I think you miss the point. HBM seems to be nothing than a premium product with a relatively short lifespan. Solving nothing but the issues of high end cards. In other words, HBM wont replace GDDR5(X) but rather be another option on the market.
Don't remember when was the last time I agreed with you, but basically this from the first page :thumbsup:You can pretty much say the same about HBM2 over HBM1.
With yields already giving 13Gbps chips I think we will be fine. HBM2 for high end, GDDR5X for performance/mainstream and GDDR5 for low end is how I see it. 4GB low, 8GB middle, 12-16GB high.
However even if GDDR5x is going to replace its predecessor it'll have to compete with the 20nm GDDR5 & as such the prices will have to come down for the former to make it the staple in the 200~300$ price bracket. Will be interesting to see where GDDR3 ends up, since we all know that AIB's still like to churn out the cheapest crap possible, & how the mem controller would have to be tweaked to accommodate GDDR5 & GDDR5x, presumably for the same (cutdown?) chip.the 3 new GPU theory works if all three of them use different kind of memory otherwise it'll likely be just the two
Will be interesting to see where GDDR3 ends up, since we all know that AIB's still like to churn out the cheapest crap possible, & how the mem controller would have to be tweaked to accommodate GDDR5 & GDDR5x, presumable for the same (cutdown?) chip.
My bad DDR3 it is, but even DDR3-DDR4 transition will still take an awful lot of time.GDDR3 has been dead for years now; everything I've seen lately is either DDR3 or GDDR5. (Or HBM, in the case of Fiji.) I would expect DDR4 to be the new standard for low end cards.
At least now we can expect a dGPU to beat most IGP's out there, but as I've outlined the transition to DDR4 will probably take longer than it should, even for the junk category.By low I mean something like a GTX750. Not the junk category below
By low I mean something like a GTX750. Not the junk category below
Is HBM better than GDDR5 in cost? I dont think so and I doubt it ever will be.
You can't be serious. As the production ramps up for any device, costs go down. HBM2 is expensive now as it is new and low production numbers. 2-3 years from now, it will most likely be used on mid range GPUs and AMD's APUs.
dGPU volume is decreasing and HBM got a higher fixed base cost. Its not some kind of law that new technology has to end up cheaper than the old.
Polaris is designed to run both on GDDR5 and HBM.
I've been saying this for some time now. I guess once a belief takes hold of minds, people are reluctant to change their views.Let me quote Raja
http://venturebeat.com/2016/01/15/a...-to-full-graphics-immersion-with-16k-screens/
"Yes. We have two versions of these FinFET GPUs. Both are extremely power efficient. This is Polaris 10 and that’s Polaris 11"
There is no confirmation that its 2 GPUs only. The 2 versions of FINFET GPUs could mean
1. GDDR5 based GPUs manufactured at GF 14LPP.
2. HBM2 based GPUs at TSMC 16FF+.
AMD have experience with manufacturing big GPUs at TSMC. With Fiji they also have experience with HBM and 2.5D stacking at TSMC . So staying with TSMC for the big die HBM2 GPU makes sense. Yields are also better with TSMC.
110 sq mm - 1024 sp, 128 bit GDDR5, 50-60w GPU.
200-220 sq mm - 2048 sp, 256 bit GDDR5, 110w GPU.
300+ sq mm - 4096 sp, 2048 bit HBM2, 190-200w GPU.
If you have a source, say so. If not you need to preface comments like this with, in my opinion, or I believe, or something. You shouldn't state something as fact if you aren't certain it's accurate.
AMD reps have said Polaris is capable of GDDR5 or HBM, there was a video interview. Which they use will depend on the market segment due to "cost issues with HBM".
This basically rules out HBM2 for anything but high-end.
It is very possible that both Polaris 10 and 11 are GDDR5 parts, as low-end and mid-range with next-gen memory compression (Tonga's already got an excellent bandwidth compression, onpar with Maxwell IIRC).
Vega could be the big one from TSMC 16ff with HBM2. This scenario does not contradict anything said publicly by AMD reps. There's "2 14nm FF Polaris, 10 and 11". Note they were very specific about saying the Polaris design was built for 14nm FF.
Polaris 10 & 11= maximize perf/w, small die size, cheap to manufacture, thus sticking with GDDR5. But it is possible Polaris 11 is HBM2, though they are being coy about it.
Vega = maximize perf, big die size, HBM2, expensive, low yields. AMD actually needs something high perf for their Firepros too, it's a market they've been gaining lately (HPC growth despite consumer dGPU loss).
TSMC is not manufacturing any Polaris GPUs. AMD is using Samsung 14FF for both. This has been clearly stated by AMD.
Hmm strange. I was under the impression that Koduri actually said similar thing. You may be right that it is not exactly true...
Oh yeah we already know that!
Vega != Polaris.
A slight update. NVidia and AMD now got GDDR5X samples.
The samples are 1Gbit and 2Gbit. That means 4-8GB for 128bit and 8-16GB for 256bit.
https://benchlife.info/micron-start-sampling-gddr5x-memory-to-amd-and-nvidia-03242016/
16gb with 1-2gbit modules, yeah right. Good luck with that not occupying 1/2 of your PCB.
GDdr5x still has a long road till it ships in densities similar to what we have with gddr5 today (with 8gbit being just available recently).