IDK man, what your saying makes an awful lot of sense, I would like to agree, but I just don't see where the benefits from HBM are. I mean perhaps next gen, at 14nm, if they build large dies, with many, many more transistors, then we might see gains, maybe... Pretty sure there were some benchmarks where people downclocked the HBM and we didn't see much of a performance hit.
Before it came out, I thought it would really pull ahead with the HBM at higher resoloutions, turns out it doesn't work like that. It is a shame, I am sure there are some advantages somewhere to it, I mean it was a long time in development, not some spontaneous decision. I just don't see it though...
@streetguru I think that the future server chipsets are supposed to support some stupid amount of ram, not that much but still. On another note, there were some companies looking at doing funky stuff with the ram, like sticking NAND flash on there. If you did that, you could potentially see them pushing against that limit... Eventually...
I am waiting for higher capacity DDR4 though, I mean for god sakes, same capacities as DDR3, and with much higher latency, in some cases not much faster.... Not really an improvement, just a scheme to sell more ram.... Saying that I also want more cores with higher IPC..... Hopefully zen won't flop, and won't be intel expensive. I really need a good 6 core with VTd to run the satan OS (win10, with customer experience feedback program)... IN VM, OFC, Because that stuff is not secure or safe...
Big MAYBE! Zen with its high corecount needs more "lanes" to make them work together. There is also a chance of HBM being combined with Zen, maybe thats a reason.
Yeah, I get what your saying, and I totally see you point of view. We were all hoping for it to be more of a success than it was. However, lemmy put another spin on it...
Fiji is heavily based on the r9 285 (Tonga), mainly for its GCN 1.2. They would have had to scale up that GPU to get it to flagship level, and simply scaling up doesn't improve performance in a 1:1 ratio. A 285 is definitely not flagship quality.
Developing Tonga architecture to take advantage of HBM may have made a huge difference in performance. With GDDR5 instead, Fiji could have just been a fast Tonga, and therefore would have not even been able to compete with the 980Ti.
Hey, it could have been just an overclocked Tonga, if it didn't have HBM. We would've been really disappointed. nVidia seems to be going in the direction of HBM too. There are other benefits other than bandwidth, like power efficiency, and scalability for the future. Really cant see it being a gimmick.
My main point is that engineers rarely put in a gimmick that could possibly hinder a design, or require massive amounts of RnD just to say "its new", but in reality not offer anything.
Like you were saying, I also think that they will perfect it in a few years, and it has the potential to gain back some market share. They have the advantage with HBM, lets see how nVidia fairs.
There's no need to go 128bit, the only reason we did go to 64bit was to accommodate hardware bit space requirements for memory mapping and other I/O's. Its far better to go with smaller packets for cpu, as they will be processed faster than long one (networking 101). Thus it would actually make the cpu slower than most, and they would need to develop something like x86_64 ~ but this time x64_128 ~ again pointless.
I think someone already patented 128bit cpu... i think ibm.
I hope AMD keeps being independent company, once they cease to do so they will fail. Only companies that could afford to buy AMD is Samsung & Microsoft. Valve is too poor to buy them up ~ sorry folk but fck valve and gabe in his spyglass.
a big thing for AMD -> APU's is going to be HSA, lets say MS will add 100% support for HSA in windows ~ (or even cpu's being able to use AMD gpu's) their gain from this would be so big, intel would have to pack their sh*t up and gib something new good and op...
by worth of total equity it may look like they can but... its not that simple. If you wanted to buy AMD you'd need about $7-10billion in cash. Total worth equity of Valve Corp is about $2billions. (so if valve sold itself it might get somewhere... near.)
mhm yeah, but why would AMD want that? There would be nothing to gain from such a move.
and lets say if - you would see plenty of proprietary shit always watching drm in drivers and what you can imagine... valve like their porn ~ especially if its yours.
just about any modern x86 cpu has the PAE instruction set which allows using 64gb ram in a 32bit linux or mac. windows doesn't use PAE due to driver issues. that said some companies like asrock have made a work around that allows you to use extra ram as a ram disk. but really the biggest thing zen needs is single thread performance. it's really sad that the AMD A10-Series A10-7870K is weaker then Intel Core i5-2500K on the cpu side. at the very least they should be matching ivy bridge right now.
Noo... There market cap is much lower than that... But I do agree valve would not buy AMD as they are in completely different industries, it would make some sense for perhaps Samsung to buy and try and revive AMD or enter into the X86 market under their own brand name.
Samsung has been developing their own custom processors for their smart devices, so they could very well be interested in combining the expertise from AMD, not to mention if they can push AMD back up again, then they can funnel all of AMDs Fab needs for processors and their vram into Samsung fabs. That could be worth it, depending on what sort of sales volumes AMD could achieve under their umbrella.
Keep in mind that would be lost sales for their competitors fabs. And the AMD experience might give them an edge in their smart device markets, and the server world.
Their market cap reflects that, it's pretty amazing how cheap it is. TBH samsung would probably do quiet well buying it out, as whatever new management they bring in is by default going to do better than what they have now. So they can increase value quiet a lot.