Howdy, Stranger!
It looks like you're new here. If you want to get involved, click one of these buttons!
Quick Links
Intel Haswell to arrive in the first quarter of this year
http://news.cnet.com/8301-1001_3-57564603-92/intel-ceo-the-pc-is-shape-shifting-into-a-tablet/
"In the first quarter we launch Haswell."
That, I think, is the big news item. Rumors a while ago had put Haswell launch around March or April, then later rumors put it closer to the middle of 2013. If Haswell is going to launch in the first qarter of 2013, Intel would have to already have a ton of chips made and know that the yields are acceptable for launch.
"The single largest generation-to-generation battery life improvement in Intel history."
That's a reference to Haswell's efforts at bringing idle power consumption way down. x86 processors have traditionally used a fair bit of power at idle, with most recent chips in the ballpark of a few watts. Atom is the exception, but its performance is dismal. Intel Haswell and AMD Jaguar cores should both bring idle power consumption way down.
"We are in the midst of a radical transformation of the computing experience with the blurring of form factors and adoption of new user interfaces. It's no longer necessary to choose between a PC and a tablet. Convertibles and detachables combined with Windows 8 and touch provide a 2-for-1, no-compromise computing experience."
He tries to sound excited about a future move from laptops to tablets and tablet/laptop hybrid devices. (It's not a move from desktops, as nearly anyone who was going to move from desktops to tablets has already moved to a laptop.) But it's a future that looks to be largely without Intel. Clover Trail Atom is an unmitigated disaster, and it's unlikely that a Haswell-based tablet will make more sense than an AMD Temash-based tablet for more than a tiny fraction of people looking to buy a Windows 8 tablet. And that's assuming that people even want an x86 tablet in the first place, as opposed to ARM.
Intel's first real shot at having a nice tablet chip comes with Bay Trail, which will have Silvermont Atom cores around the end of the year. That will be a die shrink to 22 nm, giving Intel a temporary advantage over the rest of the industry at 28 nm, but if Intel still wants to put Atom into cell phones, they might not be able to scale performance up far enough to compete with AMD Jaguar cores on the CPU side. And in order for Intel to compete with AMD in tablets, they don't just need to be competitive on the CPU side. They need to absolute crush AMD on the CPU side to try to compensate for how they're going to get slaughtered on the GPU side.
Comments
Interesting that his new buzzwords were "Convertibles" and "Detachables". I guess we can look forward to a lot of this later this year.
New motherboards, for one thing. 1155 was nice that it actually lasted through 2 generations. Haswell and Broadwell are both rumored to use Socket 1150 (or H3) for their desktop socketed versions. Heatsinks will require new brackets, as it will be differently sized from 1155/1156.
New motherboard also means new chipset: Lynx Point. Not really anything exciting here, 14 USB 2.0 and 6 3.0 (with some rumored ability to designate which physical ports are which), all SATA3 (no more split between SATA2 and 3), 8 PCI2.0 lanes (PCI3.0 is baked into the CPU, not the PCH). The only huge change will be the FDI and DMI will be merged into a new interconnection: the OPI (on-package interface) - not really that significant really, since the FDI only deals with integrated graphics. Lynx Point may also have some additional SSD optimizations, which haven't really been spelled out in great detail (in addition to the SSD caching ability).
The problem with a convertible is that it's not as good of a laptop as a real laptop, and it's not as good of a tablet as a real tablet. In tablet form, it means you're stuck with the full weight and thickness of a clamshell laptop.
In laptop form, it could be better, but I'd worry about something breaking if you're shifting it between tablet and laptop form a lot. You need it to firmly stay in place in whichever form you have it, but be easy to shift from one to the other when you want to, and without messing up the internal electronics (power and signal cables for the monitor, basically) that pass through the connectors as you shift back and forth hundreds of times. You can't disconnect the pieces entirely or the machine won't work, so there will have to be a lot of twisting back and forth. It might be doable, but I'd worry that it would be fragile and/or expensive.
