GTX 1070 have only 1920SP!

[R0H1T]

Can we give the FUD spreading a rest, please? Let's wait to see what the 1070 officially brings before insinuating that "thar be skeletons in that thar closet." Thanks.

Sure, we'll see when or if the free (& fair?) press gives Nvidia a free pass for the nth time.

 

[nitromullet]

Tough to say, but it seems like if the 1070 has the same RAM issue as the 970, they would have just put 6GB on it to begin with. It's not like that won't be enough for this card.

From looking at the GTX 1080 previews, that card still struggles at 4k. I'm thinking the target market for the GTX 1070 will be 1080p and 1440p gamers.

 

[Stuka87]

Tough to say, but it seems like if the 1070 has the same RAM issue as the 970, they would have just put 6GB on it to begin with. It's not like that won't be enough for this card.

From looking at the GTX 1080 previews, that card still struggles at 4k. I'm thinking the target market for the GTX 1070 will be 1080p and 1440p gamers.

They can't drop it down like that though. No matter what the amount of memory is, is the rops are cut the way they were on the 970, you end up with a chunk of the memory having poor bandwidth. If its done just like the 970, it would have 1GB of RAM that has very poor performance.

 

[tential]

Tough to say, but it seems like if the 1070 has the same RAM issue as the 970, they would have just put 6GB on it to begin with. It's not like that won't be enough for this card.

From looking at the GTX 1080 previews, that card still struggles at 4k. I'm thinking the target market for the GTX 1070 will be 1080p and 1440p gamers.

Honestly, if it worked last time, why wouldn't you do it again?
People won't care, they'll make excuses to justify their purchases, or they'll say they are mad, but they won't be. Really, it's still going to be a good card whether it has 7 GB of ram or 8GB of the RAM of "fast" RAM. The only people who care are us people on forums and how many of us are switching our purchase brands? Not that many. It still looks better on a shelf and thus why it's more beneficial to do it than not.

 

[tential]

They can't drop it down like that though. No matter what the amount of memory is, is the rops are cut the way they were on the 970, you end up with a chunk of the memory having poor bandwidth. If its done just like the 970, it would have 1GB of RAM that has very poor performance.

This isn't a NEW thing Nvidia has been doing though. They're not going to say "Oh this portion of ram has less performance, lets not even count it. Or lets try and explain this to customers." That's insane. It's so confusing to do for them.

It has 8GB of VRAM. It was really a non issue with the vast majority of 970 owners. The 1080 will even be more of a non issue if true.

 

[Flapdrol1337]

I really doubt it works in the same way as with the 970, if so you'd be cutting 1/4th of the bandwidth of the already slower gddr5. Performance would be awful. Besides, why not put 6GB on in that case, would be plenty.

 

[RussianSensation]

I am very disappointed.GTX1080 have 33% more SP than 1070 and 25% more memory bandwidth.Is this card really GTX1070 or some GTX1060Ultra or 1060TI?
It is most cutdown x70 card ever made.

It's much worse than that because you didn't account for the disparity in clocks

Shader performance
1080 vs. 1070 = (2560 CC x 1733mhz) / (1920 CC x 1600mhz) = 44.4% higher

Texture performance
1080 vs. 1070 = (160 TMU x 1733mhz) / (120 CC x 1600mhz) = 44.4% higher

Memory bandwidth
1080 vs. 1070 = 320GB/sec vs. 256GB/sec or 25% higher but GDDR5X will overclock to 370GB/sec, while it's highly doubtful 8Gbps GDDR5 will go much beyond 8500-8600mhz (275GB/sec). That means max overclocked, 1080's should have 30-35% higher memory bandwidth over max overclocked 1070's bandwidth.

This is NOT a x70 level card. It's NV taking a GTX660/660Ti and re-badging it as a GTX1070.

