Nvidia Turing: Everything you need to know
(Update: June 27, 2018): With invites now reportedly going out for an Nvidia event during Gamescom 2018, and a next-generation GPU event mooted for an upcoming ‘Hot Chips’ conference, it’s increasingly looking like we might finally see Nvidia’s next graphics card’s announced next month. Read on for everything we know so far.
Nvidia Turing is set to be the graphics card giant’s consumer successor to its Pascal architecture, but what on Earth does that mean and why are we even talking about it?
Given that its big GPU rival AMD has just shown its hand with its high-end gaming Vega architecture, there’s no time to waste in looking at what might be coming next. First, let’s look at what Turing is before assessing its role in the world of GPUs this year.
What is Nvidia Turing?
Turing is the code name for Nvidia’s next generation of graphics architecture. Graphics architecture is another way of describing a chip’s design; the template on which the various different products are based. Turing is the consumer successor to Pascal, which was the codename behind 10-series GPUs ranging from the bottom-spec GTX 1050 all the way up to the GTX 1080 Ti. In contrast, Turing is expected to be the platform behind rumoured cards such as the GTX 1180 (although it could end up being branded the GTX 2080).
I thought Nvidia Volta was next?
We did too! Ever since Nvidia released its massive V100 servers running Volta chips we’ve assumed that Volta would go on to be the name of the consumer-focussed graphics cards.
However, as time has gone on we’ve seen more and more clues that suggest the cards are going to end up being named ‘Turing’, suggesting that Nvidia is going to use different naming schemes for its professional and consumer hardware.
After all, the specs posted on the TechPowerUP database used the codename Turing, and Nvidia’s Twitter account recently wished a happy birthday to Alan Turing, the father of the modern computer.
It’s not an official confirmation, but for now we’re placing a tentative bet that the cards end up being called Turing rather than Volta.
Nvidia Turing release date
Officially, Nvidia wants us to think that we’ve got a while to wait. Speaking at Computex 2018, the company’s CEO Jensen Huang said that Nvidia Turing’s release will be “a long time from now.”
But more recently it’s starting to seem like we’ll actually see an announcement in August according to The Verge. The Hot Chips conference that’s due to take place then recently announced and then retracted a talk from Nvidia titled ‘Nvidia’s Next Generation Mainstream GPU’, and Nvidia is also reportedly sending out invites for a PC-related Gamescom event.
Combine this with the fact that a Lenovo product manager mentioned the existence of the GTX 11 Series during an E3 stream an it’s starting to look like “a long time from now” actually translates to, “actually two months from now, but please don’t stop buying our current GPUs in the mean time.”
Nvidia Turing specs
This is where things get both technical and woolly. We’ve had a number of different reports from a number of different locations claiming to know what features the new cards will be packing.
Most recently, a report from Tom’s Hardware suggested that Nvidia’s Turing cards will be packing the same RTX technology that Nvidia has recently used in the professional-grade Titav V and Quadro GV100.
You can read a full explanation of what this technology involves on Nvidia’s website, but in short the technology allows for real-time ray-tracing, which is a new was of simulating light that looks absolutely incredible. You can check out the tech in a video below.
The new generation of Nvidia Turing GPUs should have significantly higher clock-speeds in excess of 2GHz. A fourth generation of GPU Boost should allow clocks to be pushed ever higher under the right cooling conditions.
Finally, the recent report also suggested Nvidia Turing will be making use of the upgraded HDMI 2.1 standard. There are a number of new improvements expected with the standard, chief amongst them being that it will finally support 4K content at over 60Hz (up to a maximum of 120Hz).
However, perhaps more interesting is HDMI 2.1’s introduction of Variable Refresh Rate. Nvidia’s GPUs already support their own version of the technology called G-Sync, but at the moment this is a proprietary technology that only works over the DisplayPort connector. HDMI 2.1 support would in theory mean that variable refresh rates would also work over HDMI, and wouldn’t require users to pay the price premium for an expensive G-Sync monitor.
Recently leaked photographs of what appears to be an Nvidia Turing engineering sample suggest that the cards will use the next-generation GDDR6 VRAM, which should allow for higher clock-speeds and more memory bandwidth.
So there’s a lot of potential tech under the hood, but if you want exact specifications we have to turn to the TechPowerUp GPU database which recently saw some additions that claim to show what the successor to the GTX 1080, the Nvidia Turing GTX 1180 will be capable of.
According to these specs, the 1180 will have 3584 CUDA cores (a 40% increase over the GTX 1080), 224 TMUs (another 40% increase) and 16GB of GDDR6 RAM (a 100% increase).
Overall, this means the new card will have 13 teraflops (tflops) of computing power, which not only faster than the 1080’s 8.7 tflops, but is even faster than the Titan Xp’s 12.
The specs also suggest that the new card will be based on a 12nm fabrication process, rather than the smaller 10nm fabrication process that was originally rumoured.
The smaller the process size, the more transistors (the yes/no gates that do all the work) you can fit on a piece of silicon, which means you effectively get more powerful components without using more energy. We warned you this would be technical.
By way of comparison, 10nm is the same process size as Intel’s Cannon Lake processors.
Then again, perhaps this confusion could be to blame for the two different names floating around, Volta and Turing, which could end up having different fabrication processes. Volta could be the 12nm version, while Nvidia’s Turing could step it down to 10nm.
Check out how all the specs compare in the table below:
|GTX 1080||GTX 1180||Increase|
|RAM||8GB (GDDR5)||16GB (GDDR6)||100%|
|FP32||8.7 TFLOPs||13 TFLOPs||49%|