The Inquirer-Home

AMD and Intel speed up next generation CPU rollouts

Analysis At least something will be out in 2010
Wed Jul 21 2010, 17:51

AS SOON AS rumours were heard that both Intel's Sandy Bridge and AMD's Fusion initial entries might be announced well before the end of this year rather than early in 2011, expectations were raised that we might enter the CPU generational switch soon, or at least a bit earlier than expected. But what do the expected Sandy Bridge release at Intel's September IDF and AMD's follow-up Fusion release shortly afterward mean?

First, they don't mean as much as you'd think. Neither of these are high-end parts. The initial Sandy Bridge is the 32nm replacement for the current mainstream 45nm quad core Lynnfield CPUs in the LGA1156 socket. So yes, the LGA1366 six-core Gulftown Westmere processors will remain the nominal high end for a while more. Secondly, the new LGA1155 socket required by the new CPUs doesn't excite the 'mainboard upgrade' fans, but then they shouldn't make too much noise about it since Sandy Bridge CPUs provide much faster PCIe v2 connection to the South Bridge I/O to alleviate the old DMI bottleneck for handling USB3 or SATA3 additions. So, a mainboard upgrade might be worth it to get faster, more future proof I/O interfaces.

In fact, these initial Sandy Bridge Core i5 and i7 parts are, with their quad cores and integrated GPU, more likely to face off against AMD Fusion Llano parts directly, as both companies' new offerings couple four cores and tightly integrated graphics for the mainstream desktop and high end mobile use as well.

While the new Intel Sandy Bridge integrated GPU is expected - but not confirmed, mind you - to be somewhat slower than the Radeon HD55xx derivative squeezed inside the AMD Llano and Ontario Fusion chips, it has one supposed big advantage over the AMD parts: it is, well, "fused" together better with the rest of the CPU. Basically, the Sandy Bridge GPU core communicates directly with the four CPU cores over the shared fast - yes, quite a bit lower latency than Nehalem or Westmere - L3 internal cache, while the Llano GPU supposedly goes through the slower memory controller. Even if it is through the internal crossbar, it would be somewhat slower than sharing data with the CPU cores directly through the L3 cache.

On the other hand, all bets are off when it comes to the CPU speed comparison. If Llano uses full fledged Bulldozer core clusters, with dual integer plus single floating-point units in one block, the application performance versus Sandy Bridge will vary widely depending on the code profiles. The low power Ontario dual core Fusion chip, aimed at the ultra low power 10W - 20W subnotebook CPU market, will have an easier time, as it will fit between low-voltage versions of Sandy Bridge and the 32nm Atom refresh, yet will offer graphics comparable to the Nvidia NV240 GPU in performance, at least if you base the comparison on the Computex AMD Internet Explorer 3D demo.

The real fun, unfortunatelly, will be missed this year. It's the high end battle. Until this month, it looked like neither Intel nor AMD would rush with their top bin - that is high end desktop, workstation and server - new generation parts. These are the AMD Interlagos Bulldozer 12 to 16 core - in reality 6 - 8 "double integer, single floating-point" core blocks - high end G34 socket part to replace the "Magny Cours", as well as Intel's supposed top end Sandy Bridge Core i7 29xx and Xeon 5700 series 6 to 8 core Socket LGA2011 - yes, 2,011 pins for year 2011 - processors. We might have to wait for both even until next summer. Not an exciting prospect, especially for the bored-to-death tech press.

Right now, the rumours are that the release dates for both of these might end up to be a little earlier than this, possibly late spring. Why no rush? Well, from a high end user's point of view, the AVX instruction set extension - one of the key Sandy Bridge advantages - will take more time to be used on complex high end applications that need certifications than a game on a desktop may require. But AMD, right now badly squeezed by Westmere Xeons out of many profitable high end segments, should be keen to push the Interlagos out faster. Maybe that's what it is doing after all, and Intel, not keen to surrender any of the high margin markets, could then speed up the LGA2011 high end Sandy Bridge introduction. Wouldn't it be fun to have one (or, still better, two) next generation 6-core to 16-core CPU(s) in your high end desktop? µ

 

Share this:

Comments
@STOP IT

Firstly there is no need to swear like that, secondly do you honestly believe that the CPU just sits there doing all the work and that the socket only needs a power point and a monitor out?

