Intel Kaby Lake to be Skylake successor next year says report

[Stu C]

My Xeon/X58 combo marches on for another year!

[canopus72]

Intel are delaying Kaby Lake until late 2016 due to poor yields. See RedGamingTech youtube report. I am waiting for Zen. I have seen the block diagram and specs sheet for Zen and it looks amazing. I doubt that intel could offer anything as revolutionary as Zen. Glad to see AMD finally getting their head in the game.

[viper626]

What the hell are intel doing, skylake, kaby lake, ice lake they must have water on the brain, so what do we have here...

socket 1150
socket 1151
socket ice lake

whats next dried up lake...

[shaithis]

Intel are delaying Kaby Lake until late 2016 due to poor yields. See RedGamingTech youtube report. I am waiting for Zen. I have seen the block diagram and specs sheet for Zen and it looks amazing. I doubt that intel could offer anything as revolutionary as Zen. Glad to see AMD finally getting their head in the game.

Tone down your enthusiasm, it won't seem so bad if things don't come out as you think they will

I remember people saying the same when seeing the early Bulldozer information and we all know how well that panned out.....It's easy to be wooed by a diagram but that's a million miles away from the actual silicon performance.....on top of that the biggest difference is a move back to the non-Bulldozer design of one large integer unit per core....a.k.a what Intel have stuck to all along. Not exactly what I would call revolutionary lol.....IMO, that also points to a major failing because it seemed that AMD wanted to push compute (from GPU cores) so that the reliance on the old trusty x87 instruction set was minimised (I have seen 1 game support off-loading to iGPUs; Total War; and that only supports Intel Haswell and above......not AMD.).....that does not appear to have happened in the last 4 years and now AMD appear to be going back to providing decently performing x87 units instead.

The other "biggy": 512 bit AVX...Intel already has units out with.

Another change, SMT.....Intel have been using that since the P4!

So this revolution is what? Copying Intel? Going back to the "old way" of doing things because bulldozer didn't work out?

I am mildly optimistic for Zen and I REALLY want it to succeed but a single block diagram doesn't really say a lot and what it does say can be interpreted a number of ways.

[DeViLzzz]

2017 or later is when I will upgrade so I guess even this Intel Kaby Lake which not much is known about will not be what I am buying.

[DanceswithUnix]

BD has problems, but I think you are looking in the wrong places.

.....on top of that the biggest difference is a move back to the non-Bulldozer design of one large integer unit per core....a.k.a what Intel have stuck to all along.

Apart from Pentium 4 which split things up into that bizarre double clocked integer engine, and an SSE engine that bypassed the cache hierarchy. Oh, and Itanium which was just odd. Oh, and recent Atoms use clusters of two cores which aren't as integrated as BD but you can't get a single core they always come as a pair because that is the most power efficient way of delivering two threads. That's three Intel architectures that are not simply derived from the P3, two of which are still shipping, one of which is quite well regarded and selling well.

I don't understand people's comments about x87 either. Compare FX games performance with Phenom II, and the supposedly lame FX wins every time. In things like video compression that rely heavily on floating instructions it is a huge improvement. In fact when I bought my FX, second pass video compression was the only benchmark where the FX could beat an i7 of the day. Stuff that uses the really old x87 floating point stack operations that crippled x86 performance for decades are now so old that they are going to be doing over 100fps anyway so really only SSE/AVX matters.

Point and laugh at the cache performance by all means, and sigh at the way that FX was left to rot while architectural performance improvements were fed into the FM2 platform that no-one much cares about and will never catch AM3 as long as they stick with 4 threads.

And careful with "Copying Intel" comments. IBM beat Intel to SMT with Power a long time before, and they in turn copied the idea from the big iron Tera CPU. In fact, I think AMD are the only people out there who have never dabbled with SMT. On paper the BD architecture looked like it could have out scaled SMT which has some inherent problems of its own, but they never got that to work so guess they go back to the same boring tech that everyone else uses. Personally, I applaud them for trying at least.

[kalniel]

It took quite a while for FX to realise it's potential though - for a while Phenom II was better for gaming (if I remember rightly). That it did eventually realise it, especially in transcoding, went slightly under the radar either because of earlier results or AMD don't get the pass others do (same thing happened with GCN GPUs - the 7970 is *still* performing really well in games like TW3 - it just took a while for drivers etc. to unlock and the tech stayed relevant for longer than you'd expect/games started using more of it's assets).

But I was certainly among those who called it wrong - I thought high float and multi-threaded int was the way to go. Same thing is happening in phone space - many chips are going high core numbers, while Apple plug away at few core high int perf. and are doing at least as well in the real world.

[DanceswithUnix]

It took quite a while for FX to realise it's potential though - for a while Phenom II was better for gaming (if I remember rightly). That it did eventually realise it, especially in transcoding, went slightly under the radar either because of earlier results or AMD don't get the pass others do (same thing happened with GCN GPUs - the 7970 is *still* performing really well in games like TW3 - it just took a while for drivers etc. to unlock and the tech stayed relevant for longer than you'd expect/games started using more of it's assets).

When BD first came out the Phenom X6 1100T (the expensive, rare one that I never saw in real life) could beat it in some benchmarks. When Microsoft learnt how to schedule Windows for it and AMD fixed the BD bugs to release PileDriver, it seemed pretty obvious to me which direction I should spend my money in. Looking for an X6 on the grounds that "if it is cheap enough, I could still get one of those" the only ones available at all were very low clock and still quite expensive. That sealed it, the benchmarks said that the FX6350 was the right chip for me but bought the FX8350 anyway.

Internet lore still seems to say that FX is a step backwards from the old Phenom "stars" cores, but it is very hard to find benchmarks that back that up. In the benchmarks that I care about (the very Intel friendly World of Warcraft for example) the FX does fine. ISTR there was a few benchmarks where the i7 was already 40% faster, but thankfully not in anything I was interested in. So for me it came down to i7 vs FX, and I went for the cheaper option as the performance was close enough.

But I was certainly among those who called it wrong - I thought high float and multi-threaded int was the way to go. Same thing is happening in phone space - many chips are going high core numbers, while Apple plug away at few core high int perf. and are doing at least as well in the real world.

As I said above, I think FX has been left to rot so it can't really be used as part of that argument. If the caches were fast, if the wider and better decoupled decode of the later parts was implemented with the same 4GHz clock speed, then it might be competitive. We will never know.

As for phones, it seems that it uses less silicon and is cheaper to licence 8x A7 cores than a pair of A57 cores. Apart from possibly better battery life, I don;'t think there is a technical argument for lots of small cores in a phone.

[HassTheMass]

Zen is AMD right?

I really hope that AMD can compete soon. It kinda sucks that Intel basically has a monopoly on the gaming CPU market.