Dual-Core Pentium 4 Slated For 2Q 2005

Quantrell writes "Today is the first full day of the International Solid-State Circuits Conference in San Francisco, and Intel has announced that dual-core Pentium 4s are coming in the second quarter, one in the Extreme Edition line (no surprise there), and also the Smithfield Pentium 4 800 series, which is the next so-called consumer desktop line. No word on pricing, yet."

Lack of bandwidth?

[ajiva]

While dual processors is great and all, I'd rather see double the memory bandwidth then double the processing power. In the case of Intel processors (especially duals) memory bandwidth is severly lacking, and while DDR-2 should help a bit, I don't expect to be that impressed with the new dual cores.

Re:What kind of sockets will there take?

[David Ziegler]

That must be because you didn't read the article:

The Pentium Processor Extreme Edition will be combined with a new chipset named the Intel® 955X Express chipset, formerly codenamed "Glenwood," that includes features such as Intel® High Definition Audio, PCI-Express and faster dual-channel DDR-2 memory.

Intel will also couple its mainstream "Smithfield" processor with two new chipsets named the Intel® 945G Express chipset and Intel® 945P Express chipset, both previously code-named "Lakeport" in the second quarter of the year.

And:

And before you ask, the new dual-cores require new chipsets, so these CPUs will not run on existing systems. AMD fans, meanwhile, can still hope that their dual-core parts will run in Socket 939.

Re:Question

[bbrack]

since 90nm fab times should be on the order of 4-5 months, I'd assume Intel is running these in fairly high volumes (and has been for a month or so)

Re:Well...

[mako1138]

Not really. Intel has been playing catchup all this time, first with 64-bit and now with dual core. Opteron was built from the ground up to support more than one core, which is the beauty of it.

Here's a long discussion on the current dual core situation on Ace's Hardware [aceshardware.com]. (They use a lot of codenames. "Smithfield" I think is what this /. story refers to. "Yonah" is somewhere in the future.)

Reading around though

[Anonymous Coward]

AMD has been providing working real dual-core samples to partners for months, whilst dual-core Intel processors are apparently in short supply.

This smells of Intel running to get there first before AMD, so they aren't second again with a technology.

HyperThreading is disabled in the Smithfield dual-core product too, so expect a mere 50% overall performance increase at the same clock speed (2.8GHz, 3GHz, 3.2GHz soon afterwards) for Intel. AMD stand to gain more from dual-cores, as they have no HyperThreading equivalent at the moment, and AMD have said that dual 2.4GHz will be possible, that's two 4000+ rated processors, probably overall performance of 6000-7000+. That's a bit better than the 5000+ performance from a dual core Smithfield.

Dual core AMD will likely perform a lot higher than dual core Intel therefore.

Re:What kind of sockets will there take?

[Zocalo]

I wouldn't be shocked at all if there's an LGA775 dual-core CPU.

The question then becomes one of how well the existing motherboards would cope of course. The ideal would be for it to be entirely transparent and the dual cores are handled by the CPU in a similar manner to hyperthreading. If that's not possible then we'd be looking at a BIOS update at least, and even then it might not be possible to maximise the benefit of dual cores with out a motherboard designed for the purpose.

In any case, with AMD in a similar situation with its own upcoming dual core CPUs, it's going to interesting to see how the two companies approach it. There's going to be some unhappy customers if one company manages to enable upgrades to dual cores on current motherboards and the other doesn't, that's for sure...

Re:Lack of bandwidth?

[exley]

In that case, you'll have to talk to memory, chipset, and board people. The memory interface is still in the northbridge/MCH (memory controller hub) on Intel-based architectures, as opposed to AMD's x86-64 offerings which have memory controllers integrated into the processor. Also, when it comes to memory, bandwidth isn't the only issue; latency is also critical.

Re:Good explanation of how this will actually help

[Chirs]

It works exactly the same as an SMP system. Any OS capable of handling SMP will be able to handle this, including WinXP Pro (but not "Home").

As for app support, any time you're doing a task that is parallelizable, you may be able to benefit.

If you are running two totally different processes at once, then you get immediate benefits. (And immediate subtle bugs, if the processes share resources and weren't properly written for SMP).

