GeForce GTX 1180, 1170 and 1160 coming in August. Prices inside

[CakeMonster]

The most relevant comparison IMO is the yearly top card (Titan excluded). That has been: 980->980Ti->1080->1080Ti.

Those have average 20-30% performance increase. I couldn't care less about what the chip looks like under the hood, this is the realistic upgrade available to consumers. "Ti" in the model name is worthless for consumers as long as its not released simultaneously with the non-Ti version. But they're not, they're now just the yearly card that's faster by the same steady margin. Ti or non-Ti, they could have been labeled "2016 top card" and 2017 top card" and it would have been more useful.

As long as I have games to play I'm willing to pay for that kind of performance increase, although the nudging of the price point upwards is starting to push it now. In some sense its nice to know that we can expect that kind of steady performance increase every year from NV so that we can plan upgrades. But we seriously need to stop caring about what they name it because it makes no difference.

 

[Head1985]

That is worthless in actual practice. Comparing like that doesn't mean anything in real life. Plus the die size depends on millions of things, a ton of die size might be just to make certain chips work(if the yields are low, so more chance of getting full chips and selling at full price), a lot of it is in terms of computing for example, maybe things like physix for Nvidia, etc... it doesn't really represent mm2 per performance. If we had the SAME architecture at the same die size it would be a good representation of process and node improvements, but even that would be worthless in terms of the real world, because you can only compare same price points.

What is the point if the GTX 2070 is 100% faster than the GTX 1070 and same die size and same architecture, if it costs $1000? Is it the same range? You would need to compare the GTX 2070 to the Titan XP, otherwise the "gains" might be technically there, purely from a technical perspective, but there is no gains if you have to buy a GTX 2060 at $400 to match the 1070. That would obviously be the logical match, even if the GTX 2060 die size is say 25% smaller than a 1070.

Nv always make whole lineup with new generation.
Every DIE have their successor and predecessor.After that marketing can call them what they like and price it what they like even if new geforce name is green unicorn 10000 that card DIE will have predecessor with same DIE size category.
From smallest DIE to flagship DIE and in every category they make new DIE.
GTX780 was flagship/Big DIE.It was cutdown TITAN with 250w TDP similar to 980TI and 1080TI.GTX680 was midrange DIE similar to GTX980 and 1080.
Flagship/BIG DIE-GTX780/780TI,TITAN OG,TITAN black,980TI,TITANX,1080TI,TITANX pascal,TITANXP.
Midrange DIE-GTX680/670/770,660TI,760,980,970,1080,1070/1070TI.

Nv will target 50-55% perf gain within die size category so its pretty easy estimate new geforce performance.
GTX2080TI will be 50-55% faster than 1080TI and 30% faster than GTX2080
GTX2080 will be 50-55% faster than 1080 and 20-30% faster than 1080TI
GTX2070 will be 50% faster than 1070 and so on.

Of course they also can take midrange DIE and name it TITAN deluxe and sell it for 10000 but that DIE will be still midrange and they will have flagship/big DIE ready with much more performance.

 

[tviceman]

How else can you compare? Die size depends on a huge variation of things. If you look at Vega for example it can easily be much smaller if die space was cut down that is used for all of the compute stuff. Of course this is decided in the architecture, how you want to build it, but ultimately its the price segment that we compare.

Would it be fair to compare Vega 64 to GTX 1080ti? After all it's a lot more similar to that card in terms of TDP, die size, etc... No, because it's a $550 card, bar the mining craze.

So it easily goes 780 to 980 to 1080. How else would you compare? Even we go by your logic we have to compare the 770 to the 980 to the 1080ti. It doesn't make any sense.

In fact I think comparing 960 to 1060 is much more fair and telling, even though the 970 wasn't that far off at $330. But ultimately the 960 4GB cost about $230-240, much closer to the $250 1060 6GB.

And we already established that there was a huge architectural change in the Maxwell series that lowered power draw by a huge amount. This started off with the GTX 750 and 750ti and it was perfected with the 900 series. Since then even Pascal at 16nm only made incremental power draw improvements, purely from the process node, very little from architectural improvements.

First of all I wouldn't compare anything between AMD and Nvidia when it comes to release cadences and advances. That is comparing fruits to vegetables. Second of all, I wasn't comparing die sizes, I was merely pointing out that your comparison was based on consumer products, not architecture advances.

