GP106 speculations

[selni]

By that logic the 1080 should be about 45% faster than the 980 Ti, and yet it is only about 32% faster.

1080: 2560C x 1750Mhz x 2 = 8.96 TFLOPs
980 Ti: 2816C x 1100Mhz x 2 = 6.20 TFLOPs

8.96 / 6.20 = 1.45

So we're looking at a performance deficit of about 9% (1.32/1.45=0.91). This will more or less cancel out the 15% increase in theoretical performance, and thus GP106 should end up about even with a 970.

The various VR specific features come on top of this of course.

GPU boost makes these calculations hard (the stated clocks are not a hard cap), particularly with Maxwell being well known for the advertised speeds being very conservative.

 

[tential]

Vega has improved perf/watt over polaris according to amd slides.

 

[Silverforce11]

Vega has improved perf/watt over polaris according to amd slides.

Vega isn't Polaris GCN. It's their next architecture.

Quite clear in JPMorgan's interview, AMD claims this is their fastest ever roll out of a follow up new architecture after Polaris. Time-frames between new generations are normally at least a year or more. They are aiming for 6 months.

About GP106, agree that it should be 2 GPCs, 1280 CC. It's their easiest modular implementation. GTX 970 performance will be good, as you can imagine the power usage will be extremely low.

 

[R0H1T]

I think GP104 (1080/1070) may be more efficient (perf/watt) than Polaris 10. I am much more doubtful that GP106 will be. And it will be GP106 that is competing in Polaris 10's market segment. For an apples-to-apples comparison in the >300mm^2 segment, we will have to wait for Vega.

As I stated before, we saw this with Maxwell: GM204 was the most efficient 28nm GPU, with GM206 lagging behind in perf/watt. And back in the Kepler days, GTX 660 (full GK106) had a 140W TDP. I wouldn't be surprised to see the same thing again, especially since GP106 may be clocked as high as GP104 or even higher. As long as it fits on one 6-pin connector, it's not that big a deal for desktops; no one really cares much about whether it draws 125W or 150W when gaming. It's a bigger deal for laptops and other thermally-constrained designs, such as AIOs (iMac).

The 980 is the most efficient GPU for resolutions above 1080p, the 750Ti is the most efficient for gaming at FHD & below.

GPU efficiency is closely tied with the power that's being fed to the GPU & more importantly if it's (artificially) power limited. The stock 750Ti, being powered from the PCIe slot, comes without a power connector but there are custom models that come with additional power connector & are OC beasts.

I'd say such custom models won't have the same level of efficiency, similarly many of the custom 1080's are also slotted to draw more than 180W with additional power connector(s) & there again the much touted efficiency goes out of the window.

The point being Nvidia clocked the 1080 possibly to its limits & knowing most of the FE cards, like on review sites, won't be able to maintain the top boost clocks on their less than stellar cooler. So anyone boasting 2GHz boost clocks & unparalleled efficiency should pick one, otherwise they're misled or are being intentionally obtuse for some reason.

 

[tviceman]

If GP106 is 128-bit 32 ROPs, will GP107 be 64-bit 16 ROPs?

 

[Enigmoid]

If GP106 is 128-bit 32 ROPs, will GP107 be 64-bit 16 ROPs?

I'd guess GP 106 is 192 bit and GP107 128 bit. With increasing performance comes increasing bandwidth requirements.

GP208? will likely be 64 bit.

 

[antihelten]

GPU boost makes these calculations hard (the stated clocks are not a hard cap), particularly with Maxwell being well known for the advertised speeds being very conservative.

The clocks used are actually more or less in game boost clocks:

980 Ti boosts to an average of 1128 MHz
1080 boosts to an average of 1783 MHz

At least as long as the GPU isn't thermally limited (which seems to be a very real issue with the 1080 FE).

 

[ShintaiDK]

At least as long as the GPU isn't thermally limited (which seems to be a very real issue with the 1080 FE).

Not with retail driver and cards.

 

[antihelten]

Not with retail driver and cards.

There are plenty of examples of people with retail cards having throttling and I have seen no evidence that the driver version changes this (i.e. I have seen no comparative tests of review/release drivers).

Just because some people don't experience it throttling, doesn't mean that it isn't something that can happen. Not all GPUs/cases/games/climates are created equal.