If you are running a single multi-threaded app, you get immediate benefits. (And immediate subtle bugs, if the app wasn't properly written for SMP).

If you only run a single app, and that app has only a single thread, then you will not gain much at all.

Re:What kind of sockets will there take?

[mako1138]

AMD fans, meanwhile, can still hope that their dual-core parts will run in Socket 939.

Kevin McGrath (AMD's chief architect of x86-64) gave a talk about dual cores at my school last month. I asked him if 939 would support dual cores, and he said it would, though he didn't have a timetable. He also reiterated that we'll be seeing dual cores coming on all product lines.

Part of the reason AMD can do this, I think, is their discipline in keeping a consistent power envelope, so the motherboard and heatsink manufacturers don't have to scramble to support a new incredibly hot processor. I anticipate that Smithfield will require massive cooling.

The other reason is the memory controller is built onto the processor, as opposed to Intel's traditional arrangement of it being on the North Bridge of the chipset. Thus no change of chipset is needed (in theory).

Re:... questions ...

[djohnsto]

It will not use socket 478, but it may use LGA775. Dual-core P4's will let you execute 2 simultaneous threads at about 1.5-1.8X speed they would run on a single core P4 (given the same clockspeed). Single-threaded apps will not see a performance improvement (although you could run 2 single threaded apps and get an aggregate improvement). These will probably also be 64-bit enabled.

If you want dual-core, I would imagine Intel's will be cheaper than AMD's at intro. The Smithfield processor is in their performance mainstream segment (i.e. same as current Pentium 4 - not Xeon). AFAIK, AMD will intro dual-core with their Opteron line. Not sure when it hits the Athlon FX / Athlon 64 line.

Re:What kind of sockets will there take?

[ip_fired]

There is a good article on AMDZone [amdzone.net] that talks about how AMD is doing it's dual-core. They have put 2 cpus each with their own cache but they share the on-chip memory controller and the hypertransport links. According to AMD, there is only a 10% loss in performance by using the shared components.

Not bad considering you only have to have one socket on your motherboard to accomodate a dual processor system now. And it will even work in current motherboards, using the same 939 socket.

Re:Question of OS Software compatibility.

[ePhil_One]

Also on that same note, if you have a dual core proccessor with hyper threading, creating the illusion of 4 cpu's would this be able to run under MS windows Pro or would it need a "server" edition version of software which supports 4 CPU's

Windows 2000 SP4 and Windows XP Pro both run fine on a Dual Xeon P4 w/ HT enabled. Task Manager sees 4 CPU's as expected. Pre-SP4 systems might complain because they are unaware of Hyperthreading, but I think MS had not really gone into the overkill mode that highlight XP. Prior to XP they were pretty trusting of folks, license limits were managed via trust. In other words is an 11th client tried to connect it worked, instead of rejecting you with a nasty message about how you need to upgrade to server edition.

Re:What about Windows licensing issues?

[dabraun]

Windows XP recognizes the difference between hyperthreading and seperate processors. I have XP Pro running on a dual P4 with HT enabled - that is, task manager shows four processors. This works fine. I'm not sure if the code to tell the difference was ever backported to Win2K (which shipped before HT existed.)

Odds are that they will do the same thing for multi-core (i.e. not count each core but rather each chip - I am sure that Intel and AMD will provide a way to tell the difference just as they do with HT.)

With HT there's actually more to it than just the lisencing issue - XP treats HT processors differently - it knows that if proc 1 and 2 are really the same chip / proc 3 and 4 are the other chip that given two threads it should prefer to run them on seperate physical processors when possible. I am not sure if issues like this would apply to multi core but they might since the two cores will likely still share some things.

Pricing

[qtothemax]

No word on pricing yet

This news bit [digitimes.com] had been posted on anandtech a bit ago, and seems decently reliable and realistic. 2.8ghz for $241 isn't bad at all, pricing is right between today's prices for a 3.2 and 3.4. I personally though am waiting for AMD's dual cores which will supposedly work on my current motherboard, though it looks like at first the only dual core will be an FX processor, with the insane price that goes with that.