Rumor: NVIDIA GeForce GTX 1060 (Ti) could feature 6GB RAM

[P4-630]

NVIDIA GeForce GTX 1060 (Ti) with 6GB RAM?
Although the leaker clearly says GTX 1060 in his post, I think it’s too early to be sure if third GP104-based graphics card would be named after GTX 1060 or GTX 1060 Ti, as former GTX x60 Ti parts (GTX 660 Ti and GTX 760 Ti) were both based on GK104 silicons.

The GTX 1060 (Ti) is rumored to feature GP104-150 GPU and 6GB GDDR5 memory, whereas GTX 1070 would use GP104-200 and 8GB GDDR5. Finally GeForce GTX 1080 would come with GP104-400 GPU and 8GB GDDR5X frame buffer. So all these cards are supposedly based on Pascal GP104 GPU, but in different configurations. The 6GB variant would probably require 192-bit memory bus, which is actually not that surprising. Even GM204 had 192-bit variant, although it was only used for mobile GeForce GTX 970M and desktop GTX 960 OEM.

As we know Polaris 10 is aiming at mainstream segment. Therefore GTX 1070 or GTX 1060 (Ti) could become its direct competition. Should GeForce GTX 1060 (Ti) only feature 4GB RAM, it would look poorly in comparison to Polaris 10 with 8GB RAM. "So in my opinion decision to increase frame buffer on this mid-range part was probably dictated by Polaris 10."

NVIDIA GeForce GTX 1080 GP104-400-A1 8GB GDDR5X
NVIDIA GeForce GTX 1070 GP104-200-A1 8GB GDDR5
NVIDIA GeForce GTX 1060 (Ti) GP104-150-A1 6GB GDDR5

http://videocardz.com/59471/nvidia-geforce-gtx-1060-ti-6gb-ram

 

[RejZoR]

I'm not sure. Considering current high end cards don't really have enough grunt to utilize 4GB of VRAM in most cases, what's the point of sticking twice as much on upcoming mid end? At best, new mid end will be as fast as current high end. Fury X has 4GB, GTX 980Ti has 6GB. Fastest cards on the planet at the moment. But I know the real reason. Big memory cards sell well even if GPU can't even virtualy utilize it. I don't understand how are people after all this time still this stupid.

 

[SK-1]

This as an obvious evolution. I'm baffled at the "controversy" of adding 2gb more vram.

 

[Caring1]

I'm not sure. Considering current high end cards don't really have enough grunt to utilize 4GB of VRAM in most cases, what's the point of sticking twice as much on upcoming mid end? At best, new mid end will be as fast as current high end. Fury X has 4GB, GTX 980Ti has 6GB. Fastest cards on the planet at the moment.

I think 4Gb is the limit on the Fury X, no point adding more if it can't be used.

 

[SK-1]

I've seen tests that suggest 4GB may be a limiting factor for the Radeon R9 Fury X at 4K.

 

[Dethroy]

Fury X is limited to "just" 4GB (but look at that sexy bandwith) due to first generation HBM restrictions. But everyone reading TPU regularly should be well aware of that.
But if you look at the price of GDDR5, 6GB for a 1060 (TI) doesn't look unrealistic at all and I'd be surprised if it was less tbh.

Edit: typo

 

[RejZoR]

I think 4Gb is the limit on the Fury X, no point adding more if it can't be used.

The fact is, even with 8GB it wouldn't operate any better. Sure HBM was limited to 4GB at that time, but GPU itself doesn't have enough grunt to run more polygons, pixels, textures and effects to make use of more than 4GB. And it's not just because of HBM design. Same goes for GTX 980Ti. I don't think giving it just 4GB would limit it all that much. Maybe it would lose a frame or two here and there, but nothing that would create a huge gap like there is between different series of cards.

 

[P4-630]

With more vram you can use high quality textures in games, even if it's not an high-end GPU.

 

[Vayra86]

The fact is, even with 8GB it wouldn't operate any better. Sure HBM was limited to 4GB at that time, but GPU itself doesn't have enough grunt to run more polygons, pixels, textures and effects to make use of more than 4GB. And it's not just because of HBM design. Same goes for GTX 980Ti. I don't think giving it just 4GB would limit it all that much. Maybe it would lose a frame or two here and there, but nothing that would create a huge gap like there is between different series of cards.

This is not true. A higher amount of VRAM has advantages, especially in todays' console ports with streamed content, and mostly in relation to frame pacing *before* actual FPS, but in some cases it also allows higher FPS. And then there are modded games...

Fury X GPU core is limited by its ROPs, not by its 4GB. The whole reason Fury X falls off in 1080/1440p to a 980ti is because it clocks lower and because of its ROP count. It has absolutely zero relation to VRAM or HBM, nor is the core too weak to drive 4GB. With the 980ti, the core is also well balanced with the VRAM amount.

The only recent cards where VRAM is out of line, are the 8GB R9's.

 

[SK-1]

Fury X GPU core is limited by its ROPs, not by its 4GB. The whole reason Fury X falls off in 1080/1440p to a 980ti is because it clocks lower and because of its ROP count. It has absolutely zero relation to VRAM or HBM, nor is the core too weak to drive 4GB. With the 980ti, the core is also well balanced with the VRAM amount.

Well spoken. Cheers.

 

[RejZoR]

This is not true. A higher amount of VRAM has advantages, especially in todays' console ports with streamed content, and mostly in relation to frame pacing *before* actual FPS, but in some cases it also allows higher FPS. And then there are modded games...

Fury X GPU core is limited by its ROPs, not by its 4GB. The whole reason Fury X falls off in 1080/1440p to a 980ti is because it clocks lower and because of its ROP count. It has absolutely zero relation to VRAM or HBM, nor is the core too weak to drive 4GB. With the 980ti, the core is also well balanced with the VRAM amount.

