[TPU] 1080 TI specs leaks

[Gikaseixas]

it's actually pretty brilliant, wish I'd predicted this

Brilliant for who? the consumer who will be buying slower memory but paying higher prices still(vs 1080 vanilla)???

If true, It should have premium 12GB GDDR5X for that price and performance tier. There's nothing brilliant about it. You know, it is ok to criticize NV too, maybe they'll listen and stop playing with it's customers intelligence.

 

[tviceman]

All being said - I don't believe the rumor that it'll come with regular 8gbps GDDR5. That just seems too little bandwidth over a 1080 and if I was paying in the $700-800 range for a card I would be really turned off by not getting the latest and greatest tech on it.

Do the math. It's 20% more bandwidth over the 1080 and 21.7% more shader / ROP throughput. It's scaled up 1:1 over GTX 1080 almost exactly. It won't be anymore bandwidth starved than the GTX 1080 is.

 

[Arachnotronic]

Brilliant for who? the consumer who will be buying slower memory but paying higher prices still(vs 1080 vanilla)???

There's more of it and effective memory bandwidth over the GTX 1080 goes up, even if it's "just" GDDR5.

If true, It should have premium 12GB GDDR5X for that price and performance tier. There's nothing brilliant about it. You know, it is ok to criticize NV too, maybe they'll listen and stop playing with it's customers intelligence.

Says who? They are providing more memory than the GTX 1080 has and more bandwidth. It's a higher end part and it gets more bandwidth and more of it. What's the problem here?

Anyway, if you don't like it, don't buy it, doesn't affect me in the slightest.

 

[Gikaseixas]

Do the math. It's 20% more bandwidth over the 1080 and 21.7% more shader / ROP throughput. It's scaled up 1:1 over GTX 1080 almost exactly. It won't be anymore bandwidth starved than the GTX 1080 is.

A 1080Ti with GDDR5X is possible. Just release with clocks that won't beat Titan X(P). Customers for those cards appreciate having the best tech available to them.

 

[Arachnotronic]

A 1080Ti with GDDR5X is possible. Just release with clocks that won't beat Titan X(P). Customers for those cards appreciate having the best tech available to them.

Customers can show their appreciation for the best tech available by buying a Titan X.

Actually I guess is memory technology is what someone cares about over, say, delivered performance, a Fiji has the most advanced/expensive to build memory tech around.

 

[Gikaseixas]

even if it's "just" GDDR5

What's the problem here?

Being just GDDR5 is the problem here. Why use very old tech on a 2016 flagship product???? You call it brilliant, i call it bad engineering

 

[n0x1ous]

Ill call it a great milking technique....I dont buy this rumor though. I think it will be G5X - possibly clocked lower but probably not.

 

[Gikaseixas]

Customers can show their appreciation for the best tech available by buying a Titan X.

So customers paying $800 don't deserve best tech but customers paying less (Vanilla 1080) do. WOW

Fiji has the most advanced/expensive to build memory tech around

And it shows tech advancement, a great selling point as it should be. It's a card that despite it's limitations, trades blows with GTX 1070 in some DX12 titles.

for heaven’s sake don’t pay 200+$ in 2016 for a 3GB video card

Quote from another thread... seems that Nvidia will do anything to milk it's way into uninformed people's pockets. Again, why pay $800 in 2016 for a GDRR5 video card???

 

[Arachnotronic]

Being just GDDR5 is the problem here. Why using very old tech on a 2016 flagship product???? You call it brilliant, i call it bad engineering

"Bad engineering"? Building a product that achieves performance and cost targets is not "bad engineering." Engineering isn't about throwing all of the fanciest features into a product without an eye towards cost structure/manufacturability.

By the way, the Tesla P40 uses GDDR5 not GDDR5X. Why do you suppose that is?

 

[crisium]

Hmm, I wonder how much faster it will over the 1080. Certainly much closer than the 980 Ti over 980.

