Read more.Wants to stop AMD's next-gen flagship right in its tracks.
Read more.Wants to stop AMD's next-gen flagship right in its tracks.
Both this and the 390x need to get released asap. Can't wait to see how they compare with each other. One of them is going to get bought by me, just need them to be released!
I wonder if this indicates that GM200 yields are better than expected (leaving a bunch of extra chips to sell) or worse than expected (leaving a bunch of chips that don't quite make the grade as Titan Xs).
Doubt the actual release will have much to do with the 390X launch, only the announcements to try and spoil it as much as possible, maybe the final clocks will be set only after getting an idea of the competition, but both teams have done that before.
Really it's just following what they did with keplar - release the upper mid range chip first, then full fat titan, then a gaming version of the titan a bit later.
I think that of larger interest in the 'competition' between Red and Green is that AMD is rumored only to be announcing a product, while Nvidia is rumored to actually be launching a product - both of which are to be the high end of this generation of consumer products. It's never good to be 2nd in a 2 horse race. Actual product always trumps.
What's being ignored altogether (at least here) is that Nvidia is also working on their next gen product, which will also be an HBM product. So AMD has no real advantage in that department.
While neither company has had any real lead for a while (call it a stalemate, to be honest), is this a good way for AMD to do things? If the 390x doesn't blow the socks off of the 980ti, there's going to be some serious repercussions, for AMD, on a lot of levels.
nVidia already released the product, this is a re-release at a different market point, and great though big maxwell is, it's not new.
Quite the opposite - while nVidia are going to release an HBM chip in a generation or so, this is AMD-developed tech which is going in a this generation AMD product. nVidia are going to be at least a generation behind so AMD do have the advantage*.What's being ignored altogether (at least here) is that Nvidia is also working on their next gen product, which will also be an HBM product. So AMD has no real advantage in that department.
What alternative do you suggest?While neither company has had any real lead for a while (call it a stalemate, to be honest), is this a good way for AMD to do things?
Why do you say that? It's still a 28nm product remember, we're not going to get any sock blowing off until we get to smaller nodes, if even then.If the 390x doesn't blow the socks off of the 980ti, there's going to be some serious repercussions, for AMD, on a lot of levels.
*In memory tech at least. I have my suspicions that fuji is some kind of full-fat tonga, so core performance maybe like 2x 285s, plus any mem bandwidth bottleneck removed, maybe rooms for a few more ROPs or something with the spare die area, dunno. I haven't seen an article checking how much mem bandwidth is limiting this gen of chips, though the gains both sides are making with improvements in their compression tech seem to be paying dividends even on 28nm.
Last edited by kalniel; 21-05-2015 at 09:20 AM.
That assumes a product launch where cards are readily available, rather than just enough made to send to reviewers and a handful to disappear of the shelves. If Nvidia keep the chip volume they release low, then it doesn't matter how easy/hard they are to make.
Let's see if hard launch = hard to find cards on either side before we call a winner.
The one bit of Nvidia engineering they seem to struggle with is packaging, as shown in bumpgate. I think I would want to pass on their first generation of HBM. Mind you, it will probably be three years before this tech gets to the price point that interests me anyway.
Kindly point me to where I can put my hands on a 980ti, retail, right now.
I'm pretty sure Samsung and Hynix don't entirely agree on the exclusivity of AMD's development, but that's not really here or there (all 3 companies hold patents on different parts of the architecture, with AMD's part being *almost* exclusive to GPU tech). In addition, I'm pretty sure that most logical people won't honestly lump the 390 in the same family as the rest of the 3xx family. Different tech, different architecture, different everything. I'm not saying this is a bad thing (new tech), and the naming/numbering is all about marketing. But it isn't the same family.Quite the opposite - while nVidia are going to release an HBM chip in a generation or so, this is AMD-developed tech which is going in a this generation AMD product. nVidia are going to be at least a generation behind so AMD do have the advantage*.
As for Nvidia being a generation behind, they've already completed this generation, other than some of the obligatory custom stuff that the oem's will be kicking out for the holidays, and have already announced that HBM will be a huge part of the next generation Pascal cards, and that we should see the entire line in 2016. We shall see.
Don't announce vapor? If you don't even have some sort of mock-up for your current top-of-the-line product to display, then you may as well just be blowing smoke. As I've said before, I want AMD to succeed. REAL competition is good. All AMD has done lately with the 390x is announced a possible ANNOUNCEMENT. Tell me where that makes sense? It's not even a tease, except for the most rabid of fanbois, and even the most rabid of pitbulls will change masters if they get better food on the other side of the fence.What alternative do you suggest?
I say that because, as I pointed out above, the 390x is, technically, a generation apart from the rest of the 3xx family, and the architecture is a completely different design - one built for both speed and efficiency. After all, a major 'selling point' for HBM is that supposedly, DDR5 has reached the end of the road. So yes, considering the likely price point, there should be some doors blown off. Anything less that 10% is going to be a pure disaster, and less than 15% is going to make a LOT of claims very suspect.Why do you say that? It's still a 28nm product remember, we're not going to get any sock blowing off until we get to smaller nodes, if even then.
*In memory tech at least. I have my suspicions that fuji is some kind of full-fat tonga, so core performance maybe like 2x 285s, plus any mem bandwidth bottleneck removed, maybe rooms for a few more ROPs or something with the spare die area, dunno. I haven't seen an article checking how much mem bandwidth is limiting this gen of chips, though the gains both sides are making with improvements in their compression tech seem to be paying dividends even on 28nm.