It doesn't "easily go" from 780 to 980 unless you're looking at it from a consumer standpoint, but even then it's still a bad comparison. The 780 was a derivative of the OG Titan. The 780 was flagship kepler. The 980 was not flagship Maxwell. The GTX 980's chip was a direct replacement of the GTX 680's chip. The 780 was a significantly cut down GK110. There is no cutdown GP102 chip to insert above the 1080 but below the 1080 TI.

As you point out, Maxwell was a significant departure in architecture which helped it gain more perf/w. It also had a 33% larger die size moving from GK104 to GM204. Maxwell didn't have the advantage of being on a more advanced node like Turing will have. Turing will go the same route as Maxwell in throwing more transistors and increasing die size to help make the chip run more efficiently than just the node improvements alone. But I agree that Pascal didn't make much of an improvement in perf/w outside of the node advances, and I think that was because it as a full node jump + a new architecture. Since we're dealing with a half node jump, and one Nvidia has already made a 800mm2 chip on, I believe we'll get the 50% performance / watt TDP improvement across the board as has been seen for the last several generations.

 

[jpiniero]

It's not really a half node jump. There is a quality improvement but the shrinkage is minimal at best. GT104 should be pretty close in size to GP102.

 

[PeterScott]

Another clue the cards may be named the 20 series, rather than 11. I always preferred 20, but the rumors have gone back and forth. Also looking more like the Chip Series will be something starting with A: Ampere?

https://videocardz.com/76978/manli-teases-geforce-gtx-2080-and-geforce-gtx-2070
"Manli also registered two GeForce 20 series codenames at EEC. If this leak was to be considered plausible and it should, the new cards should be called GeForce GTX 2080 and GeForce GTX 2070 respectively".

https://videocardz.com/76967/nvidia-board-partner-registers-ampere-ga104-400-gpu-at-eec
"On July 20th, Manli Technology Group has notified EEC certification office about upcoming products. In a wall of text, we discovered GA104 graphics processor, a GPU that was neither released nor confirmed by NVIDIA.

Not only that, the exact GPU model is listed, which is GA104-400."

 

[Elfear]

First of all I wouldn't compare anything between AMD and Nvidia when it comes to release cadences and advances. That is comparing fruits to vegetables. Second of all, I wasn't comparing die sizes, I was merely pointing out that your comparison was based on consumer products, not architecture advances.

It doesn't "easily go" from 780 to 980 unless you're looking at it from a consumer standpoint, but even then it's still a bad comparison. The 780 was a derivative of the OG Titan. The 780 was flagship kepler. The 980 was not flagship Maxwell. The GTX 980's chip was a direct replacement of the GTX 680's chip. The 780 was a significantly cut down GK110. There is no cutdown GP102 chip to insert above the 1080 but below the 1080 TI.

As you point out, Maxwell was a significant departure in architecture which helped it gain more perf/w. It also had a 33% larger die size moving from GK104 to GM204. Maxwell didn't have the advantage of being on a more advanced node like Turing will have. Turing will go the same route as Maxwell in throwing more transistors and increasing die size to help make the chip run more efficiently than just the node improvements alone. But I agree that Pascal didn't make much of an improvement in perf/w outside of the node advances, and I think that was because it as a full node jump + a new architecture. Since we're dealing with a half node jump, and one Nvidia has already made a 800mm2 chip on, I believe we'll get the 50% performance / watt TDP improvement across the board as has been seen for the last several generations.

I think you and Guru are just talking about two sides of the same coin. You're talking about actual Perf/W improvements (due to architectural and process improvements) and he's talking about the end result to the consumer (i.e. current top-end card owners (1080Ti) will probably only see a 10-25% improvement when the 2080 launches). You're both correct.

 

[ub4ty]

Buying these as soon as they come out. Hopefully a quick ebay flip of some of them will pay for one to keep and personally test out for viability. I'm particularly interested in new features like a rumored single finger nv-link and potential tensor cores. I sadly expect them to gimp its full capability and functionality. So, we'll see. I don't expect details on the development sdk side to be fully detailed. So, I will likely flip my first purchase on ebay. If the details are provided and solid enough, it might be a keeper from the get-go.

