong win.

Sigh. This 35% crap again. The first graph clearly shows Titan X being 41% faster. But round down if it helps your argument LOOK better.


Regardless of your twisting and contorting, Titan X is going to be priced ridiculously too high to be a valid comparison to Fiji (this is hoping AMD doesn't go full retard in pricing). What will be the more reasonable comparison is a GTX 980 TI (if there is one) vs. Fiji.

 

[nvgpu]

This popped up on Chiphell, take it however you want.

 

[Paul98]

You may be seeing an argument where there is none. For reference, this is what I was responding to:


You stated that 980 ~ double 960 specs and TitanX ~ triple 960 specs. I agree with this, so I took it a step further show the effect via benchmarks.



The benchmarks I posted show the practical evidence of your initial argument: a single 960 performs about half of a 980 while two 960s perform about the same as a 980 (give or take 15%). Double the architecture ~ double the performance. If it were possible to do 3-way 960, we could get a preview of TitanX performance. The obvious issue with that is the lack of VRAM on the 960, but the 4GB model is coming soon that should help close the gap.

TL;DR: I agree with you and thought some benchmarks would help validate your statement. Sorry for helping?

You are missing what I said, I am saying is that you can look at the real world performance of a single 960 vs 980. Then the specs of the Titan X, 980, and 960 allow you to get a good idea of where the Titan X should fall in real world performance. You can do this by looking at how the performance scales from a single 960 to 980. Some games scale well, others not.

Edit: though I expect DX12 should help with scaling.

 

[Silverforce11]

You keep saying this but it's such a crap comparison. The 980 has way more TDP and core clock headroom. Hawaii in it's current form is being basically pushed to it's near max out of the gate, while GM204 is well below it's limits. Most 980 cards come with decent overclocks to begin with and are ~20% faster out of the box than a max user OC'd 290x.

What percentage of OC do 980 typically reach? Hawaii can go from 1ghz to 1.2ghz so thats a straight 20% OC with added vcore. Power consumption goes crazy so 1.15ghz is more the norm, or 15%. But my point re the performance gap isn't so much to debate with you R290X vs 980. It's to showcase the overall performance we extrapolate from to arrive at Titan X vs 390X.

Sigh. This 35% crap again. The first graph clearly shows Titan X being 41% faster. But round down if it helps your argument LOOK better.

Regardless of your twisting and contorting, Titan X is going to be priced ridiculously too high to be a valid comparison to Fiji (this is hoping AMD doesn't go full retard in pricing). What will be the more reasonable comparison is a GTX 980 TI (if there is one) vs. Fiji.

Please dude, take a chill pill.

Anyone with logic would compare similarly priced products in the same segment, so I agree with that. The problem is because 390X is AMD's top dog, it will be always compared against NV's top dog, regardless of price.

In GAMES, if Titan X is only ~35% faster than 980 (3dMark is comparable for similar architectures), its going to find itself behind 390X that's ~60% faster than R290X, that much should be simple to understand.

NV needs to clock it higher out the door and aim for 45-50% above 980.

 

[Cloudfire777]

This popped up on Chiphell, take it however you want.

Faster than GTX Titan Z?

 

[Silverforce11]

This popped up on Chiphell, take it however you want.

Looks about right for the 3dMark (graphics) results. ~35% in that, and ~33% in games on average.

116 x 1.33 = 154

As i posted, NV reusing the titan blower is fine but that blower drops off significantly when TDP is higher than 225, as seen in Titan Black where it already gets noisy & hot. If NV push clocks on Titan X to get 50% above 980 (170W x 1.5 = 255W), they're looking at >250W and that blower just isn't capable to remain cool & quiet.

But it should have the OC headroom to allow enthusiasts to manually crank up the blower or put it on water.

 

[antihelten]

Well, ok maybe 15% slower then.
0.85 (15%) of GTX Titan X puts it at 6312 on 3DMark11 X.
R9 290X is at 3900 (+50%).
GTX 980 is at 5309 (+20%)

Using your numbers though, it quickly becomes apparent that 3DMark is not a good tool to gauge performance between different GPU families. Your numbers indicate that the GTX 980 should be about 36% faster than the R9 290X (5309/3900), whereas in actual games its only about 11% faster (source: TPU 1440p). In other words 3DMark overestimates the performance of GTX 980 (or underestimates 290X depending upon perspective) by about 20% (1.36/1.11).

 

[Silverforce11]

Faster than GTX Titan Z?

Titan Z looks good for 3D Mark but not all games scale well with multi-GPU. So yes, it could well end up being faster on average in 22 games.

Because launch is imminent with NV's conference coming up, these leaks from China & Taiwan I feel are solid with final production samples & latest drivers. There's also no surprises so far, the numbers fall in line where they are expected to be.

What's interesting is how well it scales at 4K and in multi-GPU config, we may see the gap increase a lot from the 980 due to all the extra ROPs/TMU and plenty more bandwidth.

