'

AMD reveals their 7nm Vega Instinct MI60 and MI50 graphics cards

The world's first 7nm GPU that offers the more FP64 performance than any other graphics card

AMD reveals their 7nm Vega Instinct MI60 and MI50 graphics cards

AMD reveals their 7nm Vega Instinct MI60 and MI50 graphics cards

AMD has officially revealed their 7nm Radeon Instinct MI60 and MI50 graphics cards, the world's first 7nm graphics cards. Both of these GPUs are based on AMD's Vega architecture, though it is clear that AMD has made some additional changes under the hood to offer increased performance levels in select workloads. 

For starters, AMD has worked to deliver the industry's highest levels of FP64 compute on a single graphics card, reaching 7.4 TFLOPS on their MI60 graphics card, something that is unmatched within the industry, surpassing even Nvidia's Turing-based Quadro RTX 8000. On top of that, AMD's Radeon Instinct MI50 and MI60 graphics cards also offer support for PCIe 4.0, a first for the Datacenter GPU market. 

Thanks to AMD's use of TSMC's 7nm process node, the company has been able to condense their Vega GPU design to a die size of 331mm squared, while also adding two additional memory controllers to support four HBM2 dies. This allows the Radeon Instinct 60 to support up to 32GB of HBM2 memory.   

In terms of core clock speeds, 7nm Vega offers a significant boost over AMD's 14nm design, boosting the graphic's card's peak clock speeds from 1500MHz in the MI25 to 1800MHz on the MI60, an increase of 20%. This showcases the performance benefits of TSMC's 7nm process node, especially after considering the fact that the MI60 has the same 300W TDP as the MI25, while also requiring 8-pin + 6-pin power configuration. 

AMD has made it clear that their 7nm Vega graphics cards are not designed for gaming applications, instead, acting as a way for the company to enter the lucrative server/datecenter and machine learning markets. 
 

 Radeon Instinct MI60Radeon Instinct MI50Radeon Instinct MI25
MicroarchitectureVegaVegaVega
Lithography7nm7nm14nm
Compute Units646064
Stream Processors409638404096
Peak Engine Clock1800MHz1746MHz1500MHz
Peak FP64 TFLOPS7.4 TFLOPS6.7 TFLOPS 
Peak FP32 TFLOPS14.7TFLOPS13.4 TFLOPS12.29 TFLOPS
Peak FP16 TFLOPS29.5 TFLOPS26.8 TFLOPS24.6 TFLOPS
Peak INT8 TFLOPS58.9TFLOPS53.6 TFLOPS 
Memory TypeHBM2HBM2HBM2
Memory Capacity32GB16GB16GB
Memory Interface4096-bit4096-bit2048-bit
Memory Clock Speed1,000MHz1,000MHz945MHz
Memory Bandwidth1,024 GB/s1,024 GB/s484 GB/s
ECC SupportYes (Full Chip)Yes (Full Chip)Yes
TDP300W300W300W
PCIe SupportPCIe 4.0PCIe 4.0PCIe 3.0
Power8-pin + 6-pin8-pin + 6-pin8-pin + 8-pin

 

When looking at AMD's 7nm Vega graphics cards, it is clear that the company sees a future in AI compute, though at this time it is unknown whether or not the company plans to bring their AI performance into the gaming market, mirroring Nvidia's moves with Turing. With the 20% clock speed gains through their use of 7nm technology, I'm excited to see what the company will be able to achieve when they eventually decide to bring 7nm to gamers, especially when combined with a next-generation graphics architecture like Navi.

Sadly AMD's announcements today didn't reveal any of AMD's future graphics card roadmap, or hint at the release of any new Radeon graphics hardware for the consumer market. 

You can join the discussion on AMD's 7nm Vega Instinct MI60 and MI50 graphics cards on the OC3D Forums

«Prev 1 Next»

Most Recent Comments

06-11-2018, 19:47:07

NeverBackDown
Should be pretty good. Makes you wonder if Deep Learning/Machine Learning/AI will be applied to games like Nvidia is doing with super computers and DLSS with this heavy focus on those markets.Quote

07-11-2018, 08:28:20

Avet
AMD is good in raw compute power. But their architecture just doesn't go nice with games.Quote

07-11-2018, 08:32:49

g0ggles1994
Quote:
Originally Posted by Avet View Post
AMD is good in raw compute power. But their architecture just doesn't go nice with games.
Good job this isn't for gamers then eh? Quote

07-11-2018, 09:55:11

Avet
Quote:
Originally Posted by g0ggles1994 View Post
Good job this isn't for gamers then eh?
Well... It does eventually roll down to gamers. All consumer tech is mostly downgraded server stuff. If the base architecture doesn't do specific workloads, you can't optimize much. Same as CPU lineup.Quote

07-11-2018, 11:19:28

tgrech
Vega isn't particularly heavily focussed at gaming(It did seem to be at some point in development but then the AI boom happened and the gaming-related features like DSBR & Primitive shaders seemed to get mostly shelved), but it's hard to argue the same applies for all of AMDs chips & architectures. Traditional GCN started out in a similar position but by later revisions (Consoles & Polaris) it was pretty well optimised for gaming & generally managed better game FPS vs raw compute power (FLOPS & similar GPGPU measures) than any architecture before or since.

Navi looks like it will do the same. While the general architectures of the enterprise grade chips do trickle down, generally by the time they reach small-die variants that have much more of a life in mid-end gaming cards than enterprise cards they've been gutted of many of their enterprise grade features (Double precision compute, EMC support, ect) and had their resource allocation tweaked to better suit gaming purpose, which brings along the improved efficiency at the task through minimised "dead silicon" and better fed units. NVidia's RTX cards aren't particularly efficient gaming cards either for this reason, and the same could be seen with many of the Titan variants that focussed more on compute performance.Quote
Reply
x

Register for the OC3D Newsletter

Subscribing to the OC3D newsletter will keep you up-to-date on the latest technology reviews, competitions and goings-on at Overclock3D. We won't share your email address with ANYONE, and we will only email you with updates on site news, reviews, and competitions and you can unsubscribe easily at any time.

Simply enter your name and email address into the box below and be sure to click on the links in the confirmation emails that will arrive in your e-mail shortly after to complete the registration.

If you run into any problems, just drop us a message on the forums.