Slashdot Mirror
Intel Plans for Dual-Core Prescott CPUs in 2005
scapermoya writes "X-Bit Labs is reporting that Intel is planning to step up their introduction of dual-core processors, with the first chips to hit the market in late 2005. Intel announced this plan at the Technology for Business Today seminar, held in Washington, D.C. Looks like NetBurst is sticking around, despite what we have heard lately about a move toward the 'M' architecture. Supposedly, thanks to HyperThreading, the OS will see 4 installed processors. Snazzy."
UPDATE: A representative for Intel Corporation told X-bit labs the company had never released any precise details in regards the dual-core strategy. The information published herein should not be considered as based on official statements.
WTF?
Ars is covering this too. Ken Fisher makes it a point to mention that the person who made the claims is in marketing. He also speculates, quite logically, that bringing out dual core Prescotts in '05 would be a feat even for Intel. Worth reading for a more sobering take on the situation.
Have a Happy.
XP Pro understands hyperthreading fine, and in fact will work with a machine with dual hyperthreading CPU's. What won't work is 2k Pro. Also note that 2k3 server standard reduced the max number of physical CPU's to two for standard edition whereas it had been 4 in previous iterations of NT. Btw don't use hyperthreading on win2k even if you have enough processor licenses because it will balance evenly across the hyperthreading CPU's not realizing that they are really there for spare capacity and hence trash the instruction cache.
There are 4 boxes to use in the defense of liberty: soap, ballot, jury, ammo. Use in that order. Starting now.
Currently they license based on physical cpus, not the logical number. They had to make some minor changes when Intel introduced hyperthreading.
My guess is initially it will count as 2 cpus and as it becomes more widespread they will revert to it being counted as 1. A true dual system (4 cores, 8 logical) would be counted as 2.
Of course in another few years they will be feeling more pressure from linux so they may change the licensing a lot.
Windows XP handels Hyperthreading, so you can have 2 physical processors and 4 logical processor according to microsofts hyperthreading document.
Licensing is for physical processors only. So, even today, if you have 2 physical CPUs with hyperthreading, you are compliant. Task Manager will show 4 CPUs, but the OS can determine that only 2 are physical.
I suspect, however, that a dual-core CPU will be treated as 2 physical cpus...(+2 virtual CPUs)
Yes, it's true. This man has no dick.
XP 2003, aka server supports more - if multi cpu setups were at all common in the desktop workstation market then this would be rolled up into XP Pro, but it isn't.
Pretty much the same thing has applied throughout the NT Lifecycle, with Workstation sticking to single / dual setups and Server handling 1/2/4/8 setups.
2k server does, mind. i've got six DL380's in front of me with dual HT xeons, and task manager shows them as 4 CPUs.
2; you only get 4 by a bit of extra hardware to virtualize one as two. Hyperthreading is just exploiting the already superscalar architecture a little more.
:)
HTT has a transistor count overhead of ~5%; dual core is over 100%
You seem to be forgetting the fact that your CPU takes that much power only when it's actually working, and not when it's sitting idle. So unless you got SETI running, your power bill shouldn't be that high. Ofcourse if you have SETI running, then it could also be the aliens drawing power from your computer.
Yes, but if you had a 4 ways system with HT enabled you would really only be using the first two CPU's and their HT units, the other two physical processors would be doing nothing. Not to mention that the OS and older apps will naively balance across all 4 'CPU's' evenly despite the fact that the HT units are not fully capable processors and in fact can easily degrade performance by trashing the contents of the instruction and more importantly data cache. MS claims that windows 2000 will properly use the physical processors before the logical ones if the system is written to Intel BIOS specs but my real world experience says that there aren't a lot of correctly configured systems because I've had to restage 4 way servers from Dell, IBM, and HP and 2 way workstations from HP and Dell.
There are 4 boxes to use in the defense of liberty: soap, ballot, jury, ammo. Use in that order. Starting now.
I prefer Ars Technica's Understanding Pipelining and Superscalar Execution...
Besides, I feel HT exploits the fact that the processor is pipelined more than its superscalar nature.
tasks(723) drafts(105) languages(484) examples(29106)
Well, a few weeks ago I had to install 3 test servers - but turned out we have only 1 available server.
I almost freaked out - but then I realised that it's a dual-Xeon box ---> 4 CPUs (to the OS). Using an obscure switch in VMware config file, I manage to assign each CPU to a (virtual) Server. The fourth CPU is assigned to the host OS.
Everyone was very happy with the result, and looking forward to utilise more of this feature (and this kind of CPU) in the future.
Actually, the big jump in heat for the Prescott cores is from Intel use of only starined silicon in manufacturing. By creating a strained lattice for the silicon, you increase the likelyhood of current leakage (hence more heat). This is why AMD and IBM went with silicon on insulator and added strained silicon later (the SOI process helps to mitigate the leakage in strained silicon).
Here's a simple primer
[RIAA] says its concern is artists. That's true, in just the sense that a cattle rancher is concerned about its cattle.