GTX680 was 27% faster than GTX660Ti

Since GTX660Ti was a $299 card, NV is effectively doing this:

Reference cards: $499 GTX680 -> $699 GTX1080
Reference cards: $299 GTX660Ti -> $449 GTX1070

The biggest marketing scam in the history of GPUs accompanied by a huge pricing increase as well from already inflated prices of 2012 GTX600 series. Since the masses do not buy AMD, AMD has no cash flow for sufficient R&D and to hire the best engineers to be able to do anything about this nonsense.

PC gamers voted for years buying NV over and over and over and now they got exactly what was coming to them --> Record margins from NV = Record prices of mid-range GPUs (masquerading as high-end via marketing re-branding strategies, bifurcating a generation into two halves), while AMD is left hopelessly trying to scramble for the remaining 20% market share while still losing $$$.

Don't forget, a stock GTX670 cost $399 and outperformed GTX580 (aka Titan X predecessor) by 20% on launch day.

There is no way a GTX1070 will beat the Titan X by 20% at 1440p/1600p, further proving it's not a real x70 series card but a x60Ti re-branded to x70 because there is no competition from AMD.

None of this will change until people stop buying NV and start buying AMD. As long NV gets $ from its loyal userbase, they will continue to raise prices. As long as AMD doens't get $, they have no $ to develop next generation GPUs. This in turn allows NV to neuter performance any way they like, while raising prices. Expect GP100/102 to cost even more than $699 780Ti or $649 980Ti did. It's only going to get worse actually because AMD doesn't have money for a price war and lack the human capital/R&D resources to keep up. That means Vega should also cost a lot too. All the bickering online defending NV for decades have finally resulted in one of the worst next generation line-up of all time (i.e., Record prices of next gen mid-range cards and the most neutered x70/x75 card ever made).

 

[tential]

Ya, but what about the 1070 being on par with the 980Ti/Titan X?

If it's cut down that much can it still deliver on that claim?

 

[Bacon1]

Ya, but what about the 1070 being on par with the 980Ti/Titan X?

It will probably hit stock 980 Ti when OCing, but not OC'd 980 Ti.

 

[tential]

It will probably hit stock 980 Ti when OCing, but not OC'd 980 Ti.

I find this very hard to believe that Nvidia would not know their 1070 at STOCK clocks would be faster than a 980Ti. Why would they say that then?

Instead of being within about 20-25% of the 1080, the 1070 will be like 30-35% away from a 10800 in performance and below a 980Ti?
I find it really hard to believe Nvidia would risk that PR nightmare of releasing a slower than expected 1070 when Polaris will be RIGHT THERE. A slower 980Ti puts a 1070 in a very very awkward spot, in which a P10 GPU OCed could catch up to it at its stock clocks.

You really think Nvidia left themselves open like that?

 

[tviceman]

It's going to be 3-5% faster than Titan X and will still have a considerably better perf/$ than the GTX 1080. No one is making anyone buy anything. According to the AMD fans, Polaris is going to be Fury class speed at 110 watts for $299 or less, so your "problem" is solved right there.

 

[tential]

It's going to be 3-5% faster than Titan X and will still have a considerably better perf/$ than the GTX 1080. No one is making anyone buy anything. According to the AMD fans, Polaris is going to be Fury class speed at 110 watts for $299 or less, so your "problem" is solved right there.

I'm asking how can the 1070 be 3-5% faster than the Titan X, when it has over 40% less shader and texture performance than the 1080 according to RS's numbers....

This has nothing to do with my "problem" or about what "AMD Fans" think. Please don't derail this thread with the whole fanboy game.

My point is according to RS numbers, shader/texture performance disparity between the 1070/1080 seems to mean the 1070 can't deliver on it's performance claim of being faster than a Titan X. Which I don't believe Nvidia would leave themselves open to in the wake of Polaris announcing...

 

[Lepton87]

There is no way a GTX1070 will beat the Titan X by 20% at 1440p/1600p, further proving it's not a real x70 series card but a x60Ti re-branded to x70 because there is no competition from AMD.

None of this will change until people stop buying NV and start buying AMD. It's only going to get worse actually because AMD doesn't have money for a price war. That means Vega should also cost a lot.

