Read more.Quote:
1GB and 2GB densities available, for partners to make 8GB and 16GB graphics cards.
Printable View
Read more.Quote:
1GB and 2GB densities available, for partners to make 8GB and 16GB graphics cards.
With HBM here and reducing power consumption noticeably, I just cannot get excited about faster GDDR5....
I guess it's more a case of something is better than nothing - nvidia don't have HBM and will have to wait for HBM2.Quote:
Originally Posted by shaithis
But at the moment, the power savings are substantially more useful than the extra bandwidth.....have we seen any benchmarks yet where the bandwidth makes a difference?
Fury X vs 390X perhaps? Guess hard to tease out the effect of more stream processors - AMD will claim they wouldn't add them if they were bandwidth limited probably.Quote:
Originally Posted by shaithis
Maybe 390x vs 290x a better comparison.
Until HBM2 is available, HBM is limited to 4GB due to interposer size limits. Your choices are 4GB of fast RAM, or 8GB+ of a-bit-slower RAM. We've seen that more RAM is a more effective speed boost than faster RAM for CPU tasks, and it's unlikely that will change for current GPU tasks.
With the amount of HBM being limited and looking likely to remain that way for a while, added to that the extra costs involved GDDR5X is a good transition product (IMO).Quote:
Originally Posted by shaithis
I can't see HBM being needed or wanted for anything but higher end products at first.
The HBM is expensive for now am comfortable with GDDR5X for the low end cards, that will be great maybe intel will use it for HD graphics!
Hard to get excited over GDDR5X with HBM2 upcoming. It's like comparing Quantum Dot to OLED: it's not that the former is bad, but rather than the latter is just so much better.
NV didn't need that power drop due to memory, so only AMD is using it. It does nothing for bandwidth currently (maybe next gen cards will have a need). The point here is easier and cheaper to mass produce (see HBM1 causing shortages of AMD cards and they don't sell many with NV at 83% share of discrete, probably more this Q), and we have no idea yet on power usage though one would guess lower than previous versions. This memory has more than enough bandwidth for NV (since they seem to have better compression tech already), and NV already has low watts with their gpus. No point in going HBM/HBM2 (and HBM1 limited to 4GB) and waiting to see if we are in shortages or other issues on HBM2 like HBM1, and also no point in possible delays since NV would be 1st in line for GDDR5x as AMD most likely hasn't got the cash to work on multiple designs at this point (hence, NV being the only one mentioned for cards this summer) and is probably using HBM/HBM2 more and more to get costs down this year. Costs (and issues getting cards out the door) killed them on this crap last year (see quarterly reports).Quote:
Originally Posted by shaithis
HBM took a whole new process, while this is just a small swap out from old stuff. That's a much safer play in bad economic times. HBM lost AMD market share and higher margin sales (among other issues). The quarterly reports and market share data show this unfortunately for AMD. The sad part is, HBM/HBM2 won't get much cheaper for them if NV puts off using it. That might be one reason NV is going GDDR5x, as it hurts AMD margins if they're alone with HBM/HBM2 for the most part. Smart business. If watts are lower for the new stuff, it almost makes HBM/HMB2 a moot issue for quite some time to come.
http://hexus.net/tech/news/graphics/87533-slides-touting-gddr5x-memory-advantages-leak-online/
The slide shows lower power. Probably won't be HBM levels, but it only has to get close to help NV who already wins on the gpu side (part of the reason AMD went with HBM to get watts down on their top end stuff). With old stuff topping out at 8GB/s, you could see running stuff slower (1.5ghz) and getting 12GB/s netting more than enough for next gen NV stuff most likely while keeping power lower. NV cards show you get far more OCing the gpu than the memory by far. Surely adding 50% more bandwidth is pretty good already when highly oced cards don't get much from memory on NV's current top end. I'll take the lower price thanks, and better quarterly report from NV due to choosing that route. AMD got screwed by HBM1 and I expect the same from HBM2, by either cost or shortages or both.
Nvidia have already confirmed Pascal (their next GPU) is going to using HBM2, although to what extent is yet to be seen, it maybe only the reserve of the top end cards.
Nvidia only had an advantage in terms of power draw this time around because they moved parts of the GPU into the drivers, something that looks to be changing with the next round of GPUs.
HBM2 fixes the bandwidth limitations because (iirc) it increases the 1024 bits memory bus to a 4096 bits memory bus.
Money isn't an issues for AMD as Micron will sell GDDR5x to anyone willing to buy it, it's Micron who developed it not (afaik) Nvidia or AMD.
AMD have said Polaris is scheduled for release somewhere around July to early September.
Maybe i missed some corrections as i stopped reading after the first paragraph, :)
TBH,I don't really care what type of RAM they use,outside of technical discussions,as long as price/performance for the final cards is decent! :)