Haswell could be a nifty chip for convertibles, if there is much of a market for them. Which there might not be.
I see more of a future for detachables. That's basically a tablet with an optional keyboard dock, akin to Microsoft Surface or the Asus Transformer keyboard dock. There, you just need to pass through USB signals, as the guts of the system are in the tablet portion.
The only real disadvantage to a detachable as compared to a pure tablet is that you have to have the spots set up for the keyboard to dock, and to be able to take USB input signals through it. That's not that much of a problem.
Detachables do have a big performance disadvantage as compared to normal laptops, though. Your processor and graphics are almost right up against the monitor screen in a detachable, just like a tablet, and monitor screens don't like to get terribly hot. Intel is promising a 10 W SoC bin of Haswell for tablets, which is a lot of power for a tablet, though kind of viable.
After seeing the Ivy Bridge "7 W" Sometimes Design Power fiasco, I have to wonder what the "10 W" for Haswell actually means. A month ago, I'd have assumed that if Intel says they could make a 10 W Haswell chip, then they can. But now I'm more skeptical. They're going to take Ivy Bridge cores, increase performance, add more stuff on die to make it an SoC, add much more powerful graphics, and bring the TDP down to 10 W on exactly the same process node as Ivy Bridge? Maybe if they put a heavy emphasis on bringing down the minimum clock speeds that Haswell needs to run. Or maybe if they make it a single CPU core and really cut down graphics. If it's a 17 W chip with a Sometimes Design Power of 10 W, that's a no-go on tablets.
Meanwhile, Intel is going to have to compete with AMD Temash in tablets, which is actually designed for low power. The top Temash chip will have four Jaguar cores and two GCN CUs, all clocked really low, and be an SoC in a TDP of 8 W. The lower version will probably be approximately half of that on all counts, and 4 W has no problem going into a tablet. I'd bet on the real-world power difference at load between an "8 W" Temash chip and a "10 W" Haswell chip being a lot more than 2 W, too.
Jaguar cores aren't the solution to everything, of course. Put them in a desktop and you'd have a nettop. But they are the solution if you want an x86 chip around 5 or 10 W.
Haswell cores will have much higher single threaded performance than Jaguar cores, of course. But if you were really worried about single-threaded CPU performance in a tablet, there's nothing stopping some OEM from putting a 9 W Kabini chip in a tablet. That's basically the 4 W Temash chip, except clocked and volted higher. Will a Haswell core at whatever clocks Intel needs to hit 10 W really be so much faster than a Jaguar core at 1.7 GHz or so? Maybe, but it's not obvious. I don't think you'd want to put Kabini in a tablet, but you could.
http://www.tomshardware.com/news/Intel-Haswell-Ivy-Bridge-E-CPU,20590.html
Or maybe not. That says that desktop Haswell isn't coming until June, and even then, it will only be the high end at first. Maybe Intel wasn't talking about the "first quarter" of 2013? Or maybe laptops will come out long before desktops, as AMD did with Trinity? Or maybe the rumor is just wrong.
Microsoft / Nokia missed the gravy train with smart phones. Why buy a win phone when you can buy an Android one for the same price and have a much healthier ecosystem of sold for it? Why buy a win phone for the showing off factor when Apple has that market?
Same with tablets, too late to the table. People have played with ios and Android tablets, they realised a) they don't need a laptop to just browse the web, play little games, play dvd or listen to music. And b) they certainly don't need a win tablet as a replacement with all the bloatware and overhead, both in hardware inefficiency and user setting up time that comes with Windows.
Haswell desktop chips are coming on week 22-23, that's May/June.
Who cares about the mobile chips? They are not the bottleneck in laptops anymore.
Windows Phone has its own little ecosystem that doesn't have much that can run on it. Windows RT has the same problem.
But tablets running Temash, Haswell, or even Atom can run the full Windows 8, so you have access to everything that runs on Windows 8--such as most of the games on this site. A large fraction of the games on this site won't be playable on an Atom-based tablet, but most will on Temash or likely Haswell.
