Still waiting a 3500--4096SP Vega up to @1.50GHz HBM2 and sub 225w

I guess for sure we'll know how accurate the report is in a week or so when Nvidia releases it's earnings - but I personally think this rumor is false. For starters these architectures, the design, fabrication, testing, etc are planned out pretty far in advance. Suddenly moving the starting point of it up about 6 months is insane for a project of this scope. Especially when you consider that they usually end up delayed. So instead of a typical few month production delay, Nvidia intends on moving the launch of an entire generation up ~6 months? Seems far fetched. There are too many key technologies that aren't where they need to be yet. GDDR6 for example is supposedly ready for mass production in Q1 2018 by Micron - who is a big partner for Nvidia. I feel like if Nvidia was launching a series of cards within several months, why wouldn't they just wait until that's finished? Whatever AMD refreshes Vega/Polaris with next year will almost definitely have it. The two Titan Super Computers and Xavier, all Volta based are all slated for a 2018 launch. The Titan's were supposed to be 2017 but got delayed and the Xavier units were announced for volume shipping in 2018. We probably would have heard about it being taped out. Every major GPU release has gotten a rumor ~8-10 months prior to release of being taped out. We haven't heard anything like this about Volta. Which seems weird considering it's "supposed to launch" in the next 5 months. I also think a large part of the problem with stagnating sales is that current GPU's are simply enough for the vast majority of people. 90% of steam gamers own 1080p or lower monitors. For a long time, you'd had to constantly update your card in order to be able to game at 1080p with decent framerates in the latest games. This generation is different though - a 980/970/390/390x etc are all able to do 1080p @ 60 in most current games. Even more problematic for the industry is that 4K is the new target - but I doubt many gamers are going to upgrade. For example, the consoles generally control the "base" level of graphics. When a new console would come out - they'd just make all the graphics better and that would filter over to the PC, where you'd need a new card in order to drive that at 1080p. But this generation of consoles is shifting the target to 4K, so all the new GPU power is being utilized just to scale to that resolution. But most gamers probably won't be upgrading to 4K monitors anytime soon - thus a 980/1060/480/390/etc will all most likely be capable of running the next generation of AAA titles, at 1080p fine. So why upgrade? I don't think launching Volta early is really going to have an effect on that. I could be wrong, but I predict we won't see a Volta based consumer GPU until Q1 2018, probably around March. I think they will detail Volta this year - in fact I think we may even get some info on it at 2017 GTC in a few weeks (mostly AI/Deep Learning Focused Info) - but I don't think we'll see a real product this year. At least one that Guru3D people would be interesting in buying.

I agree, I can't see Volta coming out this year, it's too soon!

Nothing is too soon, it will be October - November for sure.

Yeah,could be a paper launch in Q3 2017.

I know that I will be skipping the 20 series of cards because I am waiting for any gaming Laptops that will be released with 6 or 8 cores. My 1070 will do be well for years to come because there will be very few new titles that I am interested in that my 1070 couldn't handle.

I guess for sure we'll know how accurate the report is in a week or so when Nvidia releases it's earnings - but I personally think this rumor is false. For starters these architectures, the design, fabrication, testing, etc are planned out pretty far in advance. Suddenly moving the starting point of it up about 6 months is insane for a project of this scope. Especially when you consider that they usually end up delayed. So instead of a typical few month production delay, Nvidia intends on moving the launch of an entire generation up ~6 months? Seems far fetched. There are too many key technologies that aren't where they need to be yet. GDDR6 for example is supposedly ready for mass production in Q1 2018 by Micron - who is a big partner for Nvidia. I feel like if Nvidia was launching a series of cards within several months, why wouldn't they just wait until that's finished? Whatever AMD refreshes Vega/Polaris with next year will almost definitely have it. The two Titan Super Computers and Xavier, all Volta based are all slated for a 2018 launch. The Titan's were supposed to be 2017 but got delayed and the Xavier units were announced for volume shipping in 2018.

Exactly this. Let's not even mention that Volta was supposed to come out in 2015 originally, and that Pascal wasn't even supposed to exist. NVIDIA seems to be doing way too many new things with Volta. New architecture, new node, new memory subsystem. I can't recall the last GPU that had all that changed at the same time, it's a giant endeavor. I shouldn't put a lot of credence on it, but the AMD Vega video from some months ago was also "mocking" NVIDIA for Volta being "late". Vega and Volta seem to be the next level of competition from both teams. Vega will come earlier on the 14nm node, Volta seems to go directly for the 10nm, and I'm 100% certain that it will have the performance crown until AMD shrinks Vega. What worries me about NVIDIA is that they try to do both tick and tock at the same time and change the GPU memory in top.