The chip all by itself is $20,000 and is about the size of ones hand. It is 4 physical, 8 logical with 144 MB (that's right...megabytes) of cache.
[RIAA] says its concern is artists. That's true, in just the sense that a cattle rancher is concerned about its cattle.
Yes, meanwhile IBM is on their second generation dual core chip, POWER5 -- now available in eServer i5 systems -- shipping TODAY.
The way POWER4 was packaged for the higher end boxes, you have what they call a Multi-Chip Module (MCM) with 4 POWER4 processors on-board. This means each MCM was an 8-way.
Now, for POWER5, they have added the Dual-Chip Module or DCM. With the i5 model 570, you can get a 1/2 way or 2/4 way box. If you buy the 1/2 way, you have one DCM installed...and if you buy the 2/4 then you get two DCMs.
POWER5 has what IBM calls Simultaneous Multithreading -- SMT, which is the same type of idea as Hyperthreading. Essentially if the application supports multithreading, it will functionally see twice the processors...but this is a logical thing...a 4 way is still a 4 way...not an 8 way.
Now, having said all that....never underestimate IBM development labs. I hear POWER6, 7 and maybe 8 are already out in development.
TGM.
Actually I believe you are wrong here. Intel split the reorder-buffer in half, split the physical register file in half, and fetches from each thread every other cycle. If I am not mistaken, the latter point means that each thread gets equal execution resources as once instructions have been fetched and decoded, dispatch and execute don't care which thread an instruction belongs too. Like I said, I think this is the case but am not sure.
From my friends in the architecture community, Intel's SMT implementation is sort of half-assed.
On the other hand, IBM's Power5 also fetches from each thread every other cycle, however it shares a reorder buffer physically (but of course not logically).
that's how it works. 2k server doesn't support over 4 CPUs, advanced server does - it's one of the reasons it's called advanced...
XP Home also sees it as two processors. I used a Sony computer at Sam's Club that had a P4 3.06 HT. I was really surprised when I saw it was XP Home. Task Manager did show two processors and allowed me "set affinity" and everything.
Win2k will work with hyperthreading, it just doesn't work well. You'll probably see better performance if you turn it off, since Win2k doesn't realize there's only one physical CPU and thus doesn't balance the loads properly.
Yes
Modder processors use MOSFET transitors in CMOS arrays (Complimentary PMOS and NMOS networks in each gate)
This which means the only time power is expended (And therefore heat created) is when a gate transisions (if a gate stays the same across multiple clocks heat is only produced at the transition into that state).
So the more clock cycles you have the more often it's probably going to be switching states - each gate creates miniscule heat and power dissipation, but there are a lot of gates.
It is true that if you have two cores you have twice the heat - but that heat is also spread out across twice the area, less concentrated heate is easier to cool. Why do you think heatsinks spread the heat out all over the place.
A dual core processor running at 2Ghz would be roughly equivilient to a single core at 4Ghz - so the demand on clock rate increase is reduced and heat created/time is lower.
If you cannot keep politics out of your moderation remove yourself from the Mod Lottery.. NOW!
I should have read the post a little better.
I was referring only to HT capable single CPU setups, not dual CPUs or a single CPUs with dual cores.
Of course, Intel/MS could do something to make XP Home treat one dual core CPU the same way it treats a HT CPU now.
No, the point of hyperthreading is to expose functional units that are not being used by the main process('s) running on the physical CPU. If you naively schedule everything across both 'CPU's' you will end up with stupid things like running parallel versions of a tight integer loop which is already maxing out the integer calculation units thus polluting the data cache and stalling the pipe MUCH more often which is a BAD thing on the P4. For more info on scheduler tweaks to accomodate HT I suggest you see this LK post by one of the Linux scheduling gods =)
There are 4 boxes to use in the defense of liberty: soap, ballot, jury, ammo. Use in that order. Starting now.
I'm oversimplifying for sure, but aren't the heat issues (and other more difficult quantum effects) primarily due to the ever increasing demand for clock speed?
For the most part, yes. Power density is a problem. Increasing L2 cache sizes is an easy way to increase performance without adding much in terms of high power density... it's mostly static power consumption. But, this is changing too as leakage power increases with shrinking dimensions.
By going dual core, you save power with pads and you can share the L2. There's also the reduced cost of motherboards by not having extra sockets, pins, VRMs, caps, etc., and increased computing density, yadda yadda.
As a layman it kind of makes sense to put 2 lower speed cores on a die rather than one faster one, and get lower power consumption and more importantly less heat production, and let the software deal with utilising it?
Very good point. One thing to keep in mind, though, is Amdahl's Law... single-threaded performance still matters even for parallizable code, and parallel programming is a pain in the ass.
But, as Power 4 and 5 decided to put two complex cores onto a single die, there was an old project by Compaq/Alpha called Piranha that used 8 simple cores on a die (I believe they were Alpha 21164s, when 21264s were the hot thing). It was optimized for database/transaction code, since there's a ton of parallelism in db code and you can get a lot of performance improvements by reducing communication misses (and communication miss latency) and increasing L2/L3 caches (which reduces communication misses). Here's the link to the paper.