Both the Titan X and the 980Ti are significantly under-clocked at reference after overclocking even the 1080 doesn't beat the 980Ti by 20%. It's actually under 15% faster. I can't imagine that the 1070 will be under 15% slower then the 1080 which means comparing overclocked to overclocked the 980Ti is going to be faster then the 1070.

 

[Bacon1]

I find this very hard to believe that Nvidia would not know their 1070 at STOCK clocks would be faster than a 980Ti. Why would they say that then?

Instead of being within about 20-25% of the 1080, the 1070 will be like 30-35% away from a 10800 in performance and below a 980Ti?
I find it really hard to believe Nvidia would risk that PR nightmare of releasing a slower than expected 1070 when Polaris will be RIGHT THERE. A slower 980Ti puts a 1070 in a very very awkward spot, in which a P10 GPU OCed could catch up to it at its stock clocks.

You really think Nvidia left themselves open like that?

It will be faster in VR, due to the new abilities, but for normal gaming? I don't think it will be.

 

[tviceman]

I'm asking how can the 1070 be 3-5% faster than the Titan X, when it has over 40% less shader and texture performance than the 1080 according to RS's numbers....

This has nothing to do with my "problem" or about what "AMD Fans" think. Please don't derail this thread with the whole fanboy game.

My point is according to RS numbers, shader/texture performance disparity between the 1070/1080 seems to mean the 1070 can't deliver on it's performance claim of being faster than a Titan X. Which I don't believe Nvidia would leave themselves open to in the wake of Polaris announcing...

Agreed. Everything in all reviews shows Nvidia did not exaggerate about 1080's performance, and there is no way there will be a 30% gap between 1080 and 1070. There wasn't even a 30% gap between GTX 980 and GTX 980 TI.

1070 OC probably won't be able to catch 1080 stock, but that is something Nvidia has been doing more of so no surprise there.

 

[Mopetar]

There were some people speculating in other threads that Polaris 10 would be in the same ballpark as a 1070 at 1920 SPs just based on them having roughly equal die sizes. Obviously it doesn't quite work out to 75% of SMs = 75% of die area as the memory controllers, etc. eat up some of that room, but it comes close enough for napkin math argument purposes.

 

[flopper]

I find this very hard to believe that Nvidia would not know their 1070 at STOCK clocks would be faster than a 980Ti. Why would they say that then?

Instead of being within about 20-25% of the 1080, the 1070 will be like 30-35% away from a 10800 in performance and below a 980Ti?
I find it really hard to believe Nvidia would risk that PR nightmare of releasing a slower than expected 1070 when Polaris will be RIGHT THERE. A slower 980Ti puts a 1070 in a very very awkward spot, in which a P10 GPU OCed could catch up to it at its stock clocks.

You really think Nvidia left themselves open like that?

They didnt plan for the change of strategy from amd.

 

[Rvenger]

GTX1070 diagram

Can GTX1070 even have 256bit and 64Rops.Just look how much cut it is...

Where did you find this picture? Were you the same individual who created it on overclock.net?

 

[maddie]

There were some people speculating in other threads that Polaris 10 would be in the same ballpark as a 1070 at 1920 SPs just based on them having roughly equal die sizes. Obviously it doesn't quite work out to 75% of SMs = 75% of die area as the memory controllers, etc. eat up some of that room, but it comes close enough for napkin math argument purposes.

Certainly it seems possible. It looks like there actually might be a fight between Nvidia and AMD despite the different targeted performance brackets. Some of us have been saying this for a while. Full P10 might = 1070 [cut 1080]. Let the competition begin.

The 1080 model, because of its price, will be targeting the very upper echelon of gamers. The rest should have a choice between the two GPU camps.

Maybe Nvidia priced the 1080 so high knowing that they could not supply enough and decided to milk as much as possible with the limited quantities available. The downside to this is when yields increase, they can't improve the specs of the 1070.

One option, is that they might insert a 1070Ti model in several months time.

 

[wilds]

If GDDR5X versions of the GTX 1070 ever come into fruition, it may perform a bit more like an X70 should. I would gladly buy a 7GB 1070 SLI if it had GDDR5X instead of a 1080.

