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

[Malogeek]

There is a 5gb 1060? I've not seen that.

I think it was china market only.

 

[LTC8K6]

I had not seen this before:

https://www.engadget.com/2018/08/15/nvidia-turing-quadro-8k-video-playback/

NVIDIA's Turing GPUs can process 8K video in real time

NVIDIA's latest Quadro RTX GPUs with advanced ray-tracing tech cost between $2,300 and $10,000. Land said that 24fps 8K has been tested on the $6,300 P6000, but added "we are confident that the lowest-end Turing will [also] work well." Interestingly, the tech will eventually allow for 8K playback on current-generation cards like the NVIDIA GeForce GTX 1080 Ti. "This is more a code thing than a hardware thing," said Land.

RED also strongly hinted that the new tech will work on NVIDIA's incoming consumer GPUs, too. "The release of this is tied to the new hardware but everyone on NVIDIA will eventually win," said Land. "The story is only half complete ... there are a slew of new 'consumer' NVIDIA cards coming."

 

[LTC8K6]

There is a 5gb 1060? I've not seen that.

https://www.guru3d.com/news-story/gigabyte-geforce-gtx-1060-5gb-windforce-oc-spotted.html

 

[maddie]

1060 now, or wait two months for 2060 or 2050?

Wondering what the best bang for $300 bones would be.

Used card prices, if you're so inclined, will tank.

 

[LTC8K6]

It's almost pointless to do the smaller cards since they won't really but Turing with the the RT/Tensor HW. So you can just do some rebranding for lower end:

2060 is rebranded 1080 family
2050 is rebranded 1060 family.

Done.

Though I don't believe it will be a short release.

My bet is 2000 series will have a 2 year run, with 7nm die shrinks phased in.

I think the 2060 and 2050 are Turing chips. I think NV is going with an all new lineup.

Well, I hope anyway.

 

[crisium]

RTX 2080 is cutdown GPU...
https://videocardz.com/77369/nvidia-geforce-rtx-2080-ti-features-4352-cuda-cores

2944 CUDA cores RTX 2080? I guess the Quadro gets the non-defective chips and the gaming cards get the scraps. I expect 2304 shaders for the 2070 but I'm prepared for less at this point. And don't forget the rumour of 224-bit 7GB for the 2070.

 

[Karnak]

RTX 2080 is cutdown GPU...
https://videocardz.com/77369/nvidia-geforce-rtx-2080-ti-features-4352-cuda-cores

If true I don't see how the 2080 will outperform/be on pair with the 1080ti. Either VERY high clock rates or there are a lot of changes to the architecture. Or nothing special and it simply won't catch the 1080ti.

Maybe for Nvidia Turing was more about focusing ray tracing instead of only more performance.

 

[n0x1ous]

I wanna see the new reference cooler...all these AIB leaks......

 

[maddie]

If those shader #'s are correct, then the % increase of the 2080Ti model [3584>4352 = 21%], is greater than the non-Ti 2080 [2560>2944 = 15%]. The performance gap increases.

 

[jpiniero]

If true I don't see how the 2080 will outperform/be on pair with the 1080ti. Either VERY high clock rates or there are a lot of changes to the architecture.

Might be a little bit of both. Perhaps the 8% faster and $649 is accurate.

 

[DigDog]

If i survive my current ordeal ill buy a new SSD, buy+install Witcher 3, and run it at 1080 on my 770 GTX. My guess is that it will run smoothly enough that i won't need to upgrade for another generation (or two).

 

[PeterScott]

RTX 2080 is cutdown GPU...
https://videocardz.com/77369/nvidia-geforce-rtx-2080-ti-features-4352-cuda-cores

Which was expected.

The important thing in that link, is that Videocardz is now calling the specs confirmed, instead of rumored.

2944 CUDA cores RTX 2080? I guess the Quadro gets the non-defective chips and the gaming cards get the scraps. I expect 2304 shaders for the 2070 but I'm prepared for less at this point. And don't forget the rumour of 224-bit 7GB for the 2070.

Those are mighty nice scraps. I wonder if it gets a full slate of RT and Tensor HW.

 

[moonbogg]

If i survive my current ordeal ill buy a new SSD, buy+install Witcher 3, and run it at 1080 on my 770 GTX. My guess is that it will run smoothly enough that i won't need to upgrade for another generation (or two).

What ordeal?

 

[happy medium]

How many power pins really has no relevance on how much power the cards use.
My gtx 960 4gb had a 6 pin and a 8 pin.