You need to think before you post, im not going to list everything that the CPU doesnt do and needs access to because frankly i dont think you will listen but lets take a couple large example, PCI-E how many channels? where do they go? South bridge? memory channels?

If you increase any of those in any significant way they you will need to change where they connect to the CPU, you cant increase PCIe channels if you dont have anything to connect them to.

With the demise of the north bridge you will have to have socket upgrades more often, thats life, if you dont like it dont upgrade, or get over it. dont get me wrong im sure there is an element of forcing upgrades but this is not one of those cases.

posted by : Darren, 25 July 2010 Complain about this comment
Upgrades and Obselecence

If they are changing the architecture of how the CPU works, e.g. adding GPU architecture on-die or adding instruction sets, then even if it did 'fit' in the motherboard because it had the same number of pins, it wouldn't work. CPU manufacturers almost HAVE to change the pin counts when they make significant changes to the chips so people don't stick new parts into old MBs and wonder what's wrong.

This sort of whining happens every time there's a socket change, as if we are all surprised that they 'went and changed the sockets on us', even though it happens every couple years. Frankly, unless you need to have top-of-the-line performance, just skip every other architecture change, and get a new box every 3-4 years or so. The whole idea of 'future-proof' is nonsense, just get used to it.

posted by : JonB, 24 July 2010 Complain about this comment
STOP IT

I wish if at least journalists would stop justifying POINTLESS socket (and thus motherboard) upgrades.

How can Intel claim they care about the environment when I have to dispose of my old mainboard and get a new one just because of ONE FUCKING PIN?

Yes, that's right. There is no other advantage. There are no new features, no smarter design, nothing. Everything that affects performance is in the CPU now. Mainboard is just a CPU adapter to the mains outlet and the monitor.

And yet you should pay $200 for a new one. WHY???

WHY CREATE MORE TOXIC WASTE BECAUSE OF ONE PIN???

posted by : Me, 23 July 2010 Complain about this comment
More cores the better

Bringo n the 16 core chips I will put them to work! Don't cry if your noob apps don't use your quad core cpu... why the F would word be multi threaded anyway? Quit crying, save yourself some cash, and get a Dual core

And why the f*ck would AMD but an Nvidia core on their chips when they own ATI? Retard... besides, ATI nVidia :P

posted by : Dan, 22 July 2010 Complain about this comment
For netbooks?

Sounds like Ontario fusion is for netbooks, but with 5500ish graphics. Battery life? Don't know, sounds interesting though.

posted by : hmm, 22 July 2010 Complain about this comment
Multi/Single thread mix problem

The problem is that a lot of code is now mixed some multi threaded some not.
this can pull down the use of the 16 cores significant.

Let assume that 90% of your workload is multi threaded (very optimistic)
but 10% is still single threaded and sequential so you have to wait for it before the next can be taken like de/encoding or simulations.

The time that all 16 cores are active is equal to:
1/(16/0.9-16+1)=0.36 or 36%!!! or about 1/3 of the time, 2/3 it is working single threaded!

if you run a lot of heavy application at the same time it would have an advantage though.

posted by : kedas, 22 July 2010 Complain about this comment
16 cores. Quite frankly, no

"Wouldn't it be fun to have one (or, still better, two) next generation 6-core to 16-core CPU(s) in your high end desktop? "

Quite frankly, no. I am still lying in the gutter (had to run back home to type this) after buying a quad cpu some three years ago and am still flabbergasted by the dearth of applications using more than 1 (one) CPU.

Notepad, no. Word, no. But those weren't expected. But even some mathematics oriented programs have had a hard time stepping over the multicore line. Several games do it now, some better than others, and some still don't at all. It is a misery for most of us. A hype out of proportion for just a little to no good at all, unless you like sorting a 43 TB logfile of a chat-server.

But, WTF, gimme that damn 16 core CPU and I'll try it out. It might get me chicks.

posted by : Dr No, 22 July 2010 Complain about this comment
Where is nVIDIA Fusion with AMD

I want to see nVIDIA version of APUs or AMD CPUs on die with nVIDIA GPUs. Competition is good, right?

posted by : Sotka, 22 July 2010 Complain about this comment
aboutus
Advertisement
Subscribe to INQ newsletters
Advertisement
INQ Poll

Authorities in several countries raided Megaupload recently, shut down all of its services, seized hundreds of servers and arrested several of its executives on criminal charges.

Do you think the move was justified?