 

[ultima_trev]

If this is just speculation, I'm picturing this...

1080 is ~30% faster than 980 Ti
1070 is ~10% faster than 980 Ti

I envision 1060 will be 5% slower than 980 Ti, with 4GB of GDDR5X, a sub 100 watt power draw, sub 60 centigrade thermals at load and priced at $180 where it will undercut and outperform RX 480. Months following, nVidia will soon have a monopoly on the dGPU market.

I claim at least 3/4 of my prediction above will be true.

 

[mohit9206]

If this is just speculation, I'm picturing this...

1080 is ~30% faster than 980 Ti
1070 is ~10% faster than 980 Ti

I envision 1060 will be 5% slower than 980 Ti, with 4GB of GDDR5X, a sub 100 watt power draw, sub 60 centigrade thermals at load and priced at $180 where it will undercut and outperform RX 480. Months following, nVidia will soon have a monopoly on the dGPU market.

I claim at least 3/4 of my prediction above will be true.

You're picturing it wrong. 1060 won't be anywhere close to 980ti. Because if it is then that would cannibalize 1070 sales.
No one would buy $400 1070 if $250 1060 is only 20% slower. Nvidia wouldn't want that so just like 960 was very far from 970,so will 1060 from 1070. At best similar performance to 980.

 

[selni]

The clocks used are actually more or less in game boost clocks:

980 Ti boosts to an average of 1128 MHz
1080 boosts to an average of 1783 MHz

At least as long as the GPU isn't thermally limited (which seems to be a very real issue with the 1080 FE).

Funny you mention that, as that's a reference 980 ti sitting right on the thermal cutoff (83°C).

Fair point though - assuming that's the exact card various reviews compared performance to and there's not too much variance in chip characteristics it's close (and that's probably the case for quite a few sites), but better cooled cards can hit much higher clocks.

 

[antihelten]

Funny you mention that, as that's a reference 980 ti sitting right on the thermal cutoff (83°C).

Fair point though - assuming that's the exact card various reviews compared performance to and there's not too much variance in chip characteristics it's close (and that's probably the case for quite a few sites), but better cooled cards can hit much higher clocks.

Better cooled cards can certainly hit higher clocks, but that goes for both the 980 Ti and the 1080.

As such I think the only proper thing is to compare reference to reference (as is done in the links I posted), or compare aftermarket to aftermarket. Currently there aren't that many reviews of aftermarket 1080s out there, so it's a tad harder to do the later (although not impossible).

 

[William Gaatjes]

Nvidia might use partial disable again with Pascal as they did with the Maxwell based GTX970.
Have no idea how they improved Pascal in detail. But it is a handy feature to harvest partially defective chips.

http://www.anandtech.com/show/8935/geforce-gtx-970-correcting-the-specs-exploring-memory-allocation

 

[dark zero]

No...no that again... nVIDIA didn't learned from what they did before?

 

[ShintaiDK]

Nvidia might use partial disable again with Pascal as they did with the Maxwell based GTX970.
Have no idea how they improved Pascal in detail. But it is a handy feature to harvest partially defective chips.

And you base that speculation on what?

The next SKU coming is a full GP106.

 

[Sweepr]

And you base that speculation on what?

The next SKU coming is a full GP106.

My opinion as well.

 

[Bacon1]

Not with retail driver and cards.

Yes they do, its been shown many times. Just because your card can barely maintain boost clocks while at 90% average usage doesn't disprove throttling.

https://www.reddit.com/r/pcmasterrace/comments/4mlfme/_/

He returned his SLI 1080 because they were super hot and constantly throttled.

 

[tential]

No...no that again... nVIDIA didn't learned from what they did before?

That they make money hand over fist?

 

[ShintaiDK]

Yes they do, its been shown many times. Just because your card can barely maintain boost clocks while at 90% average usage doesn't disprove throttling.

https://www.reddit.com/r/pcmasterrace/comments/4mlfme/_/

He returned his SLI 1080 because they were super hot and constantly throttled.

Yes, 63C and 53C idle is so credible when the card idles in the 30s.

You can do better. But I am sure we are going to hear the same "story" over and over until Vega releases. Its a long wait for you.