I also think a large part of the problem with stagnating sales is that current GPU's are simply enough for the vast majority of people. 90% of steam gamers own 1080p or lower monitors. For a long time, you'd had to constantly update your card in order to be able to game at 1080p with decent framerates in the latest games. This generation is different though - a 980/970/390/390x etc are all able to do 1080p @ 60 in most current games. Even more problematic for the industry is that 4K is the new target - but I doubt many gamers are going to upgrade. For example, the consoles generally control the "base" level of graphics. When a new console would come out - they'd just make all the graphics better and that would filter over to the PC, where you'd need a new card in order to drive that at 1080p. But this generation of consoles is shifting the target to 4K, so all the new GPU power is being utilized just to scale to that resolution. But most gamers probably won't be upgrading to 4K monitors anytime soon - thus a 980/1060/480/390/etc will all most likely be capable of running the next generation of AAA titles, at 1080p fine. So why upgrade? I don't think launching Volta early is really going to have an effect on that. I could be wrong, but I predict we won't see a Volta based consumer GPU until Q1 2018, probably around March. I think they will detail Volta this year - in fact I think we may even get some info on it at 2017 GTC in a few weeks (mostly AI/Deep Learning Focused Info) - but I don't think we'll see a real product this year. At least one that Guru3D people would be interesting in buying.

I'm around the same. A generous Q1 2018 will most likely be the Volta launch date. Now, Vega is launching in June the latest this year, and GloFo will have 7nm out in 2H2018. I can see Volta getting the performance crown at around March 2018, with a Vega 7nm refresh at around holiday 2018. I wish they synced for once :infinity: As for games themselves: GPU power has already stopped offering exponential increases since more or less the 28nm days. It will be the way that the GPU is programmed that will bring better visuals and performance, not necessarily the amount of hardware in the GPU itself. Vega and Volta will most likely be the most "similar" GPUs out from both AMD and NVIDIA. They will both have tile-based rendering, exceptional pixel and triangle throughputs, beefy command processors (a first for NVIDIA), HBM and a more advanced memory controller. We don't know if Volta's controller will be able to do what the Vega MMC is able to do (access even network storage), but I don't believe it won't. It might bite me back in the a*s to say this, but I'm almost certain that despite the 10nm shrink both Vega and Volta will be closer in frequency than what Maxwell and Hawaii were. Even the 2nd generation of Polaris has closed the frequency gap quite a bit (1.5GHz oc clocks vs 2.05GHz oc clocks now, while it was 1.3 vs 2.05). Vega will close this even more. The winner here will be the company with the best driver and the best developer penetration. Some might consider this an NVIDIA slam dunk, but AMD has EA, SEGA, SquareEnix, they have ported a forward renderer and they have a whole team working on the Unreal Engine, and now they also have Bethesda. What's left for NVIDIA is basically Ubisoft, if we're talking about AAA publishers. On the driver front things are still bad on the basic feature level (no adaptive, half, or even enforceable vsync/triple buffering yet), but AMD does have features that NVIDIA doesn't. Chill is a bigger deal than a lot of people in here believe, and the same goes for Relive which is an excellent piece of software. Hey, there are even tessellation controls there and Wattman is miles ahead from what NVIDIA has which is nothing. In the end things are muddy. In a sense both companies are distancing themselves from traditional gaming. NVIDIA is knee-deep in Ai research, and Volta's design will be informed by that, same as Vega's design as AMD is keen in entering that market with their compute designs and advanced memory controllers which would fit like a glove. In the grand scheme of things this will mean that games will become much more compute heavy and they will leave draw call handling to the GPU itself. The market share might not show it, but AMD's design approach has effectively "won". Compute is more important than graphics performance. I leave for the end something else. Despite the talk around it, the Ryzen CCX design is a crazy success. It practically means that what AMD has to care about wafer-wise is to produce a fairly small quadcore which then can be used in infinite configurations. I can see them do something similar with GPUs eventually, where they use a lot of smaller cores with a small latency penalty. GPUs are controlled via their driver and the command processor and they aren't as latency sensitive as CPUs are. This would be much easier to be accomplished compared to a CPU for these reasons, and integrating HBM means that the Infinity Fabric could have insane transfer speeds. That would mean that nobody has to worry about yields again and that we also can have immense GPUs.

Irony. The hardware has become more interesting than the software it's meant to run. We hit a card series that can do console ports with the 1080/70/60. Everything now is overkill. Because let's face it....AAA PC-exclusives died 10 years ago. I wait for the project scorpion sheep to claim it's console power will require this upgrade lol.

Irony. The hardware has become more interesting than the software it's meant to run. We hit a card series that can do console ports with the 1080/70/60. Everything now is overkill. Because let's face it....AAA PC-exclusives died 10 years ago. I wait for the project scorpion sheep to claim it's console power will require this upgrade lol.