 

[Paul98]

You keep saying this but it's such a crap comparison. The 980 has way more TDP and core clock headroom. Hawaii in it's current form is being basically pushed to it's near max out of the gate, while GM204 is well below it's limits. Most 980 cards come with decent overclocks to begin with and are ~20% faster out of the box than a max user OC'd 290x.



Sigh. This 35% crap again. The first graph clearly shows Titan X being 41% faster. But round down if it helps your argument LOOK better.


Regardless of your twisting and contorting, Titan X is going to be priced ridiculously too high to be a valid comparison to Fiji (this is hoping AMD doesn't go full retard in pricing). What will be the more reasonable comparison is a GTX 980 TI (if there is one) vs. Fiji.

There is a single graph there at 41% all the rest are less than that.

 

[Ramses]

Titan X is going to be priced ridiculously too high to be a valid comparison to Fiji (this is hoping AMD doesn't go full retard in pricing). What will be the more reasonable comparison is a GTX 980 TI (if there is one) vs. Fiji.

This seems logical and in line with the previous cards from both outfits.

 

[parvadomus]

These days R290X is ~10% behind a 980.

It's not like at Maxwell's launch where the gap was bigger and the R290X lagged behind the 780ti, now its faster (or Kepler tanked, whichever you prefer).

R390X from the same leaks is ~60% faster than R290X.

To put that into perspective, if Titan X is only ~35% faster than 980, it will lose. Titan X needs to be ~50% faster than 980 to secure a strong win.

Edit: If NV is targeting it around 225W, it should be around 35% faster than 980, due to the 980 being a 160-180W part, keeping efficiency the same, 35% extra puts it right into that mark. Pushing higher clocks (OC) should get 50% faster but obviously at increased TDP, well beyond what the Titan reference blower is capable of, and it'll get hot & noisy (which is not something NV wants!).

My bet is AMD will win no matter what. Their chip is not bandwidth bottlenecked nor is consumption constrained.

Looking at Tonga then:

(4096sp/1792sp) = 2,285
2,285 * 54 = 123,4 * 1,1 (tonga runs @ 918Mhz) = 135,74 for 4096 cores running @ ~1Ghz

135,74 / 83 = 63% faster than 290X
135,74 / 92 = 47% faster than 980

But remember that score is based on a 285 with possible bandwidth bottlenecks..

 

[Cloudfire777]

Another leak. Same power consumption as GTX Titan. Its a 250-260W TDP card.

 

[happy medium]

This popped up on Chiphell, take it however you want.

Does that mean its 53% faster than a gtx970/290x?
If so I see no way for the 390x to be 53% faster than a 290x unless its another 340 watt (gtx480) hot box.

 

[Silverforce11]

Another leak. Same power consumption as GTX Titan. Its a 250-260W TDP card.

Versus 390X for context. Titan X uses 37W less than 390X. Not as big a power gap as from the R290X vs 980.

The same source has said many months ago, 390X is faster than "full-fat" GM200. Looks like they may be right. The interesting thing is the OC potential. Exciting times ahead (even if its 28nm!!).

 

[Cloudfire777]

Versus 390X for context. Titan X uses 37W less than 390X. Not as big a power gap as from the R290X vs 980.

The same source has said many months ago, 390X is faster than "full-fat" GM200. Looks like they may be right. The interesting thing is the OC potential. Exciting times ahead (even if its 28nm!!).

Who said those slides was of R9 390X other than wccf speculating it was that? It could have been Titan X as well, and they were not far off

I guarantee that the R9 390X draw a lot more than +30W than Titan X. Unless its a brand new architecture which still havent been confirmed or denied by anything yet. All we know is that it will come with Hybrid Cooler. Thats all we know. Everything is assumptions based on that. Hybrid cooler = hot = High TDP = Same architecture as 290X. Thats the assumptions which doesnt fit with the slides being from 390X

 

[kagui]

so you guys think this guy was correct?
DG Lee
http://udteam.tistory.com/688

and 390x

 

[Enigmoid]

3% difference

Given drivers and game selection (the positioning of the cards have changed) the two are pretty much equal.

Don't know the final clocks either. The 390X could have picked up a lot of driver optimizations or it might not. Either way, I would be conservative and say they perform roughly the same.

 

[Silverforce11]

Who said those slides was of R9 390X other than wccf speculating it was that? It could have been Titan X as well, and they were not far off

I guarantee that the R9 390X draw a lot more than +30W than Titan X. Unless its a brand new architecture which still havent been confirmed or denied by anything yet. All we know is that it will come with Hybrid Cooler. Thats all we know. Everything is assumptions based on that. Hybrid cooler = hot = High TDP = Same architecture as 290X. Thats the assumptions which doesnt fit with the slides being from 390X

If you think its Titan X with the 295W? lol

You can't guarantee jack and you know it. It's all a guess.

We know so little of the next iteration of GCN, everything & anything is possible.

The reference R290X was hot & high TDP. AMD copped major flak for it. It makes sense for them to go with a great reference design this time around, particularly since they've had amazing results with the R295X2.