AMD Has 7nm GPUs Running in Labs, Sampling Later This Year

Author Photo
Apr 26, 2018

AMD’s President and CEO Dr. Lisa Su made an exciting revelation yesterday during the company’s 1st quarter 2018 earnings conference call indicating that 7nm GPUs are much closer than we might have thought.

Dr. Su announced that AMD has working 7nm GPUs running in the Radeon Technologies Group labs right now. Specifically, 7nm Vega which is expected to begin sampling later this year.

amd-radeon-feature-wccftechRelated Rumor: AMD Navi RX 3080 to Launch Mid 2019 & Take On RTX 2080

AMD President & CEO Dr. Lisa Su – Q1 2018 Earnings Report

“I’m also happy to report that our next-generation 7nm Radeon Instinct product, optimized for machine learning workloads, is running in our labs and we remain on track to provide samples to customers later this year.”

7nm Vega in the House

Earlier this year at CES AMD announced that it’s working on a 7nm Vega based graphics processor optimized for Artificial Intelligence (AI) and Machine Learning workloads. Dr. Su’s remarks yesterday evening confirm that the company has gotten silicon back from the fab and is currently testing the chips.

later in a question and answer segment of the earnings call, Su also confirmed that 7nm Vega is manufactured at TSMC, rather than Globalfoundries. A decision that was likely influenced by the maturity of TSMC’s 7nm manufacturing process.

amd-radeon-feature-wccftechRelated AMD Navi RX 3080, 3070 & 3060 Specs & Prices Leaked – RTX 2070, GTX 1070 & GTX 1060 Challengers at $249, $199 & $129

AMD Q1 2018 Earnings Call

Timothy Arcuri – UBS Securities
“I think you said that the 7nm product is in the lab and it’s going to launch later this year. That’s the product at TSMC, correct? And I’m just wondering on that front do you feel comfortable that you can get capacity from that vendor?”

AMD President & CEO Dr. Lisa Su

“Our foundry strategy is to use both TSMC and Globalfoundries on the first 7nm product, we are using TSMC for that product and we have a very strong relationship with them. And so, we do see good momentum on it from what we see and I’m not concerned about capacity.”

AMD’s Infamous 7nm Vega 20

Earlier this month we spotted a peculiar piece of code in a Linux patch that makes reference to Vega 20 by name. The patch introduced support for more than 50 new Vega specific hardware-level features and we stated that this indicates that AMD could be approaching final post-silicon testing and validation of Vega 20. Thanks to Su’s announcement yesterday, it turns out that’s exactly right.

We brought you news of Vega 20 in as early as January of 2017. So we have known about AMD’s plans to introduce a high performance Vega chip based on the 7nm process for quite a while. Confirmation of such a project, however, did not come until exactly a year later at this year’s CES back in January.

Although the company still does not refer to its new 7nm Vega chip as Vega 20, previously leaked internal AMD slides clearly describe a 7nm based “Vega 20” GPU that’s specifically designed for AI and intended for market entry in 2018.

Vega 20 differs from Vega 10 in several key aspects. In that it’s designed to deliver 8 times the double precision compute performance, double the memory interface width and support for up to 4 HBM2 stacks for a total of 32GB of HBM2 vRAM and 1TB/s of memory bandwidth.

Vega 20 will be AMD’s first chip designed specifically for AI and machine learning workloads and it’s set to compete with NVIDIA’s Volta based accelerators, including the Quadro GV100 and Titan V.

AMD Polaris 10, Vega 10 and Vega 20 GPUs

GPU Polaris 10 Vega 10 Vega 20
Year 2016 2017 2018
Graphics Card RX 480, 470 Vega FE, 64, 56 TBA
Process 14LPP 14LPP 7nm
Transistors In Billions 5.7 12 TBA
Stream Processors up to 2304 Up to 4096 TBA
Performance 5.8 TFLOPS Up to 13 TFLOPS 20 TFLOPS
TDP 150W Up to 290W TBA
Memory 8GB GDDR5 Up to 16GB HBM2 Up to 32GB HBM2
Speed 2000 MHz
(8Gbps )
950 MHz 1250 MHz
Memory Bus 256bit 2048bit 4096bit
PCI Express 3.0 3.0 TBA
Bandwidth 256 GB/s 484 GB/s 1.28 Terabyte/s