Titan X Announced

[ShintaiDK]

So many flaws with that statement.
*textwall removed*

No the issue is you reject reality and try and substitute it with your own.

AMD is hammered into the ground in CPU, they are now being hammered into the ground in the GPU segment with a massively shifting marketshare.

And it is for the exact same reason in both, performance/watt in relation to the other component suppliers. Aka Intel and nVidia in this case.

Then the rest of your textwall of excuses and wishful future hopes is irrelevant. The only thing that matters is revenue and profit and the R&D budget to fuel the designs. Something AMD lacks all 3 of and we see the result.

 

[Cloudfire777]

GK104: 3.54billion Transistors. 1536 cores. 23046875/core
GM204: 5.2billion Transistors. 2048 cores. 25390625/core

GK110: 7.1billion Transistors. 2880 cores. 24652777/core
GM200: 8.0billion Transistors. x cores

80000000000/x = 26600000 (roughly due to higher density with Maxwell like shown with GM204)
80000000000 = 26600000x
x = 80000000000/26600000
GM200 Transistor Count: 3007
Nearest SMM: 23, Rounding up to 24

GM200: 24SMM, 3072 Cores. 384bit, 12GB VRAM
Thats my estimation :thumbsup:

 

[Cloudfire777]

I ran into the TITAN X again at the NVIDIA booth and was able to confirm a couple more things. First, the GPU will only require a 6+8-pin power connections, indicating that NVIDIA is still pushing power efficiency with GM200.

http://www.pcper.com/news/Graphics-Cards/GDC-15-NVIDIA-Shows-TITAN-X-Epic-Games-Keynote

 

[tviceman]

GM204 was a lot bigger than GK104 though. They don't have much room to grow from GK110, ~600mm2 is the limit at TSMC.

I'm expecting ~35% faster than 980 at ~780ti/Titan power levels (~220W), but they do have plenty of TDP room to clock it higher at stock for extra performance.

If it has all the same native hardware compute capabilities that GK110 has, then you are probably right. But if it is cut back by design in some capacity, then you're under estimating. Regardless, being only 35% faster than a gtx980 at gtx780 TI power levels would be a noticeable decline in perf/w over GM204. Kepler did not suffer a decrease in perf/w with it's bigger chips, and thus far neither has Maxwell when comparing GM107 to GM204. Also, it's true that GK110 did not have to worry about it's die size limit when it came to design, but it did have limited headroom to keep it's power draw at Nvidia's desired levels. GM200 will reverse these problems - having very limited die size increase to work with but having much more headroom to clock higher. So in the end it's got the same potential over GK110 that GM204 had over GK104, IMO.

 

[Silverforce11]

And it is for the exact same reason in both, performance/watt in relation to the other component suppliers. Aka Intel and nVidia in this case.

It's not that simple. Perf/w is a useless metric in isolation.

The problem for AMD is their CPU have worse perf/w AND worse performance outright. You cannot do well when you suffer that double whammy.

Before Maxwell, AMD was competitive with Kepler (R290X is these days faster than 780ti!), it had performance but worse perf/w. Then Maxwell just made that into a double defeat with the 980 having outright performance lead and massively more perf/w.

Their problem now is being late with a new gen GPU. If R390X arrives and is 250W (water, cool & quiet) with 50% performance gains, it will be very competitive.

There's no doubt in my mind, if AMD CPU actually was faster than Intel but had worse Perf/W they would be in a MUCH better place now.

 

[Silverforce11]

Being only 35% faster than a gtx980 at gtx780 TI power levels would be a noticeable decline in perf/w over GM204. Kepler did not suffer a decrease in perf/w with it's bigger chips.

Nope, do your calculations.

980 = ~180W. I just looked at all the recent 980 review, all the MSI, Asus, Gigabyte etc are ~180W.

180 x 1.35 (keeping perf/w identical!) = 243W.

If you think the 980 is a 160W part, then 35% perf & tdp increase = ~216W.

Last I saw, 780ti is ~220W average load in games (TPU, computerbase.de, kitguru etc). R290X is ~240W.

 

[tviceman]

Nope, do your calculations.

I don't need to when I know I'm right.

GK110 780 TI is 45-50% faster than GK104's 680 and 770.

GK110 780 TI is as efficient, if not more, than GK104.

Therefore GM200 will likely be more than 35% faster than GM204 if it's consuming the same power as a 780 TI.

 

[Cloudfire777]

Take GTX Titan and GTX 680 from the 1440p.
+37%.

So 35% is the minimum we will see I think from GTX Titan X over 980 most likely.
With 3072 cores like I calculated above, it should be 50% faster. But Titan X is probably not as agressive clocked, so 40% ish above GTX 980 is my guess

 

[Silverforce11]

You would have to compare the earlier launch results rather than years later. The reason for this should be clear as older stuff get left behind in optimizations or rather, newer stuff can be improved further.

