Now despite probably not happening, and probably not being useful at all.
What would happen? If AMD releases it and is set up for the potential far far future of 128-bit processors also acquiring that patent as well.
I'd say it gives them about 5 minutes of marketing material.
I don't even think we're near the limit of 64 bit right?
It would let them make a motherboard like a mile square and fill it up with RAM though right? Or would it need to be 10/100/1000 miles square to hit the limit? I need to go look up some math
Rather than ranting about how the back to back launch of their last two generation of GPUs fills me with 'confidence' and pointing out how they have failed to meet delivery and performance targets again and again, I would like to point out they have just restructured the GFX department, such that it can almost function as an independent company....
This does not fill me with confidence for their sucess, apart from their custom emended solutions, it's their only profitable department, and it's lost half its marketshare, probably in large, due to the funds being syphoned off to the other departments, causing a shortage in RnD resources....
This could mean they are preparing for a takeover (where someone doesn't want the GFX department), too sell it off for money to fund their CPU department, or because they are going to go bankrupt.... And they want to be able to sell everything off so they can return something to their shareholders.... Not good...
Does it do anything? Its faster And what does that do? It means it goes faster, its a higher number, that means it's better, see we are cutting edge.... But the benchmarks show that all that extra very expensive bandwidth makes no difference. and now your card is very expensive and still worse than the competition.... HBM is the future, AMD the way it's meant to be played...
They can do that, but replace HBM with 128bit.... And then claim it's a premium product.... I can see it happening. TBH I don't see how they could screw it up, I mean Intel is dedicating what half their die to the dGPU and people buying high-end CPUs don't see any gain from that, AMD doesn't need to do that, they can also build larger dies and accept lower margins, if that means they can deliver a product which performs and is priced fairly close to Intels then it's worth it as Intel has very high margins.
They can do a combination of this, remember that they are better at making GPUs than intel, so they can get away with less space there, and use that on their CPU segments which are not as competitive. If they can't get up there with intel then I think they are going to have to look at being aquired... Go begging for someone to buy them... And then watch a lengthy battle with Intel for the X86 license.
Fury should be way faster than it is, they just have unoptimized drivers it seems, and it is a first gen kinda thing. I believe it outperforms the 980ti if there's no AA at play.
It's also AMD's fastest card atm. and the Fury X*2 might be something half decent if they get the pricing right.
I was thinking that their GCN architecture has trouble scaling... If that is the case it's a serious issue, they will need to fix that for the next generation... Unfortunately Street, the Fury X2 is going to be bad...
With CFIRE its going to be about 1.95x FuryX performance, with lower OC potential, but its going to be more like 2.5x the price, you are going to pay a 'premium' for it... Because AMD/NVIDIA do that now with dual GPU cards, and they do that with 'performance kings'....
Meanwhile you are getting like 60-80% of the performance of the OCd 980ti, that equates to a 17-56% performance boost over the 980ti, when scaling is good, if that... You would be better off getting two fury non-Xs (for whatever that costs, I know it's much less), or more sensibly two, 980Tis, which have 6GB Vram each. And banking that extra cash, or putting it towards a better monitor or audio experience or something.
Maybe it's why nvidia hasn't come out with a dual card for this generation, AMD is going to end up with 2 + Fury X*2 the 690 was nvidia's last dual GPU card right? and I guess there's the titan Z that I don't think anyone bought
Yeah that's probably why. nVidia doesn't put much RnD into it, most likely. But if you look at it another way...
AMD made a dual card last generation because its very best single GPU was a whole tiers worth of performance behind nVidia's best. So by making one card with 2 GPUs they could compete at nVidia's flagship level. 1 flagship vs 1 flagship. I think AMD needed to put a lot more RnD into crossfire because they would fall behind on the flagship front. They loved mentioning that they have the fastest card out, but we knew it was 2 GPU's.
AMD agreed that the agreement is null if they are aquired, probably to prevent a hostile takeover, I imagine the courts will have to overturn it, or there will be government intervention, because nobody wants a complete monopoly.. And yeah I heard that about Xfire scaling too, but only recently, no idea where it is coming from, but with performance like that its hard to justify.
Used to think people were idiots buying the 980tis and whatnot, so expensive. Until I saw Jayz benchmark that I posted, and the two 970s which are very nice and 1/2 price, barely hold their own against the single 980Tis, when OCd, which are a single card with no SLI issues, added latency, etc..etc.. I honestly didn't think such a big die would go so high, as normally the larger the die, the harder it is to push high frequencies. TBH Nvidia did a good job with the maxwell generation.
Is it worth paying 100% of the competitors price tag to get 60-80% of the performance, with less Vram, or 200-250% of the price, for 117-156% performance, no.... I mean really why bother... Yes it's technically going to be 'faster', does that mean you should buy it, no... Buy two 980tis for less and get more performance... So it's sad to see them release such products, that they are marketing in a similar way to the ZEN processors...
@Streetguru Don't think there is an actual rumor that its going to be 128 bit. But, I'm pretty sure you were talking strictly in hypothetical anyway. Just posing a 'what if' statement.
Looked a bit into that 128 bit "rumor" (lets imagine). That has got to do with something in the processor, not the processor itself. That's talking about FMACs. They are rumored to be 256 bit, or something. Thought I'd make it clear.
@Jacobite Yeah, what you say about the agreement is true. But there is actually nothing standing in their way if they would like to make a new agreement. They just have to go through the motions again. Stuff you said about the graphic cards is pretty interesting, too.
I think this is how it is going to go down. The mindset of these manufacturers is usually that engineers go develop, build, optimize, etc. the best thing they can, then pass on key points, stats and cool sounding numbers to the marketing team to sell the stuff.
That's what they were saying when nVidia was going through that GRAMgate crap. The engineers just didn't make it clear, or the marketing team misinterpreted the info.
I would think that very rarely will the engineers just make something into a bigger number to make it sound cooler. It would hinder performance / increase costs if it doesn't help. Unless the "improvement" is simple and tangible, like putting 8gb on a 390x / 390, instead of 4gb.
If its complex architecture stuff, like the processor being in 64bit vs 128bit, they would probably leave it to the engineers. If they think it improves performance, it probably would be put in.
I would say that's part of it, yes. You have to assign a memory address to each byte in memory. If its 32bit, then the biggest number you can imagine is 2^32 ( or 4,294,967,296 ). Therefore you cannot possibly perceive the 4,294,967,297 memory address's! Crazy talk! Increase that to 64 bit, you now can imagine 1.84×10¹⁹, and therefore more memory locations you can access. That's how it goes, if I'm not mistaken.
so then we just need like a really big processor, a really big motherboard, and then all the RAM ever produced by mankind running at 1mhz to fill up the 128-bit limit
Have we even made enough RAM to do that? because it's gotta be a pretty big number.
It would take a monumental effort and lots of money to fill up a 64bit processor's worth (256TB). But in the future you could probably buy 100 TB DDR-1000 dimms, so 3 would be sufficient. (Though I heard they are dropping the DDR standard in the future, so might be called something else)
Well, they can raise this limit to 16 Exbibytes in newer implementations of x86-64. So newer implementations of a 64 bit processor can adress 16 x 1,000,000,000 GB, or 2^64 bytes. ATM vanilla x86-64 can only address 2^48, or 256TB. So 64 bit processors have a lot they can do yet.