Does that mean there will be a huge demand for Windows 8 tablets? Not necessarily. The hardware to do it properly isn't available yet, so we don't really have a good gauge on it. But it's not going to be saddled with the same "hardly anything will run on it" drawbacks as Windows Phone or Windows RT.
Actually, Haswell is going to be an excellent chip for laptops. For the first time, we'll have a high performance CPU architecture that uses virtually no power at idle. Just swapping out Ivy Bridge for Haswell while changing as few other things as possible could reduce idle power consumption by a couple of watts, and potentially extend battery life by an hour or more--and while actually increasing performance a bit.
I don't see this laptop market. People probably still use laptops for business but for home use tablets are massively out selling laptops. It's one of the reasons win 8 Is flopping even worse than Vista.
Low idle power consumption only helps when the machine is idle. Desktops and laptops are essentially idle most of the time. Servers aren't.
Servers are actually in the middle of a transition to microservers. More cores clocked slower beat fewer cores clocked faster if your workload scales well to more cores. So the idea is to get the weakest cores that are powerful enough for what you need, and then get as many of them as you need. That means that there is space for a lot of different cores to each be the best for some server workloads.
That's why, in addition to Piledriver cores, AMD is going to make Jaguar-based Opteron processors. In addition to Sandy Bridge and Ivy Bridge cores, Intel is going to make Atom-based server processors (though I'm not sure if they'll get Xeon branding). There are even ARM-based servers from Calxeda and Marvell, though current ARM CPUs are only 32-bit, which is a problem for a lot of server workloads. ARM v8 will fix that sometime around 2014, and I'd expect to see ARM in a lot more servers then.
Some servers do still need high single-threaded performance, and they'll continue to use Intel's high performance architectures such as Haswell.
There are also concerns about how much the cores need to communicate with each other, and how much memory they need to be able to address. A 32-bit CPU caps you at 4 GB, and Avoton Atom will only get 8 GB, and either of those are perfectly fine for some workloads and a complete non-starter for others. But anyway, servers are complicated.
I haven't been following computer hardware well. Do we have any info on the TDP at max load of Haswell processors?
My brother wants to buy a Windows tablet, and I told him to wait until they put something in them that uses less power than an i5. Is there anything on the 6-month horizon which will be significantly better at power consumption for tablets than the i5 while still doing as good of a job as the AMD A-core APUs? Those are the best mobile x86 processors I know of right now. I'm eager to see Intel make something better in all aspects.
If you want a Windows tablet sometime this year, then you want AMD Temash. It really is that simple. There will be two Temash chips. One is a quad core processor with Jaguar cores at 1 GHz with a GPU with two GCN CUs at an unknown clock speed, probably around 200-300 MHz, and a TDP of 8 W. The other is a dual core with a TDP of 4 W, so probably the same clock speeds, but only half as many cores and 1 CU.
For comparison, a Radeon HD 7750, AMD's lowest end desktop card of the current generation, uses the same GCN architecture, but has 8 CUs clocked at 800 MHz. Of course, it also has a very tablet-unfriendly TDP of 75 W.
You're not going to get performance anywhere near an A10-4600M in a tablet anytime soon. Or at least not in a reasonable tablet--meaning, to exclude things like the Razer Edge Pro that is a "tablet" thicker than some clamshell laptops and costs $1750 if you buy the accessories for it.
For a given architecture, performance is roughly proportional to power consumption. When you try to clock things really low, you can do somewhat better than that, perhaps cutting power consumption in half while only losing 40% of your performance. You can also do a little better that proportional if you have a different architecture designed for low power consumption. But laptops can easily handle 35 W, while for a tablet, you want something closer to 5 W.
As for when it will launch, AMD has promised Temash in the first half of this year. Some rumors say it will launch at CeBIT in March, and others at Computex around the start of June.
-----
Incidentally, TDP = Thermal Design Power. The listed TDP on a chip is basically the chipmaker's way of telling OEMs, if you build a cooling system that can dissipate this much power safely, we promise that will be enough. So "TDP at max load" should really just be TDP.