 

[crisium]

But they tend not to overbuild reference cards, maybe because you want maximum PSU comparability. If reference has 6+8 for the 2080 I'd expect higher power usage than the 1080. They don't want an RX 480 situation if you guys remember that 2 years ago.

 

[sze5003]

How many power pins really has no relevance on how much power the cards use.
My gtx 960 4gb had a 6 pin and a 8 pin.

I wasn't worried about power use just worried that I didn't have the right cables but I should be alright. If I have to move or swap them around it will just be a slight inconvenience.

 

[PeterScott]

But they tend not to overbuild reference cards, maybe because you want maximum PSU comparability. If reference has 6+8 for the 2080 I'd expect higher power usage than the 1080. They don't want an RX 480 situation if you guys remember that 2 years ago.

The 2080 likely has a bigger chip than the 1080ti, along with, and RT and Tensor cores which could all be active at the same time. It isn't unexpected that it would guzzle power more like a 1080Ti than a 1080.

 

[LTC8K6]

RTX 2080 reference is 210W if you believe the leaks...

 

[maddie]

The 2080 likely has a bigger chip than the 1080ti, along with, and RT and Tensor cores which could all be active at the same time. It isn't unexpected that it would guzzle power more like a 1080Ti than a 1080.

You think the Tensor cores would be in use a lot? Doing what?

 

[PeterScott]

You think the Tensor cores would be in use a lot? Doing what?

They are used to denoise Ray Tracing. It's pretty much they only way they can do real time ray tracing with HW of this power. You can't do enough rays to get a clean image, so you must Denoise it.

So if you are going to play a game with Ray tracing, it is going to use tradtional shaders elements, RT HW and Tensor cores to denoise the Raytracing.

 

[happy medium]

The 2080 likely has a bigger chip than the 1080ti, along with, and RT and Tensor cores which could all be active at the same time. It isn't unexpected that it would guzzle power more like a 1080Ti than a 1080.

I read somewhere that its 418mm.

 

[PeterScott]

I read somewhere that its 418mm.

I read guesses of that.

But it has about 66% of the units of the 754 mm2 monster which would be closer to 500mm2.

 

[Head1985]

2944 CUDA cores RTX 2080? I guess the Quadro gets the non-defective chips and the gaming cards get the scraps. I expect 2304 shaders for the 2070 but I'm prepared for less at this point. And don't forget the rumour of 224-bit 7GB for the 2070.

It looks more and more like adoredTV leak.
RTX2080-cutdown GPU
RTX2070-some crap x06 GPU and not cutdown RTX2080...

 

[neblogai]

I read guesses of that.

But it has about 66% of the units of the 754 mm2 monster which would be closer to 500mm2.

I did not watch the nVidia Siggraph, so I'm not sure about that info that RTX8000 die is to have 72 SMs (with 64 cuda cores each). Was that official? If so- and RTX5000 has different core arrangement (24SMs x 128 cuda cores)- then this arrangement probably less place than one on the big die.
Also- that would mean, that nVidia has built custom, different GPCs for each: TU102, 104 and 106, and maybe even more, if some smallest (notebook) dies are made on 7nm.

 

[PeterScott]

I did not watch the nVidia Siggraph, so I'm not sure about that info that RTX8000 die is to have 72 SMs (with 64 cuda cores each). Was that official? If so- and RTX5000 has different core arrangement (24SMs x 128 cuda cores)- then this arrangement probably less place than one on the big die.
Also- that would mean, that nVidia has built custom, different GPCs for each: TU102, 104 and 106, and maybe even more, if some smallest (notebook) dies are made on 7nm.

Not sure why you are assuming different SM arrangements, or why one or the other would take up significantly different amounts of die space for the same CUDA count when they deliver the same function in the same generation.

3072/4608 CUDA = 2/3rds (~67%) count, and most likely requires ~67% of the die area.
256/384 Memory bus = ~67%, so about ~67% of the memory controllers, and similarly 67% of the die space for memory controllers.
384/576 Tensor cores = ~67% of tensor cores.

In almost every way this is 2/3(67%) of RX8000.

754*.67 = ~500 mm2.

The RTX5000 (RTX 2080) - TU104 is essentially 2/3 of RTX8000 (RTX 2080Ti) (TU102?) functional units, so it should be about 2/3 of the die area, so about ~500mm2 area, give or take minor amounts.

I have yet to see any reason why RTX5000 (RTX 2080) - TU104 die would be significantly smaller than 500mm2 given the size of the bigger brother, and the unit counts on each.