I am quite bummed out about the 1070 though. Because of its specs, I expect the lower product stack to be as disappointing.

But much later near Titan P/1080 ti, I hope we see some weird combos. Maybe a 2 GHz+ 1060 with 4GB of GDDR5X.

Another weird, outlandish idea: An OEM releases a GTX 980 ti variant with GDDR5X. How hilarious would that be? Nvidia would be pissed seeing that steal the 1080's thunder.

It seems to me that NV doesn't want a lot of diversity in their lineup. We won't be seeing the likes of confusing SKU's such as GTX 650 vs GTX 650 Boost or 560 ti vs 560 ti 448.

X60 ti and X70 were merged to become the 1070. I have a feeling aftermarket 1070's will be a lot more exciting because of overclocking and different memory.

 

[GodisanAtheist]

As a (proud, mind you) HD 7950 owner, either camp that can double my current performance for $200 new has my money. A lot of people like to argue about pushing boundaries and 4k and VR but for the rest of us unwashed peons a nice performance boost at the ever stagnant $200 bracket would be all the reason we need to upgrade.

The 1070 at 980 ti performance is fine, and that performance for the (eventual) sub $400 price is OK too. Still a little too rich for most of us though.

I suspect AMD will be getting my cash two nodes in a row here (I usually flip back and forth) if the Polaris arch delivers on its price / performance one two punch.

 

[VirtualLarry]

As a (proud, mind you) HD 7950 owner, either camp that can double my current performance for $200 new has my money. A lot of people like to argue about pushing boundaries and 4k and VR but for the rest of us unwashed peons a nice performance boost at the ever stagnant $200 bracket would be all the reason we need to upgrade.

The 1070 at 980 ti performance is fine, and that performance for the (eventual) sub $400 price is OK too. Still a little too rich for most of us though.

I suspect AMD will be getting my cash two nodes in a row here (I usually flip back and forth) if the Polaris arch delivers on its price / performance one two punch.

I'm a 7950 owner as well, and I agree with your sentiment.

 

[raghu78]

I am thinking the GTX 1070 vs GTX 1080 will resemble a 660 Ti vs 680 or GTX 760 vs GTX 770. 25-30% slower and just about on par with Titan X. But given how well GTX 980 Ti scales with OC and the amazing clocking headroom a 980 Ti OC will still give the 1070 OC a run for its money.

We need to see if Polaris is good enough to bring some price cuts to Nvidia's upcoming products. The lack of competition has been damning for the last few years as Nvidia is laughing all the way to the back.

 

[Arachnotronic]

We need to see if Polaris is good enough to bring some price cuts to Nvidia's upcoming products. The lack of competition has been damning for the last few years as Nvidia is laughing all the way to the back.

Doubt it anytime soon. NVIDIA wouldn't have priced the way they did if they thought they'd have to cut prices anytime soon, they'd just look dumb in that case.

The team that NVIDIA has doing competitive intelligence is just killing it these days. 980 Ti rendered Fury X basically DOA for all intents and purposes (slightly faster @ stock, much better OC'ing headroom, more/cheaper VRAM, etc.), GTX 970 was very well priced, and so on.

 

[Azix]

GeForce GTX 1070
16 nm "GP104" silicon, 7.2 billion transistors, "GP104-200-A1" ASIC
1,920 CUDA cores, 15 out of 20 streaming multiprocessors enabled on the GP104 silicon
120 TMUs, 64 ROPs
256-bit GDDR5 memory, 8 GB standard memory amount
Max GPU Boost frequency 1600 MHz
6.75 TFLOP/s single-precision floating point performance
150W TDP, single 8-pin PCIe power connector
3x DisplayPort 1.4, 1x HDMI 2.0b
2-way SLI with SLI HB bridge support

How does it do 6.75 Tflops? My calculation gets under 6.2 and thats based on the max boost clock. The 1080 on the other hand has the right 8.2 tflops based on the base clock. What's up? It may not even be a 6 tflop card at its base clock.