1080 Ti vs 1080 is only a 21.7% TFLOP advantage and only 20% memory bandwidth advantage (default boost clocks)
1080 vs 1070 is 37.2% TFLOP / 25% memory (default boost clocks)

If we are to assume similar max OCs, then the comparison makes the 1080 fairly evenly spaced from its brothers:

1080 Ti vs 1080 + 30% Shaders / 20% bandwidth
1080 vs 1070 + 33.3% Shaders / 25% bandwidth

This is different than last gen. As we know 1070 has been removed from 1080 a little more than 970 vs 980. While last gen 980 was closer to 970 than to the 980 Ti, things have changed. The 1080 is closer to the 1080 Ti than the 1070 now.


If anyone wants to estimate 1080 Ti performance, you can start with knowing for sure it'll have a smaller lead over the 1080 than 1080's lead over 1070.

I actually like cards evenly spaced with performance, I guess the extra VRAM will be another differentiation for the 1080 Ti too and help it stand out since it's somewhat closer to the 1080. At max OC though there will be a nice even spread of 1070-1080-1080 Ti.

 

[Gikaseixas]

Engineering isn't about throwing all of the fanciest features into a product without an eye towards cost structure/manufacturability

Great engineering is about throwing all of the fanciest features into a product with an eye towards cost structure/manufacturability specially when you aim to charge that much for it.

By the way, the Tesla P40 uses GDDR5 not GDDR5X. Why do you suppose that is?

Does that makes the RX 480 a great product too???

 

[IllogicalGlory]

This leak is fake, not to mention it's a month old, as pointed out by a poster on the first page.

 

[railven]

Hmmm...

Wonder how close it will come to Titan Xp? Has Titan Xp shown more OC muscle over the current GTX cards?

Kicking myself for buying the GTX 1080 when I did. Just two weeks later manager gave me OT out the rear-end. Blew it all on an X99 system though haha.

 

[raghu78]

How much of a performance increase is there between a 2600K and a 6700K and what is the performance delta between the GTX 560 and the GTX 1070?

You can also play today's games just fine with a 2600K, but a 560 is unusable today.

dude that tells us more about the nature of the software running on these 2 different devices. Games keep demanding ever more GPU horsepower as developers keep pushing the envelope in the quest towards photorealism and life like graphics. Since graphics is an emabarrasingly parallel problem which adapts well to GPUs and GPU performance increases 1.75-2x at every process node/architecture change its much easier to see performance improvements in GPUs. Since games keep requiring ever more powerful GPU hardware we cannot play today's games on GPUs of 2010/2011. For CPUs the majority of client software is still single threaded (even though there is a healthy transition towards multithreaded) and increasing single thread performance is damn hard, especially if you are starting off such a strong architecture like Sandy bridge. Even multithreaded software on desktops rarely uses 8+ threads (there are very few even if thats the case). Server software is a different animal altogether and scales to hundreds of cores/threads. Thats why server users see huge benefits in performance at each process node/architecture transition. Just compare the performance of a sandybridge ep vs haswell ep vs broadwell ep at the same TDP (btw i am talking about the highest skus).

Just to point out the obvious I am pretty sure a gamer who bought a 2600k in 2011 and is running at 4.5 Ghz would have no problem running today's games and running the latest GPUs without bottlenecking. That is incredible value for money. Try telling that about a GPU from 2011.

 

[DamZe]

If they keep the GDDR5X in there and ship it with 8GB vRAM, then that is an OK compromise, an increased bus width and shader count will be enough to set I apart, yet not infringe in that holy 12GB Titan X territory. Whoever was buying ax X80 Ti labeled GPU was never concerned about vRAM, evidenced by the 780Ti's almost laughable 3GB onboard framebuffer with and 700$+ price tag, and it was still selling like crazy. 8GB in this day and age is more than enough, even for 4K. Not having GDDR5X in there reeks of milking. But we'll see.

ANYWAYS if its priced at around the 800$ mark then to whoever paid 700$+ (And god have mercy, think about the prices outside the US) for the regular 1080, you have my sympathy.

 

[n0x1ous]

If they keep the GDDR5X in there and ship it with 8GB vRAM, then that is an OK compromise, an increased bus width and shader count will be enough to set I apart, yet not infringe in that holy 12GB Titan X territory. Whoever was buying ax X80 Ti labeled GPUs was never concerned about vRAM, evidenced by the 780Ti almost laughable amount for a 700$+ price tag, 8GB in this day and age is more than enough, even for 4K. Not having GDDR5X in there reeks of milking. But we'll see.

