These are the leftovers (not passing the binning process) of the full dies I guess. 7nm must still have yield issues.
I must say that marketing wise, AMD seem to have all the bloody morons there. What a nightmare their naming options.
This doesn't point to yield or naming problems at all if the primary intention here is just a "one off" product set for Vega at 7nm.
The name isn't an issue if there will only be exactly one of these 7nm Vegas brought to the consumer side. So they don't need a 7xx series or something with multiple model numbers if there is just one.
Vega 56 , Vega 64 , Vega VII
until they are all retired later, probably this time next year. ( maybe earlier if things get ahead of schedule).
I don't think there is a huge yield issue here. What they need is more affordable cards. The MI50 and MI60 will sell but that is a limited volume market due to the prices. At some point that will likely either plateau or even go backwards. The lower priced Vega VII will help fill out the volume when the MI's versions peak out. There is probably is a threshold of wafer starts they need to hit to have 'priority' in the queue at TMSC and the packaging ( combo of GPU and HBMv2 assembly ). The MI's have fat profits but not volumes.
If the MI cards happen to do completely miss expectations then perhaps they would need a "Vega VII Plus" later in the year , but I suspect they pretty have that as a "Plan C" option. [ Plan B being drop the effective prices on the MI series a bit to generation some more demand. ] Similar if Navi has some roll out hiccups later in the year at larger sizes. The other major issue is the uptick on the software associated with the MI cards. If that stalls then the MI card sales will stall.
We'll see when the final stats come out but if the Vega VII is being run at higher clocks and higher TDP then this isn't binning and selling rejects. (current specs put the clocks higher. The TDP is probably up also) That's binning stuff for peaker, 'hot rod' loads. Typically those aren't rejects. Certainly there will be a few 1-4 CU failure 'rejects' thrown into the pile but that probably won't be the core basis of the binning. Same for infinity link failure 'rejects'; some but not core volume.
The MI's will be binning for more sustained loads characteristics. ( also can pick up DisplayPort subsystem bugs and ones that don't clock all the way up. The Vega VII clocks could have been the targets and but are much more stable at the MI's rates. )
They either copy competitor's naming schemes (sure, it's so much easier for us all) or come up with nonsense one.
if this isn't a broad 'line up' then that may come with Navi. Which is the point of their 2019 offerings. Just not until more so the second half than the first half.
Radeon VII (seven). Seven for 7nm? Wow, that's creative, and tied to the process node (sort of).
More than several months ago they said they said they weren't doing anything in Consumer space for Vega 7nm. I think they are primarily jumping into a pricing umbrella 'hole' that Nvidia is holding open to them. Much of the NVidia RTX was thrown at non traditional graphics ( not at generalize computation or traditional graphics computation pipeline ). That opened a window to be closer with a placeholder for a substantively longer period of 2019. That's fine.
If Nvidia had done something to completely leave the 7nm Vega in the dust (on pricing and performance) I suspect AMD may not have done a Vega VII. AMD would have rolled out with more affordable MI cards and avoided the consumer side for additional volume. So Vega 7nm would have been MI Instinct only exactly how they were forecasting a year ago.
The other factor is the bottom falling out of the crypto mining craze. There is going to be are larger number of used top end cards floating around out there. AMD needs a stopgap product that is incrementally faster than all that lower priced used stuff out there relatively soon. [ At this point it is far more painful for them to float the Vega 56/64 for 9-10 months of 2019. ] The Vega VII volume should help the MI card's margins because can spread the costs out over a wider base.
P.S. To loop this back closer to the topic of the Mac Pro ..... I have doubts this makes much of a difference to Apple's plans. They have been so "asleep at the wheel" with Rip van Winkle tactics that AMD doing a 'shift' on Vega 7nm has a very good chance of going over Apple's head if these was done in May/June 2018-Nov 2018 planning/execution by AMD . If Apple was asking for Vega 7nm for Mac Pro / iMac Pro and AMD was resisting them a bit then it will work out, but Apple's ability to keep up to fluid changes is highly suspect.
[doublepost=1547246370][/doublepost]
Preliminary info on the Vega VII suggest it to be the same MI60 with video output, no PCIe4 no Infinity Link, while seems VegaVII to have a custom xGMI bridge inferior to the one on MI60, or at least coming later as the MI60 market gets cold (driven by FP64 performance, a small niche)
No, the current specs match the MI50 , not MI60 . Specs for MI50:
"...
GPU Specifications
GPU Architecture Vega20
Lithography TSMC 7nm FinFET
Stream Processors 3840
Compute Units 60 ..."
https://www.amd.com/en/products/professional-graphics/instinct-mi50
For the MI series the "50" and "60" is
not the number of CU units. 50 is just a smaller (implicitly more affordable) number than 60. It isn't coupled to the GPU core count.
The MI60 only has 4 more CUs. It has a bucket load more HBMv2 RAM. ( 32GB versus 16GB). That's a big difference in terms of cost (harder to make in addition to more capacity) and on large data sets workloads ( more data kept local).
What AMD is doing to taking the more affordable MI50 and selling it at an even more affordable price in the Vega VII by lopping off some functionality (to make more unattractive to machine room deployments) and up-clocking it a bit to run a "hot rod" state by default.
