Nvidia Volta Megathread

[renz496]

nvidia prototype board with GDDR6 on board:

https://videocardz.com/76659/nvidias-next-gen-graphics-card-with-gddr6-smiles-for-camera

 

[-Fran-]

Uhm... 3 power connectors... Proto-board indeed.

I wonder if that is going to be 386bit or 256bit for GDDR6. Still, interesting to see one in the wild already.

Cheers!

 

[goldstone77]

First Nvidia GTX 1180 PCB prototype leaked with 12 GB 384-bit GDDR6 VRAM

According to insider sources, Nvidia's desktop GPU team is already testing GTX 11xx prototype PCBs and the final specs should be announced in mid-July. One of these prototype PCBs was already leaked on reddit. The board does not feature the GPU, however the spec sheet includes 12 GB of GDDR6 VRAM with a 384-bit bus, three 8-pin power connectors and a second gen NVLink.

Apparently, the card uses Micron memory chips, and judging by Micron’s own spec sheets, the 12 GB installed on the prototype card should yield around 672 GB/s total bandwidth, which would be a tad faster than what Titan V can output right now. Micron’s specs also allow for 14/16 GB memory options, so Nvidia might opt to include more memory for the Ti versions later on.

https://www.notebookcheck.net/First-Nvidia-GTX-1180-PCB-prototype-leaked-with-12-GB-384-bit-GDDR6-VRAM.311656.0.html?utm_source=dlvr.it&utm_medium=twitter

 

[goldstone77]

THE ART OF SUPERCOMPUTING WAR
June 25, 2018 Timothy Prickett Morgan

https://www.nextplatform.com/2018/06/25/the-art-of-supercomputing-war/

 

[renz496]

When it comes to accelerator nvidia no doubt lead the pack. for supercomputing FP64 support is a must. AMD last GPU with true FP64 support was Hawaii. There is still no clear indication about their next FP64 GPU. it seems they want to focus more on AI related stuff but even that start being fully dominated by nvidia. Some supercomputer might employ their Epyc CPU instead of intel xeon but when it comes to accelerator i still think majority of HPC client will use nvidia tesla instead of AMD Firepro.

 

[YoAndy]

I hope they have a problem
Nvidia May Have a GPU Inventory Problem

There are reports that Nvidia might have a GPU inventory problem on its hands. If true, it would explain why Jen-Hsun told the press that new GeForce cards wouldn’t arrive for a “long time,” when other sources we’ve spoken to have implied a much quicker release time frame within the next few months. As always, stories about rumors, inventory levels, and unannounced products should be taken with lots and lots of salt. Say, a GeForce GTX 2080 worth?

https://www.extremetech.com/gaming/272173-nvidia-may-have-a-gpu-inventory-problem?utm_medium=notification

 

[goldstone77]

Judging by how cheap video cards are getting, there is an expectation of Nvidia's next generation much sooner than we were lead to believe. 1180 has been flaunted around in rumors.

 

[renz496]

I hope they have a problem
Nvidia May Have a GPU Inventory Problem

There are reports that Nvidia might have a GPU inventory problem on its hands. If true, it would explain why Jen-Hsun told the press that new GeForce cards wouldn’t arrive for a “long time,” when other sources we’ve spoken to have implied a much quicker release time frame within the next few months. As always, stories about rumors, inventory levels, and unannounced products should be taken with lots and lots of salt. Say, a GeForce GTX 2080 worth?

https://www.extremetech.com/gaming/272173-nvidia-may-have-a-gpu-inventory-problem?utm_medium=notification

nvidia might have a bit problem on this front but i don't think it was really an issue to them.

 

[goldstone77]

Friday June 29, 2018
NVIDIA 11 Series Coming to Lenovo Desktops Fall 2018

DSOGaming is reporting that at E3 a lenovo representative told YouTuber "Brainbean" that the NVIDIA 11 series, up to the 1180, would be in the Lenovo Legion desktops in Fall 2018. This is the first we're hearing the name, and timeframe of the upcoming cards.

https://www.hardocp.com/news/2018/06/29/nvidia_11_series_coming_to_lenovo_desktops_fall_2018

 

[goldstone77]

NVIDIA "GT104" Based GeForce GTX 1180 Surfaces on Vietnamese Stores
by btarunr Monday, July 2nd 2018 05:54

Among the specifications of the GeForce GTX 1180 are 3,584 CUDA cores, and 16 GB of GDDR6 memory across a 256-bit wide memory interface. The memory is clocked at 14 GHz (GDDR6-effective), which works out to 409.6 GB/s of memory bandwidth. Pre-launch prices, just like most specifications, tend to be bovine excrement, which in this case converts to a little over USD $1,500, and isn't really relevant. What is, however, interesting is the availability date of September 28.

https://www.techpowerup.com/245657/nvidia-gt104-based-geforce-gtx-1180-surfaces-on-vietnamese-stores

 

[-Fran-]

If AMD doesn't have a proper response to the 1180, I might just get an nVidia again (been a long time since the 670!). That time-frame is not too bad for me, to be honest. Also, I'm hopeful miners won't have GDDR6 figured out for the first round of video cards that come out with it! xD

Cheers!

