TR Forums

Hey, don't shoot the messenger; he's not under a 'deny' NDA....

https://wccftech.com/amd-project-canis- ... nture/amp/

Anyone who's allowed to speak know anything (other than the date...)? And has anyone sent the paramedics to administer CPR to Chuckula?

Yeah... What day is it today?

Would it be... The first of the month?
Where the month is April?

But tell ya what: If these things really launch this year, then I'll buy you one! Price is no object!

Process 10nm ++ (Intel)
Architecture Canis 10
Stream Processors 10240
Core Clock (Boost) 1800 MHz (2441 MHz)
Memory 32GB HBM4
Bandwidth 4 TB/s
Peak Performance 50 TFLOPs
Die Size 400mm2 (x2)
GPU Type MCM
Interconnect Unknown (TSV based)
Feature Level Direct3D 12_2
TDP 350W
Power Connectors 3x 8-pin
Expected MSRP $999
Launch 2H 2018

lol c'mon dpaus.

I just can NOT believe the skepticism you heathens are showing.

Except Chuckula: I accept your offer

dpaus wrote:

I just can NOT believe the skepticism you heathens are showing.

Except Chuckula: I accept your offer

Well it is April Fool's Day today, so...

Obviously an April Fool's joke but the general theme of what is coming in the future isn't too off. Splitting up a monolithic die across an interposer will likely happen within a few generations. nVidia has released a paper on that very subject last year and AMD has been dropping hints that it is exploring the same concept. In all reality this will probably be how the future midrange, high-end and crazy-stupid-but-my-ego-wants-one enthusiast/professional class chips are segmented: just how many dies are stitched together. Low end gets one die but harvested with fewer units enabled, midrange gets a fully functional die, high end two dies and top-of-the-line-OMG-I-want-it gets four compute dies.

Utilizing Intel fabs is unlikely but that is what everyone said about Kaby Lake-G. Just enough doubt to make it a good inclusion for April Fools. Other things that are off is that it'll use HBM4 where as HBM3 would be more appropriate given the timeline. Also future GPUs should be transitioning over to PCIe 4.0 in this next round.