Sapphire lists its Radeon VII 16G HBM2 graphics card

[kalniel]

Please correct me if I've missed something but I'm not seeing references as to why this card has fewer stream processors, compute units and texture mapping units in comparison to the Vega 64. Surely if they remained the same or even increased them it would improve the performance further and maybe take better advantage of the 16 GB of HBM2 RAM?

I am aware it's the professional workstation card but surely the same would apply? Unless it really comes down to cost cutting and expediency as the professional card just needs more RAM?

You mean compared to M160 right? Vega 64 is a previous generation so can't be compared like for like. And the answer is the same reason that nVidia do exactly the same for the 2800ti vs the Titan RTX - the price for workstation cards is much higher, but if you allow for a few mistakes in the fab process and scavenge not quite the full compliment of stream processors then you can still end up with a fast chip, just one that doesn't make the grade for a M160 or titan RTX.

[DanceswithUnix]

Please correct me if I've missed something but I'm not seeing references as to why this card has fewer stream processors, compute units and texture mapping units in comparison to the Vega 64. Surely if they remained the same or even increased them it would improve the performance further and maybe take better advantage of the 16 GB of HBM2 RAM?

I am aware it's the professional workstation card but surely the same would apply? Unless it really comes down to cost cutting and expediency as the professional card just needs more RAM?

Rumour has it that AMD's current generation can't use more than 64 CUs, hence that has been the limit since the Fury cards.
Rumour also has it that Navi should remove that limit.

[Vanqueur]

Rumour has it that AMD's current generation can't use more than 64 CUs, hence that has been the limit since the Fury cards.
Rumour also has it that Navi should remove that limit.

Thanks, that is what I wanted to know.

You mention the rumour around AMD not being able to use more than 64 CUs on current gen but this new card will only have 60 CUs, 3840 SPs, and 240 TMUs.
Both Vega 64 and the Fury R9 have 64 CUs, 4096 SPs and 256 TMUs.

Unless what Kalniel mentioned about binning is in play?

[Ozaron]

Thanks, that is what I wanted to know.

You mention the rumour around AMD not being able to use more than 64 CUs on current gen but this new card will only have 60 CUs, 3840 SPs, and 240 TMUs.
Both Vega 64 and the Fury R9 have 64 CUs, 4096 SPs and 256 TMUs.

Unless what Kalniel mentioned about binning is in play?

GPU chips which can reliably, safely run all 64 CUs in a production environment become professional cards, which are sold for far more.

Radeon VII uses 60 CUs because this way the many chips which can only reliably use 60 of their 64 CUs, due to manufacturing defects, can still be sold off (in a market AMD would otherwise be completely ignoring) and the difference in performance between 60 and 64 is marginal anyway. Vega 56 with undervolting and overclocking is noted for catching up with Vega 64 in gaming benchmarks, for example.

The dodgy CU is simply disabled and firmware is swapped for the consumer oriented alternative (less performance in CAD, more in games and 3D apps) to transform from MI60 to Radeon VII.

Either R9 Fury and Vega 64 were built to be full-spec consumer GPUs, or demand for the business-focused version was low enough that AMD could afford to sell these simultaneously (given supply problems and driver issues with both series, I would expect the latter). Such is not the case for Radeon VII. AMD's more interested in developing the Navi architecture now than creating a brand new maximum spec GPU for gaming, based on Vega.

[Vanqueur]

Either R9 Fury and Vega 64 were built to be full-spec consumer GPUs, or demand for the business-focused version was low enough that AMD could afford to sell these simultaneously (given supply problems and driver issues with both series, I would expect the latter). Such is not the case for Radeon VII. AMD's more interested in developing the Navi architecture now than creating a brand new maximum spec GPU for gaming, based on Vega.

Thanks - Good to know as now need to decide whether to buy the Radeon 7 or wait 6-9 months for NAVI...