The only recent cards where VRAM is out of line, are the 8GB R9's.

And you just confirmed exactly what I was saying. If new mid end will be as fast as current high end (which is the most likely scenario), how does 6GB and 8GB make any sense if the old high ends can't utilize it properly? Unless if you expect double GPU performance which is very unlikely thing to happen.

 

[Mussels]

its all about that bus (no treble)

looking at nvidias card history, they've had plenty of odd sizes. 768MB, 1.5GB, 3.5GB (lol).

It was probably a choice of 3GB or 6GB - so they went with 6.

 

[Vayra86]

And you just confirmed exactly what I was saying. If new mid end will be as fast as current high end (which is the most likely scenario), how does 6GB and 8GB make any sense if the old high ends can't utilize it properly? Unless if you expect double GPU performance which is very unlikely thing to happen.

What Mussels said. The 8GB on the R9 3xx is just a doubling of VRAM without actually any work on the bus width. Of course that's out of line. But on the top end cards, this is not the case and thát is what you were saying. Regardless, even when it's out of line, there are cases where it has a niche that makes it useful. Modded games, but also Crossfire. Previously, we saw Kepler 4GB 770's in SLI that definitely made use of the additional 2GB.

At the same time, 10xx Geforce will be a new arch so you never know how powerful that core will really be compared to its VRAM.

I think the real bottom line is, you don't really ever have 'too much' of VRAM. It doesn't hurt performance - but it cán help.

Edit: I might add as well, that 6GB on the 1060ti isn't all that weird either. Even if its core is less powerful than 980ti or 'just as powerful' as a regular 980, it may very well bump into 4GB limitations before it's core bound. And it will definitely run into trouble with just 3GB, at least I do hope it's going to be more grunt than a 780ti... which is generally a tad faster than the current "4GB" 970. The longer I think of it, 6GB isn't odd at all.

 

[RejZoR]

its all about that bus (no treble)

looking at nvidias card history, they've had plenty of odd sizes. 768MB, 1.5GB, 3.5GB (lol).

It was probably a choice of 3GB or 6GB - so they went with 6.

With 512bit bus found on R9-290/390, you don't really need to work more on the bus. The same reason why AMD could afford rebranding it to R9-390. It has such huge bus width it needs no optimizations or changes. Unlike starved GTX's that need to rely on compression. Without it, they'd be totally useless. Giving it 8GB of VRAM was just as excuse so people can't say it's identical card.

 

[Vayra86]

With 512bit bus found on R9-290/390, you don't really need to work more on the bus. The same reason why AMD could afford rebranding it to R9-390. It has such huge bus width it needs no optimizations or changes. Unlike starved GTX's that need to rely on compression. Without it, they'd be totally useless. Giving it 8GB of VRAM was just as excuse so people can't say it's identical card.

Apples and apples... You're applying high end logic to mid-range cards. Not going to work You and I both know mid range is 'mainstream buyer' and that means they buy into bigger numbers. You're also trying to put a negative twist on everything, while the fact is that Nvidia offers a far more efficient arch at the moment than its twice-rebranded ancient GCN counterpart. So it's real great, that 512 bit bus, but it's a massive and costly piece of the die that increases power consumption without offering a tangible performance benefit... And it's not like they don't need to work on it (because they really do), it's a case of 512bit being the max achievable, or feasible limit. That's why we have GDDR5X now... It's not, and it never has been, the VRAM that puts AMD on advantage - you said this yourself.

Back in the heat of the GTX 6xx versus HD7950/7970 war, it was NEVER the bus width that put AMD on advantage. It was a higher core clock. The VRAM was a stubborn motherfucker to OC and barely gave an FPS boost.

Reason AMD can rebrand these cards again, is because they had nothing else on offer, they just kicked their big chips down a tier, that's how they stayed competitive. But Hawaii is still good ol' Hawaii...

 

[Dethroy]

Don't start a catfight about silly VRAM amounts and their usefullness. Everyone will probably agree that VRAM, right now, is an almost irrelevant factor in the fps formula. In other words: VRAM amount is far from being a bottleneck currently.
But if increasing the VRAM amount will add little to the cost of the gpu, then there's really no need to stay conservative, especially now with VR and DX12 around the corner it's probably safe to assume that we will see an increase in VRAM usage in the not too distant future.

 

[RejZoR]

No, I'm appliyng 15 years of experience with hardware and knowing how they shuffle around performance ranges. In 90% of cases, current year high end -> Next year mid end. Current year mid end is next years low end. It's always the current year high end that raises the bar higher for everything from scratch. Even if they don't rembrand them, they scale performance this way. There are rare cases when this pattern goes out the window like in case of HD5800 series from AMD where they basically doubled performance of last generation, but that's a rare anomaly...

 

[SK-1]

I guess I'm one of the few who never associate more friggin vram with higher fps. More V-ram to me is indicative or a greater ability to handle higher resolutions and rendering more textures. More FPS...crank up the clocks.

 

[RejZoR]

Back in my days, more VRAM was even associated with lower performance. Remember GeForce 4 Ti series how models with 128MB of VRAM actually performed worse than those with 64MB (or was it 128 and 256MB?) because the bigger VRAM models had lower speed memory?

 

[Mussels]

Back in my days, more VRAM was even associated with lower performance. Remember GeForce 4 Ti series how models with 128MB of VRAM actually performed worse than those with 64MB (or was it 128 and 256MB?) because the bigger VRAM models had lower speed memory?

its still a thing. I've got a 7750 4GB here with deliciously slow ram, but so much of it!