At the launch of Titan X I expect it to be ~35% faster than the 980. Years later, it may well end up at 40-50%.

This is at the launch of the 780ti, at the relevant resolution when comparing to the 770, to give you context.
http://www.techpowerup.com/reviews/NVIDIA/GeForce_GTX_780_Ti/27.html

Perf/w is very similar to Gk104. So its clear NV's bigger die does not suffer worse perf/w (it suffers perf/mm2 instead, which isnt a metric thats important to gamers)..

So Titan X retains GM204's perf/w, extends it by 35%, it therefore should have ~35% more power use than the 980.

 

[tviceman]

Take GTX Titan and GTX 680 from the 1440p.
+37%.

So 35% is the minimum we will see I think from GTX Titan X over 980 most likely

While it's possible Titan X could be cut down, I don't think it will be. Everyone and their mom knows Fiji is in the pipeline and probably has working samples. Also, there won't be nearly as much time to slow roll GM200 as there was with GK110. On top of that, we're looking at an EXTREMELY mature 20nm process. I think Titan X will be fully functional GM200 chips.

 

[tviceman]

This is at the launch of the 780ti, at the relevant resolution when comparing to the 770

snip

Did you forget how to do math? 45% faster at 1440p which is exactly within the limits of what I said. I love how all the ultra 4k butt lovers run up with benchmarks at random resolutions to support an argument you try to create. Go through my post history talking about benchmarks and notice how I always stick with 1 fair-ground resolution - one that isn't unadopted (4k), and one that isn't over-bashed (1080p).

If you really think GM200 is only going to be 35% faster at 1440p+ then good for you, stick with it. But I proved that your theories based on power draw are off so adjust your theories.

 

[Cloudfire777]

While it's possible Titan X could be cut down, I don't think it will be. Everyone and their mom knows Fiji is in the pipeline and probably has working samples. Also, there won't be nearly as much time to slow roll GM200 as there was with GK110. On top of that, we're looking at an EXTREMELY mature 20nm process. I think Titan X will be fully functional GM200 chips.

We will have to wait and see.
I do think Volta is ready for launch next year with 16nm FinFET so this launch is close to that. But this is Nvidia, they may have some units disabled on the GM200 for one more card.

I dont think Nvidia have anything to fear about AMD`s Fiji. If the power consumption charts is semi correct, the water cooling bundle hinting toward a hot chip, I think AMD is locked in with GCN1.x with much less efficiency than Maxwell, and they need new node to compete against GM200.

Its 300W+ Fiji at $500-$600 vs 200W+ $999 Titan X that is 10-15% faster maybe. Then perhaps GTX 980 Ti later with lower price and less VRAM until Volta replace Maxwell in 2016

 

[tviceman]

We will have to wait and see.
I do think Volta is ready for launch next year with 16nm FinFET so this launch is close to that. But this is Nvidia, they may have some units disabled on the GM200 for one more card.

I dont think Nvidia have anything to fear about AMD`s Fiji. If the power consumption charts is semi correct, the water cooling bundle hinting toward a hot chip, I think AMD is locked in with GCN1.x with much less efficiency than Maxwell, and they need new node to compete against GM200.

Who knows. The water cooling rumor could have been just that. Or could be AMD's new default choice for a dual card. Who really knows what kind of performance improvements and/or efficiency improvements HBM will bring. If it can't help improve efficiency that much, then it'll end up being a way over hyped feature because AMD can't really afford to launch a bigger, faster chip without noticeable efficiency improvements.

 

[nvgpu]

Pascal in 2016 you mean, Volta is slated 2017 for the 2 new supercomputers for US govt.

http://www.anandtech.com/show/8727/nvidia-ibm-supercomputers

 

[Techhog]

I was really hoping AMD would beat this to market. They need to be on top at least temporarily at some point. :/ Interested in seeing the performance I guess... 6GB 980 Ti later this year?

Before people complain about this post, I can't really get very excited about what is likely to be a $1200 card.

 

[Silverforce11]

If you really think GM200 is only going to be 35% faster at 1440p+ then good for you, stick with it. But I proved that your theories based on power draw are off so adjust your theories.

How is my theory on power draw are off? You just said so yourself that GK110 is as efficient as GK104. I said so as well.

We can agree GM200 to be as power efficient as GM204.

I said if its 35% faster than 980, it has to therefore use 35% more power. How is that off?

You want to claim the 780ti is 45% faster than the 770, fine, at 1440p where the 1080p targeted chip suffers more, go for it. I don't really care as that is history.