 

[goldstone77]

If AMD doesn't have a proper response to the 1180, I might just get an nVidia again (been a long time since the 670!). That time-frame is not too bad for me, to be honest. Also, I'm hopeful miners won't have GDDR6 figured out for the first round of video cards that come out with it! xD

Cheers!

Nvidia GPUs are best at farming Zcash, which overclocking the GPU is better vs. memory while under volting, so they could, theoretically, go straight into service. AMD GPUs is memory sensitive while farming Ethereum. But the price of BTC will dictate mining interest. As long as BTC price is low you won't have to worry about miners. There have been reports of small mining firms folding, and larger ones are not upgrading or slowly upgrading.

 

[goldstone77]

The NVIDIA Titan V Deep Learning Deep Dive: It's All About The Tensor Cores
by Nate Oh on July 3, 2018 10:15 AM EST

https://www.anandtech.com/show/12673/titan-v-deep-learning-deep-dive

 

[renz496]

this is a real deal:

https://videocardz.com/76953/nvidia-geforce-gtx-2080-1180-pcb-has-been-leaked

 

[-Fran-]

Interesting to see the nV-Link headers there. There might be more this time around than just "SLI is back". I would think that might be for Quadros as well, given the nv-links it has.

Cheers!

 

[renz496]

it might end up being something else together instead of another SLI tech. something that does not need AFR or SFR like how our multi GPU being implemented all this time. nor it need DX12 specifically to function 😛

btw VCZ being barrage with stuff regarding new GeForce right now....

 

[-Fran-]

I think there are 2 main takeaways for the "Volta generation" from nVidia and they won't be pure "GPU" enhancements. There will be that as well, but my take is nVLink for consumer and hardware schedulers for nVidia. Those two are really interesting to me at least, even if "brute performance" doesn't go up two-fold.

Cheers!

Heh, one of the things I was told seems to be true... Let's see the one about hardware schedulers with the uArch.

Cheers! 😛

 

[goldstone77]

[video="https://www.youtube.com/watch?v=ms7HQ7rckpA&feature=em-uploademail"][/video]

 

[goldstone77]

Nvidia Live NOW@SIGGRAPH2018
https://www.ustream.tv/NVIDIA?linkId=100000003235371

 

[goldstone77]

Nvidia Turing architecture arrives in high-end Quadro RTX GPUs
Even if you don't care about workstations, it strongly hints that the Turing-based GeForce cards are imminent.
BY
LORI GRUNIN
AUGUST 13, 2018 4:31 PM PDT

https://www.cnet.com/news/nvidia-turing-architecture-arrives-in-high-end-quadro-rtx-gpus/?ftag=COS-05-10aaa0b&linkId=55513966

 

[goldstone77]

NVIDIA Quadro RTX 5000, 6000, 8000 & Turing
https://videocardz.com/77260/nvidia-announces-quadro-rtx-6000-ray-tracing-gpu

 

[renz496]

so the architecture really are called Turing. before this we thought only the tensor cores will be handling the ray tracing stuff. it seems nvidia coming up with brand new "core" for it (RT) to handle ray trace and rasterization at the same time (combined with compute and AI). the image they show at GTC using 4 titan V to ray trace the short star wars trailer. now they just need one GPU to do it (Turing).

another thing is VRAM pooling in multi GPU system. before this it was like this:

4+4 = 4

now with NVlink it is possible to make it like this:

4+4 = 8

this is for professional use though (Quadro). not sure if nvidia will introduce new type of SLI capability with this.

....another G80 moment?

 

[renz496]

btw speculation time. so we got:

RTX5000
RTX6000
RTX8000

RTX5000 probably based on GT106 that will be powering the new 2060
RTX6000 probably based on GT104 that will be powering 2080 and 2070
RTX8000 probably based on GT100 that will going to power 2080ti.

so the "breakthrough performance" most likely referring to ray tracing performance. but in current games the leap most likely still fall within what majority of us had expected.

 

[goldstone77]

 

[King Dranzer]

btw speculation time. so we got:

RTX5000
RTX6000
RTX8000

RTX5000 probably based on GT106 that will be powering the new 2060
RTX6000 probably based on GT104 that will be powering 2080 and 2070
RTX8000 probably based on GT100 that will going to power 2080ti.

so the "breakthrough performance" most likely referring to ray tracing performance. but in current games the leap most likely still fall within what majority of us had expected.

Quadro RTX5000 will be workstation equal to Geforce RTX2080
Quadro RTX6000 ~ RTX8000 will be workstations equals to Geforce RTX2080Ti

Both RTX6000 and RTX8000 will perform similarly. Only benefit of RTX8000 is that it packs in double the Memory of RTX6000 which benefits it for larger workload.

RTX2070 will be a level below RTX2080 and cannot be paired. GTX2060 will not even support Ray tracing.