Vega 8-GPU for laptops uses DDR4, not HBM2
Click here to post a comment for Vega 8-GPU for laptops uses DDR4, not HBM2 on our message forum
Ricepudding
Bit of a shame it doesn't use HBM2. although DDR4 is fast (least it can be) i'm sure most vendors would use slow 2133 or 2400mhz meaning they would bottle neck in games/programs. Would be interesting if they could make APU's like they do in the consoles for laptops for people who mostly just want to light game on the go.
sverek
Could CPU even support HBM2.
kruno
Yes,i don't see why not, only reason to not use it for CPU is relatively bad latency of HBM .
Herem
Not sure why all these stories are popping up saying this is a surprise. None of the AMD APUs have ever used graphics memory before.
These are cheap parts for low spec machines, using expensive memory doesn't make any sense at all. If HBM2 becomes main stream with price parity then it may happen, but until then there's no chance of seeing it at the low end.
rl66
yes the graphic part can use HBM2 and the CPU regular mem... like in some high end SOC.
I guess it is more a cost choise... HBM2 is still pretty expensive, look at the price of the desktop vega over NVidia.
heffeque
I'm wondering who expected a 15W mid-range APU to have HBM.
Texter
The IGP's running out of RAM anyway regardless of the type. nVidia's MX150 has 2GB GDDR5 so it can easily double performance over this.
Silva
If Vega can run on DDR4, what the fuck is stopping AMD from using GDDR5 on mainstream Vega? After the mining craze, we need a new sub 200 USD GPU because RX570 prices are absurd right now...
Evildead666
i would suspect the 256Mb bit is just what the integrated GPu is consuming at that point in time.
It should be able to dynamically allocate much more than that to the Vega-8 iGPU.
sammarbella
Good question for Raja...
...Upss he's working at Intel now!
Loophole35
That is a good question isn't it.
Texter
Yeah that's what I thought, got distracted and a couple of minutes later posted something stupid LOL.
Evildead666
This was specifically designed for an APU, which is using DDR4.
they could fit a GDDR5 controller onto vega for the lower end boards, if its economically worth it.
if they have still got to move a lot of RX5xx's, then they won't.
then again, if the performance difference is slight between the vega+GDDR and Polaris+GDDR, then it is not worth it this Gen either...
kruno
Power consumption, GDDR5 requires more power (significantly ) then HBM2,and considering that Vega is already power hungry using GDDR5 memory would make things even more worse then it is.And second thing is as seen in different reviews Vega is memory constrained,so you would need to put boatload of GDDR5 chips to get HBM2 throughput.
rl66
i think that some of you make a mistake with the GPU chip and the GPU architecture...
AMD can do a GPU based on Vega with whatever they want if mem controler is compatible...
The problem with actual Vega line is that it's with HBM2 and so the mem controler is not compatible with other mem than HBM2 and on top of that this mem is onboard the chip
So if it happen the new "Vega" would be a radicaly different GPU... but it is possible (NVidia have done it with Pascal but on the other direction for computing)
Kaotik
No, it's not a good question.
It's quite simple really, AMDs GPU architectures are more or less modular these days, you can mix'n'match different IP blocks -together (with some possible limitations, but now that they have Infinity Fabric it's more simple than before, examples of different IP blocks include the "GPU-block", VCE, UVD, memory controller etc).
Vega 10 has HBCC, which is HBM2-memory controller. It can't use GDDR or DDR-memory. That, however, doesn't mean that every Vega has to have HBCC, like in case of Raven Ridge, they're sharing the same DDR4-memory controller with the CPU. They could make a Vega with GDDR-memory controller too, but it would be completely separate project and chip.
This is also why we're currently not sure whether the dGPU on that AMD+Intel-combochip is actually Vega or Polaris. HBM and timing hints Vega, but old leaks suggest Polaris, and nothing would prevent AMD from making a Polaris with HBM-memory controller (whether it's same as in Fiji or same as in Vega is another thing).
Oh, and AMD said somewhere around Polaris launch outright that they have the capability to use which ever memory technology makes most sense for each chip, referring to the modularity of the architecture. This of course has to be decided before you start laying out your chip, but it can be done.
sammarbella
From there your reply was not so easy to understand. 🙂
Kaotik
Sorry. Let's see if this makes it easier:
AMD designs their GPUs from Legos.
They have Polaris-Lego, Vega-Lego, UVD1-Lego, UVD2-Lego, VCE1-Lego, VCE2-Lego, GDDR-Lego, HBM-Lego and so on.
You can build the full GPU using any of these Legos.
We have currently for example Polaris-Lego + UVD1-Lego + VCE1-Lego + GDDR-Lego, let's call that Polaris 10
Then we have Vega-Lego + UVD2-Lego + VCE2-Lego + HBM-Lego, let's call that Vega 10
We could also build Polaris-Lego + UVD2-Lego + VCE1-Lego + HBM-Lego for example, or Vega-Lego + UVD1-Lego + VCE2-Lego + GDDR-Lego and so on. Mix and match.
From real life easiest examples can be found from GCN3 (also known as GCN 1.2 and 3rd generation GCN and so on).
Fiji (Fury-series), Tonga (R9 285, R9 380-series) and the integrated GPU in Kaveri APU are all based around GCN3 and similar other elements, but each has different memory controller: Fiji has HBM, Tonga has GDDR and Kaveri has shared DDR memory controller with the CPU
sammarbella
Thanks for the teaching effort, i enjoyed LEGO games when i was a kid.
It seems AMD has the same problem i used to have: that LEGO piece i needed to complete my perfect starship was always lost.
AMD lost his LEGO performance piece years ago. 😀
kruno
Well AMD's current problems with graphic designs are more of a manufacturing design problems.GloFo 14nm FinFet node is more designed around low power low clock designs then around high clocks.Amd tried to offset low clocks by basically overclocking and overvolting chips and that didn't turn around so great. Vega was planned for last year but GloFo didn't have process ready,and the reason why amd didn't go to TSMC is long term contracts they have with GloFo which basically penalize them if they don't meet agreed volume.And about second thing memory controller is independent unit, instead of controller that support HBM2 they could put for instance DDR2 or DDR3 controller it doesn't matter which one it is buuuut you lose bandwidth if you go with some other memory configuration and that affects performance.