The facts remain that NV's bigger chip maintains similar efficiency as their smaller chip. We are therefore extrapolating into GM200.

How much power do you want to place on the 980? 160W or 180W? All it is, is a simple multiplication from that, and we arrive at ~220-240W for a 35% faster GM200. If you want it 50% faster, you would have to multiply that power by 50%.

 

[lavaheadache]

I wonder if it'll be more powerful than a Titan Z.

 

[tviceman]

How is my theory on power draw are off? You just said so yourself that GK110 is as efficient as GK104. I said so as well.

We can agree GM200 to be as power efficient as GM204.

I said if its 35% faster than 980, it has to therefore use 35% more power. How is that off?

You want to claim the 780ti is 45% faster than the 770, fine, at 1440p where the 1080p targeted chip suffers more, go for it. I don't really care as that is history.

My link showed very little performance degradation between 1080p and 1440p with the 770 relative to the 780 TI. But that obviously didn't compute with you. So what you are telling me is that when you say GM200 is going to be 35% faster, you mean at 1080p. A resolution no one is even going to be talking about with GM200 to begin with. A resolution GM204 wasn't "targeted" for to begin with. Gotcha.

Seriously dude, you gotta think about what you say.

 

[Cloudfire777]

Pascal in 2016 you mean, Volta is slated 2017 for the 2 new supercomputers for US govt.

http://www.anandtech.com/show/8727/nvidia-ibm-supercomputers

Oh yeah, sorry about that, my bad. I remembered they changed the name
http://www.anandtech.com/show/7900/nvidia-updates-gpu-roadmap-unveils-pascal-architecture-for-2016

 

[Silverforce11]

So what you are telling me is that when you say GM200 is going to be 35% faster, you mean at 1080p. A resolution no one is even going to be talking about with GM200 to begin with.

Gotcha.

Nope.

What I am saying is perf/w will be similar so whatever % increase you expect it to be, it will correspond with a same % power increase. Capiche?

You can skew it ugly with the 770 vs 780ti comparison by upping resolution. NV has presented the GK104 based SKUs as targeted at 1080p, we know its performance tanks as resolution is upped. Why stop at 1440p? If you compare 3x 1080p or 4k, the 770 looks awful. Your link showed a very recent review. Definitely compare them on launch results to get a clearer picture of when GM200 comes, rather than after years of driver optimizations or lack of for older products.

I don't know what resolution NV targets the 980 as, I didn't watch their launch presentation.

 

[Cloudfire777]

I dont think they have much more room than 3072 cores if you look at the die size vs GK110 (2880 cores). Maybe 2SMM/256 more cores reserved for a GTX 980 Ti later. But thats the maximum

 

[NTMBK]

Is it confirmed to be Maxwell and not GK210?

EDIT Nvidia blog says Maxwell. Bring on GM200!

 

[DooKey]

I'm looking forward to this card. Other than the anticipated high price there's really nothing to complain about.

 

[tviceman]

I don't know what resolution NV targets the 980 as, I didn't watch their launch presentation.

Then why did you pick 1080p when you know the 980 is the fastest GPU on the market and thus making comparisons at 1080p is worthless?

Nope.

What I am saying is perf/w will be similar so whatever % increase you expect it to be, it will correspond with a same % power increase. Capiche?

You can skew it ugly with the 770 vs 780ti comparison by upping resolution. NV has presented the GK104 based SKUs as targeted at 1080p, we know its performance tanks as resolution is upped. Why stop at 1440p? If you compare 3x 1080p or 4k, the 770 looks awful. Your link showed a very recent review. Definitely compare them on launch results to get a clearer picture of when GM200 comes, rather than after years of driver optimizations or lack of for older products.

I see now. You arbitrarily pick a resolution and stick with it. When talking about how fast AMD is, you stick with 4K. Because that is where AMD looks best. When comparing Nvidia products, you do 1080p. Because that is what shows the smallest gains between Nvidia chips. And since you're worried about the lack of recent driver optimizations for older products, lets go back a few months then!

http://www.techpowerup.com/reviews/NVIDIA/GeForce_GTX_980/26.html

3x 1080p: 770 - 65, 780 TI - 94......44.6% diference
1440p: 770 - 65, 780 TI - 94..........43.9% difference
1080p 770 - 66, 780 TI - 93...........40.9% difference

OMG Where is the massive drop off in 770 performance????? There isn't. Now do you see why I picked 1440p? Because 4k isn't adopted yet, 1080p isn't what GM200 buyers are going use, and the 770 was quite capable of 1440p gaming without massive performance degradation beyond typical pixel scaling.

How many times do you I have to prove you wrong?

 

[tweakboy]

Im on this like mayanoise in subway sandwich