 

[tviceman]

I think you and Guru are just talking about two sides of the same coin. You're talking about actual Perf/W improvements (due to architectural and process improvements) and he's talking about the end result to the consumer (i.e. current top-end card owners (1080Ti) will probably only see a 10-25% improvement when the 2080 launches). You're both correct.

Yes to all of this.

 

[french toast]

Without flipping back through pages, have we got any ideas/stats on what we are looking at with Turing?.

 

[Samwell]

Without flipping back through pages, have we got any ideas/stats on what we are looking at with Turing?.

It might be a graphics card. But we're not sure, as there's also the name Ampere floating around.

That's all the information we have beside wild speculations.

 

[french toast]

It might be a graphics card. But we're not sure, as there's also the name Ampere floating around.

That's all the information we have beside wild speculations.

Ok, what about the next consumer graphics card? Any specs leaked?

 

[PeterScott]

Ok, what about the next consumer graphics card? Any specs leaked?

Post 130 above. Ampere (GA104) is almost certainly the name along with GTX 2070/80 as those are actually registered by a board partner.

 

[french toast]

Right ok, so we really don't know anything other than 12nm and gddr6.

 

[mv2devnull]

Post 130 above. Ampere (GA104) is almost certainly the name along with GTX 2070/80 as those are actually registered by a board partner.

A partner that has "2070" and "2080", but no "1070", "1080", nor their "Ti" variants.

 

[Samwell]

Ok, what about the next consumer graphics card? Any specs leaked?

We pretty much don't know anything. No real leaks beside the pcb leaks. All other stuff might be just fake. As we're 14 days from the presentation, i'm sure it'll just be an announcement and hard launch will be in September.

 

[FreshBross]

We pretty much don't know anything. No real leaks beside the pcb leaks. All other stuff might be just fake. As we're 14 days from the presentation, i'm sure it'll just be an announcement and hard launch will be in September.

Yup, best we can do is wait.

 

[amenx]

RTX 2080 (to signify the ray tracing capabilities of the new cards), not expected to be massive leap over Pascal.
https://www.youtube.com/watch?v=ms7HQ7rckpA

 

[crisium]

New week, new rumour. GT104... no wait GA104... actually TU104. GTX 1180... I mean GTX 2080... by which I really mean RTX 2080.

Actual facts should be coming soon hopefully.

 

[Tweak155]

RTX 2080 (to signify the ray tracing capabilities of the new cards), not expected to be massive leap over Pascal.
https://www.youtube.com/watch?v=ms7HQ7rckpA

Expected to be "breakthrough performance":

https://segmentnext.com/2018/08/10/nvidia-gtx-1180-performance-galax/

Pretty sure you can find a rumor to support anything right now.

 

[Head1985]

x06 GPU as GTX/RTX2070 thats like new crazy level of greed by Nvidia.Also the gap will be insanely huge between 2070 and 2080 around 35%.

EDIT:23x smx for 2080 thats 2944SP so its probably also cutdown.Probably from full 3584 core.So this cutdown 2944SP RTX2080 should be 2070 basicaly.But again greed and monopoly strikes again.

 

[jpiniero]

x06 GPU as GTX/RTX2070 thats like new crazy level of greed by Nvidia.Also the gap will be insanely huge between 2070 and 2080 around 35%.

EDIT:23x smx for 2080 thats 2944SP so its probably also cutdown.Probably from full 3584 core.So this cutdown 2944SP RTX2080 should be 2070 basicaly.But again greed and monopoly strikes again.

I would say the perf estimates are fake but I've been saying that the perf/$ is not going to be great compared to the current situation, especially when you factor that the x104 die is going to be near the size of GP102.

 

[CakeMonster]

Well if its the usual yearly 20-30% performance increase who are we to complain?

Unless they raise the price of course. Which they would nev.. oh well.

 

[alcoholbob]

Hmm so it looks like Titan V CEO edition will probably be rebranded as Titan RTX. 200GB/s more memory bandwidth, more ROPS, a clockspeed boost and probably an updated cooler will get a 15% ish boost.

 

[Sonikku]

The 2080 will have an "official" price of $700 with a "Founders Edition" price of $800. Except, demand will be so high that all variants will sell for the latter even after the founder editions have run their course.

How do I know?

Pure speculation.

 

[DigDog]

I would need to live like a heremit for 2 months eating only boiled rice for two months to be able to afford one.