 

[Bacon1]

Yes, 63C and 53C idle is so credible when the card idles in the 30s.

You can do better. But I am sure we are going to hear the same "story" over and over until Vega releases. Its a long wait for you.

Lol keep trolling

He posted pictures of his setup here:

https://www.reddit.com/r/pcmasterrace/comments/4mczh8/today_i_upgraded1z/

That's the thing though, I didn't overclock at all and I didn't expect to push two 1080s in SLI to 100% when it comes to 4K since benchmarks had average FPS in the 45-55 FPS range from a single card. Due to throttling I am barely getting that from two cards.
Just a bit disappointing really.

He was very disappointed in his cards. Sorry to burst your bubble but the 1080 FE is a flounder's edition not founders.

 

[IntelUser2000]

GPU boost makes these calculations hard (the stated clocks are not a hard cap), particularly with Maxwell being well known for the advertised speeds being very conservative.

1. Nvidia on their 1080 presentation said 1.7x effective bandwidth over 980 because of the following:

1.4x bandwidth
1.2x compression benefit
Total = 1.7x

1080 = 320GB
980Ti = 336.5GB
980 = 224GB

So 1080 has effectively 380.8GB/s of memory bandwidth. That's only 13% increase in memory bandwidth over 980 Ti.

2. There's only 64 ROPs on the 1080 versus 96 on the 980 Ti.

Let's compare the full specs.

1080: 8.96TFlops, 112GT/s, 380.8GB/s effective memory bandwidth
980 Ti: 6.195TFlops, 105.6GT/s, 336.5GB/s memory bandwidth

44.6% increase in flops, 6% increase in texture fill rate, 13% increase in memory bandwidth. Different games benefit from different things but unless 980 Ti was so imbalanced, the lack of increase on the fillrate and memory bandwidth might be holding it back? 32% increase is actually impressive considering the specs. Of course despite improving marginally on some things the cost is just as much as the previous gen flagship.

Against the 980, it has 80% more flops, 45% more texture fillrate, and 70% more effective memory bandwidth. And it achieves very close to 70% gain.

Based on this I'd call Pascal actually more efficient with equal spec than Maxwell Gen 2.

 

[antihelten]

Yes they do, its been shown many times. Just because your card can barely maintain boost clocks while at 90% average usage doesn't disprove throttling.

https://www.reddit.com/r/pcmasterrace/comments/4mlfme/_/

He returned his SLI 1080 because they were super hot and constantly throttled.

To be fair something very weird is going on with his setup.

Shintai already mentioned the rather insane idle temps, which shouldn't really happen unless the fans are turned off (which they shouldn't do with the stock fan curve).

Secondly he reports his cards reaching 92 degrees, even though 82 degrees is the thermal limit, and the cards should throttle down if they go above that.

Thirdly his fans apparently goes up to 100%, but the stock fan curve should max out at 55%.

All in all, his experience may be true, but in that case he must have heavily tweaked the power/temp/fan settings of his GPUs. Either that or he has two duds.

 

[Cookie Monster]

Lol keep trolling

He posted pictures of his setup here:

https://www.reddit.com/r/pcmasterrace/comments/4mczh8/today_i_upgraded1z/

He was very disappointed in his cards. Sorry to burst your bubble but the 1080 FE is a flounder's edition not founders.

I like how one users experience means its the de facto experience for all 1080 users..

Posts like this comes off as very aggressive and you come off as someone trying to downplay a product as much as possible. Not sure where the beef is because people who buy it enjoy it minus a few exceptions of course (like this particular user).

Everyone here knows theres something wrong with that reddit users setup. No video cards regardless of brand in this day and age should idle that high to begin with.

edit - sorry for the OT. But to keep the topic on track, I think the GP106 might maintain a full 256bit bus + 20SM (half of GP104). Definitely it will replace the 970/980 series cards.

 

[antihelten]

edit - sorry for the OT. But to keep the topic on track, I think the GP106 might maintain a full 256bit bus + 20SM (half of GP104). Definitely it will replace the 970/980 series cards.

Would be a bit weird if GP106 maintains the full 256bit bus and also uses GDDR5X, if it has half the shader count of GP104. On the other hand I could see Nvidia going for a 256 bit bus if they combine it with GDDR5 instead.