I've yet to upgrade to a DX11 card, so the next generation isn't even on my radar And I don't seriously stress out over things like bumpgate. It wasn't the first time there was lousy fab, and it won't be the last time - plus it's far enough back in the past as to not even be a bump in the road anymore. My primary (and really only) issue with AMD is their historically, and continually, bad drivers. You and I both have the same issue - AMD drivers and Linux are like oil and water, and they seemingly haven't resolved all of their issues on the Windows front, either. They could make the best hardware in the history of hardware, but if it won't play nice with everything else, it may as well just be a paperweight.
It's the same chip as the Titan X, which has been out a while.
I'm not sure what your point is here. The current AMD line up is a real mix of families.I'm pretty sure Samsung and Hynix don't entirely agree on the exclusivity of AMD's development, but that's not really here or there (all 3 companies hold patents on different parts of the architecture, with AMD's part being *almost* exclusive to GPU tech). In addition, I'm pretty sure that most logical people won't honestly lump the 390 in the same family as the rest of the 3xx family. Different tech, different architecture, different everything. I'm not saying this is a bad thing (new tech), and the naming/numbering is all about marketing. But it isn't the same family.
That's my point - their generation is out, and it's not HBM yet.As for Nvidia being a generation behind, they've already completed this generation, other than some of the obligatory custom stuff that the oem's will be kicking out for the holidays, and have already announced that HBM will be a huge part of the next generation Pascal cards, and that we should see the entire line in 2016. We shall see.
nVidia lead the way on this one - even going so far as to brandish empty card packaging as a real card. No, it's good that they announce future products, or we wouldn't have anything to talk about or look forward too.Don't announce vapor? If you don't even have some sort of mock-up for your current top-of-the-line product to display, then you may as well just be blowing smoke.
I fully agree with your dislike of announcements about announcements - that's more down to places like Hexus for pandering to it though.As I've said before, I want AMD to succeed. REAL competition is good. All AMD has done lately with the 390x is announced a possible ANNOUNCEMENT.
However it's not like nVidia hasn't done exactly the same.
What claims have AMD made about the performance of the 390X so far? I think for you to just pull a 10% = pure disaster claim out there is really a bit silly - there's nothing I've seen from AMD saying that such a thing would be a disaster.I say that because, as I pointed out above, the 390x is, technically, a generation apart from the rest of the 3xx family, and the architecture is a completely different design - one built for both speed and efficiency. After all, a major 'selling point' for HBM is that supposedly, DDR5 has reached the end of the road. So yes, considering the likely price point, there should be some doors blown off. Anything less that 10% is going to be a pure disaster, and less than 15% is going to make a LOT of claims very suspect.
As I mentioned, I don't think the 390X is a completely new architecture design with any particularly new emphasis on speed and efficiency (surely that's what all cards have an emphasis on?) AMD's GCN is mature and a well-proven design, they're unlikely to throw it away.
Not going to bother with all the quotes again.
1) The Titan is a slightly different version of the 980ti, with the same architecture. The 390x is not a slightly different version than the 380, or even 290. You can't really argue that it is - otherwise, it wouldn't be sporting an entirely new memory architecture.
2) My point is, generations aren't a mix of families, and vice versa. You're either a close variant of the same architecture, or you're a different family.
3) It doesn't matter who led the way. 2 wrongs make us look like fools debating on who is worse. You asked for the solution. I provided the ultimate answer. Simply don't announce vapor.
4) A couple of articles here lately have been disturbing, not so much for their content but for their lack of vetting. And #3.
5) HBM is orders of magnitude faster than DDR5, at least as far as bandwidth and actual transfer goes. Has AMD made any claims? Not directly - but they are playing up the advantages (which I would do, too. Why make a new product otherwise?) But the truth is, if it doesn't perform well, it's going to be a PR disaster. Hitching your wagon to a star that gets sucked into a black hole is some serious bad juju. Personally, I hope it does well, as it will benefit not only AMD, but the entire industry, from cpu to gpu to (if I have to) apu...
And I respectfully disagree. If you are completely changing out your memory architecture, you have to make changes to the gpu if only to be able to connect to it. There will also have to be changes made to communicate, handle the bandwidth changes, and a myriad of other details that are way more technical than I can discuss intelligently. This, to me, makes the 390x something that is no longer part of the 3xx family - at least in any normal definition of a tech family.
Well I have had an R9 285 for a few days now, but only used it to play games under Windows for which it has been excellent. I think it came down in the end to "I don't like any of the options", but of the options I didn't like I went for the 285. Have been putting off the Linux reconfig, I guess I should do that over the weekend, so watch this space.
There were rumours of it being a 285. That would make some sense, I googled for a die shot of the 285 and didn't find one so I guess it was never released by AMD. The die size doesn't add up to it's capabilities, it is too big. There is something going on, either it has a lot more shaders to unlock or something else, like a second memory interface lurking unused.
Last edited by DanceswithUnix; 22-05-2015 at 07:35 AM.
The R9 285 has shaders fused off - the M9 295X has the same chip and has more shaders.
DanceswithUnix (22-05-2015)
There are currently 1 users browsing this thread. (0 members and 1 guests)