It wont be 8GB because of the 384 bit bus. 6 or 12 and it wont be 6 so that leaves 12

 

[HurleyBird]

I wonder if anything would stop add in makers from substituting in GDDR5X

 

[Sushisamurai]

It's pure supply and demand folks. Obviously, they're shifting their S&D curve. The few sales they've lost to higher prices have been made up by their higher revenue. So long as there is demand, prices will remain as they are. It's the basic fundamental principles of a capitalist economy

 

[Head1985]

Hmm, I wonder how much faster it will over the 1080. Certainly much closer than the 980 Ti over 980.

1080 Ti vs 1080 is only a 21.7% TFLOP advantage and only 20% memory bandwidth advantage (default boost clocks)
1080 vs 1070 is 37.2% TFLOP / 25% memory (default boost clocks)

If we are to assume similar max OCs, then the comparison makes the 1080 fairly evenly spaced from its brothers:

1080 Ti vs 1080 + 30% Shaders / 20% bandwidth
1080 vs 1070 + 33.3% Shaders / 25% bandwidth

This is different than last gen. As we know 1070 has been removed from 1080 a little more than 970 vs 980. While last gen 980 was closer to 970 than to the 980 Ti, things have changed. The 1080 is closer to the 1080 Ti than the 1070 now.


If anyone wants to estimate 1080 Ti performance, you can start with knowing for sure it'll have a smaller lead over the 1080 than 1080's lead over 1070.

I actually like cards evenly spaced with performance, I guess the extra VRAM will be another differentiation for the 1080 Ti too and help it stand out since it's somewhat closer to the 1080. At max OC though there will be a nice even spread of 1070-1080-1080 Ti.

Yeah because 1080TI is not true TI card.
Its cutdown from already cutdown TITANXP.FULL GP102 have 3840SP.So TITANXP is true 1080TI card selling for 1200USD and 3328SP version is something like GTX780 was and selling as GTX1080TI.

 

[Headfoot]

Despite all the protestation, I am in the prime market for a 1080 Ti, and I will not buy one if it comes with the same GDDR5 tech that came on the 980 Ti at a slightly higher stock speed. I am waiting for 1080 Ti and Vega and I will be buying one of the two. There is absolutely ZERO chance I will spend $700 or more and not get the latest memory tech that other cards in their lineup are shipping for. I dont care. That's a lot of money to spend knowing you're getting old tech. If Vega is a turd too, I'll wait out yet another generation of cards if I have to.

I believe nVidia understands this and I hope they ship with GDDR5X. They may not but I think its more likely that they ship with GDDR5X on a 1080 Ti if they price it at $800. If it comes in cheaper then I think GDDR5 would be more likely. I'd say my confidence level is only a 60/40 chance though because its certainly possible nVidia would do something like that these days.

 

[b-mac]

Not the point, but whatever.

I want an NVIDIA Titan X for $1, but NVIDIA is too greedy to sell it to me at that price

Haha what an awful post. No one is arguing they deserve a new Titan XP for $1.

 

[swilli89]

Haha what an awful post. No one is arguing they deserve a new Titan XP for $1.

Seems to be a lot of hyperbole and rhetoric being thrown around here in an attempt at proving argument points. We probably should avoid using these tactics as we are getting force fed this type of discussion by both Schillary and Drumpf.

 

[Madpacket]

If rumours are true, it just looks like a forced product segmentation which is kind of smart on Nvidia's part - at least in theory. In reality the people who buy these types of cards are well aware GDDR5 and GDDR5X are different. They want the latest and greatest tech (regardless if the card has more memory bandwidth). It'll probably backfire on Nvidia but who knows, they sure know how to make lots of money by copying Apple's business model.

 

[jpiniero]

As long as it's performance competitive with the top Vega consumer card, nVidia can get away with GDDR5. I think what nVidia wants to avoid is what happened with Maxwell, where people stopped buying the Titan once the 980 Ti showed up, so they want the Titan XP to be unquestionably faster. Gimping the memory bandwidth will do that.

And I do think it'll be $899+.

 

[moonbogg]

My bet is the leak with DDR5 was intentional to lower expectations. Then when they release the actual card with DDR5X, people will flip the hell out and start shipping their money to Nvidia on large pallets.