I LOL'ed, thanks.
AMD Vega 10, Vega 11, Vega 12 and Vega 20 Rumors and Discussion
I LOL'ed, thanks.
D
Deleted member 13524
Guest
To be honest, I don't think anything at all is changing with AMD's roadmap.
The only thing that happened this week was a PR move that gave an official name to what appeared in AMD's roadmaps as "14nm+". It didn't exist so far, but now we know they're calling it 12LP.
I don't think they'll "transition" Vega 10 or Vega 11 to 12LP either. If they do, it'll come under a rebrand like "RTX Vega 64" or something, but they won't simply transition the GPUs to a new process without saying anything about it.
They might use it for Ryzen refreshes (Zen+, not Zen 2), and consequently they could be making Zen+ APUs with that process (if they're not doing so already), hence claiming there will be "Vega GPUs" with the new process.
Or maybe 7nm is too late for Vega 20 so they're pushing it forward using 12LP. Truth be told, AMD's only offering for FP64 compute at the moment is the 4-year-old Hawaii.
I don't see it that way. Vega is a different name than Navi and that alone should trigger investigation. Pricing between Vega 64 and Navi 64 will be different.
Consumers might pass on bargains like getting a Vega 64 that performs better than e.g. a Navi 48 that happen to be both priced at $380, but I don't see how someone will buy a $380 Vega 64 thinking it performs similarly to a $500 Navi 64.
The best part about including the architecture codename in the consumer card name is that people won't be tricked into buying an old-generation card because it has a new-generation name. Like e.g. people who bought a Pitcairn R7 370X in early 2016 thinking it would be a good match for their new FreeSync monitor.
Future features should just say e.g. "compatible only with Navi GPUs or newer".
Oh boy was that cringeworthy...I give you the HD 6850 and HD 6870 as an upgrade to the HD 5850 and HD 5870
DrYesterday
Newcomer
Vega 11 - 28 & 32. These look just like half of Vega 10.
Is it possible AMD is taking Vega dies that have manufacturing faults and using the good half of the die, with one stack of HBM2? Or has AMD designed a new die?
Is it possible AMD is taking Vega dies that have manufacturing faults and using the good half of the die, with one stack of HBM2? Or has AMD designed a new die?
Whole lot of rosey...
Bondrewd
Veteran
Vega11 means new die.
digitalwanderer
Legend
Ooops, sorry. Had no clue. Never mined and I don't really run compute applications so I learned something new. Thanks!
Bondrewd
Veteran
Which honestly makes no sense.
It seemed like AMD was making an earnest effort to reset/streamline their naming around the time of Polaris release, which was a good time to do so in light of "Fury" muddling the waters and rebranding controversy. Of course, the whole scheme went right into the garbage bin with advent of 580 cards and now "Vega" branding.
BTW, I have to agree a comment above about generation-specific branding potentially undermining the Radeon brand which is very well established via copeous marketing resources poured in.
digitalwanderer
Legend
Mebbe they should bust themselves up in to two divisions. One for CPUs and one for the GPU stuff. They could keep AMD for the CPU stuff since it's so well known and liked right now with Zen, but what could they possibly call their GPU unit that would bring back a certain nostalgia/hope/enthusiasm for the brand? Some name that everyone wouldn't be hATing on....
Bondrewd
Veteran
It's still going to be Radeon.
Heck it's three times more Radeon now, being RX, Pro and Instinct.
you have to explicitly disable Crossfire/SLI, as for many/most
All compute situations / applications .... CFX /SLI are only something relative of gaming ( as it is related to frame / frame calculation ), for all other situations.. think to multiple CPUs ..( most interessant is to know how memory pool can be shared in computing today between GPUs ... as today ( as with Raytracing ), the memory is not a big pool.. ( if i have 2x 8GB gpus, i dont have 16GB available, but in reality only 8GB as each gpu access their own memory pools .... and with today CG, memory pool size start to be an endemic problem .. )
Last edited:
License Qualcomm Adreno tech from all their old gang that went over there. It just hasn't been the same without sireric. Except from what I understand their drivers suck too.
Frenetic Pony
Veteran
Well, it'll support double rate fp16 and raster order views, which is nice. But current HBM 2 costs do make it seem questionable.
Why do people presume Vega 11 has to use HBM? AMD stated before that they have capability to support either HBM or (G)DDR depending on which specific product needs and after Vega-launch that Infinity Fabric makes connecting different IP-blocks easier. Vega 11 could just as well be with GDDR-memories.
Not connecting 'Radeon' to 'Vega' is probably a good thing for the Radeon brand...
In my opinion, people won't like products with 4GB of memory, even though with HBM2 and HBCC it is more than enough and won't be worse than 8 GB GDDR (it's probably even better). So if they really release such product, there should be large text on the box saying:
4 GB HBM2, up to 24 GB of memory
or whatever they decide to be upper limit for these cards
4 GB HBM2, up to 24 GB of memory
or whatever they decide to be upper limit for these cards
Not really. VRAM is VRAM, try using HBCC on a system with 4GB or 8GB System RAM and see how glorious this is compared to a 8GB VRAM card.