If you take the console direct-to-metal improvements and the little custom things they both have, it's a generally good rule of thumb that you'll need whatever the console has +50% to get similar/better performance on the PC. That means that the new GPU performance target should be at around 8Tflop with at least 8GB of VRAM, and a CPU that can comfortably handle at least 8 threads. The Jaguars on the Scorpio are so customized that they aren't even really called like that any more, and they now run at 2.3GHz.

Titan P / 1080 T owners wont mind volta early launch at all, they either have the cash to back it up or they dont belong to the club so who cares 😉 I, myself as owner of 1070 bought shortly after premiere woldnt mind early voltra launch either, 1070 gave me alot of bang for the buck and really supported my upgrade to 1440p 32 inch monitor - games look awesome, jaw dropping at times 😉 If 2070 provides another 70 % perfomarmance boost over last gen like 1070 was compared to 970 I will just sell my 1070 for 200-250$ and spend another 200$ for new card... (I`m living in 2.5 world country btw 🙂

Irony. The hardware has become more interesting than the software it's meant to run. We hit a card series that can do console ports with the 1080/70/60. Everything now is overkill. Because let's face it....AAA PC-exclusives died 10 years ago. I wait for the project scorpion sheep to claim it's console power will require this upgrade lol.

There isn't a single big title I care about in 2017. I sold my 1070 so I needed a GPU. What I have is pretty much overkill for anything but 4k. So... yep. Irony.

Titan P / 1080 T owners wont mind volta early launch at all, they either have the cash to back it up or they dont belong to the club so who cares 😉 I, myself as owner of 1070 bought shortly after premiere woldnt mind early voltra launch either, 1070 gave me alot of bang for the buck and really supported my upgrade to 1440p 32 inch monitor - games look awesome, jaw dropping at times 😉 If 2070 provides another 70 % perfomarmance boost over last gen like 1070 was compared to 970 I will just sell my 1070 for 200-250$ and spend another 200$ for new card... (I`m living in 2.5 world country btw 🙂

70% is literally 1080 Ti performance, I really hope it's faster than that.

70% is literally 1080 Ti performance, I really hope it's faster than that.

The 1070 is the same performance of the 980Ti.

people in this site take rumors too serious. does anyone remember rumors about Vega? which according to rumors it was supposed to come out October last year. and many other rumors which turned out to be completely false. but I guess people never learn.

The 1070 is the same performance of the 980Ti.

Once overclocked the 980 Ti pulls well ahead.

That will be strange, what about 1080Ti owners? Im sure the new 2080 will be faster than the 1080Ti, especially if it has HBM2 and well we know Nvidia can do it (1080 vs 980ti) but 1080 never came so soon after 980Ti

So what? 1080Ti will not magicaly loose performance over night, there will be a only a newer faster card. I just need it to be suficcient for four years or so, which will be in QHD resolution no problem.

But the 1070 overclocks extremely well lol?

Then there are cases like this: https://youtu.be/AkE8mtVv_yg?t=5m53s

But the 1070 overclocks extremely well lol? I wouldn't call well ahead a few frames here and there.

Max oced 2.1ghz = ~ 1400mhz 980ti, that said most 980ti do at least 1450mhz+ pulling away further.

Irony. The hardware has become more interesting than the software it's meant to run. We hit a card series that can do console ports with the 1080/70/60. Everything now is overkill. Because let's face it....AAA PC-exclusives died 10 years ago. I wait for the project scorpion sheep to claim it's console power will require this upgrade lol.

True. But we can´t forget that some people want the best and greatest and can afford it, so Nvidia has to cater them. And we can´t forget those who have 4K or 144Hz screens or like VR. For those the extra power is welcome. For me the worst part is that nowadays most games are boring as hell not that they are console parts... On the good side, my new 1070 is going to last me a few years without any worries.

@PrMinisterGR Great post above.

Max oced 2.1ghz = ~ 1400mhz 980ti, that said most 980ti do at least 1450mhz+ pulling away further.

They are within 5% of each other. I'll take my 1070 because of how cool it runs. Don't get me wrong the 980Ti is a banging card. I wonder what the 1070 would have looked like with GDDR5X memory.

Max oced 2.1ghz = ~ 1400mhz 980ti, that said most 980ti do at least 1450mhz+ pulling away further.

If I would have an 980ti there would be no point to upgrade to a 107o, diffrence is to small to justify that move. From perspective of a person buying new card no one in thier right mind would choose 980ti over 1070 tho. Driver updates favor pascal cards, 1070 got 2 more gigs of ram too, so overall it`s much more future proof.