Dual Core Intel Extreme Edition to have Hyperthreading

[USMC2Hard4U]

Intel plans to deliver two separate dual-core products and dual-core-enabled chipsets for its Pentium® processor-class families in the second quarter, including the Pentium® processor Extreme Edition. The Intel® Pentium® Processor Extreme Edition will include Hyper-Threading Technology, providing the ability to process four software "threads" simultaneously.

w00t!

 

[StArWaRrIoR]

I saw that earlier, but the fearsome factor in this equation is the cost of the processor. I can easily see that being over $1,000.

I guess I'll have to wait and see.

 

[tornadotsunamilife]

with this hardware, it seems more like a marketing feature than a useful feature. but i may be proven wrong when some benchamrks get released

 

[DR_K13]

I can put 4 instances of Folding@Home on that Computer.......Sweet!!

 

[Ballz2TheWallz]

I can put 4 instances of Folding@Home on that Computer.......Sweet!!

now if only we could have a quad dual core HT processor solution,we could run 16instances

 

[lithium726]

is it just me, or does 4 logical processors on one 800mhz FSB sound horribly bottlenecked, so much so that it wouldnt give much real world performance? not to mention that with processors you run into diminishing returns and you wont gain nearly as much going from 2 to 4 (logical) processors as you would going from 1-2

itll definatly be interesting to see what these things, with all that cache im really intereted in the wafer size and the heat output.. prescott is already bad enough, and there is no way in hell they are gonna fit two 13u gallatin cores on a single chip.

 

[robberbaron]

is it just me, or does 4 logical processors on one 800mhz FSB sound horribly bottlenecked, so much so that it wouldnt give much real world performance? not to mention that with processors you run into diminishing returns and you wont gain nearly as much going from 2 to 4 (logical) processors as you would going from 1-2

itll definatly be interesting to see what these things, with all that cache im really intereted in the wafer size and the heat output.. prescott is already bad enough, and there is no way in hell they are gonna fit two 13u gallatin cores on a single chip.

Pretty sure the EE's are all going to be >1066 FSB from here on out. That would be 266MHz of bandwidth for each processor, just like a Thoroughbred right?

 

[xonik]

That's beside the point. In any case, the shared cache and memory bandwidth will be some annoying bottlenecks, though.

 

[robberbaron]

That's beside the point. In any case, the shared cache and memory bandwidth will be some annoying bottlenecks, though.

Interesting how they put a nice feature on it that you don't even have to turn on, and all you people are doing is just nitpicking the hell out of it.

I somewhat agree though. It's just that intel isn't about to swallow their pride and move to an integrated memory controller and use HyperTransport. It's not like AMD has an exclusive licence on it or anything, but it's not something Intel would do.

That's beside the point. In any case, the shared cache and memory bandwidth will be some annoying bottlenecks, though.

The cache wouldnt be too bad. Each Gallatin has 2MB of L3 cache, so 2x2MB/4= 512Kb for each virtual processor

 

[lithium726]

Pretty sure the EE's are all going to be >1066 FSB from here on out. That would be 266MHz of bandwidth for each processor, just like a Thoroughbred right?

ah, right, forgive htat, im used to p4's having the 800bus, completly forgot about the 1066..

still, wtih two HT netburst cores, thats gonna be some seroius bottlenecking

The cache wouldnt be too bad. Each Gallatin has 2MB of L3 cache, so 2x2MB/4= 512Kb for each virtual processor

the dual core EE will almost certainly not be a gallatin, the wafer would be gigantic (relitivly speaking)

 

[robberbaron]

ah, right, forgive htat, im used to p4's having the 800bus, completly forgot about the 1066..

still, wtih two HT netburst cores, thats gonna be some seroius bottlenecking


the dual core EE will almost certainly not be a gallatin, the wafer would be gigantic (relitivly speaking)

Then that means that they'll either
A) Use the prescott core which as of now has the 800MHz FSB and just stick a bunch more cache to it like they did with the Northwood EE
B) Make a totally new 90nm that runs at 1066MHz fsb.

 

[xonik]

Interesting how they put a nice feature on it that you don't even have to turn on, and all you people are doing is just nitpicking the hell out of it.

That's what we do in a tech forum...

The cache wouldnt be too bad. Each Gallatin has 2MB of L3 cache, so 2x2MB/4= 512Kb for each virtual processor

1 MB*

While essentially true, the problem with shared cache is that the threads have to wait in line in order to use the shared cache. So it's not 1 MB per processor as you imply, but rather, 2 MB per processor, but only one access at a time per cache pool. And I am not sure that the two caches can be accessed asynchronously.

 

[Stugots]

Pretty sure the EE's are all going to be >1066 FSB from here on out. That would be 266MHz of bandwidth for each processor, just like a Thoroughbred right?

thoroughbred's had 2x133mhz, these will have 4x266mhz...

 

[lithium726]

thoroughbred's had 2x133mhz, these will have 4x266mhz...

he was kidding...

 

[Dan_D]

Both the 800 series and the Extreme Edition will support HT. However the EE will be 1066FSB and the 800 series will be 800MHz FSB.

As far as the dual core being bottlenecked I don't think it will. Here's why. The Xeon right now shares a memory bus, and shares all the I/O of the motherboard, still the dual and quad Xeon's have kickass performance in multi-threaded applications. Therefore, there will be no difference between a dual core Pentium 4 and a Dual Processor Xeon. You'll have 4 virtual CPU's in either case. The Pentium 4 however, should be faster as the bus between the cores will no doubt be MUCH faster.

I think this will actually be a fairly good performance processor. With some outstanding cooling, we can no doubt expect stellar results, plus the new 955x chipset will offically support DDR667 ram which will be a nice performance boost there.

Intel might actually be getting it's shit together. No one thought the Pentium 4 would go anywhere until the Northwood came out. Then it was a very attractive enthusiest CPU. Then Prescott came out and fell short of expectations. Now I think Prescott will get implimented a bit better in it's dual core form and we should be looking at pretty good performance.

I look forward to the dual core Smithfields, but even so I'd still rather see a dual core Dothan with 4MB L3 and SSE3 instruction support. Add in EM64T and you'd have a badass chip sitting on the 955X chipset.

 

[USMC2Hard4U]

I seem to think this processor will rock. It will allow for you to not skip a beat when it comes to multitasking.

 

[rayman2k2]

That's what we do in a tech forum...1 MB*

While essentially true, the problem with shared cache is that the threads have to wait in line in order to use the shared cache. So it's not 1 MB per processor as you imply, but rather, 2 MB per processor, but only one access at a time per cache pool. And I am not sure that the two caches can be accessed asynchronously.

xonik, im pretty sure that each core has its own individual 1MB cache.


i cant wait for pressler though, i already planned it as my first foray into dualcore

 

[Dan_D]

xonik, im pretty sure that each core has its own individual 1MB cache.


i cant wait for pressler though, i already planned it as my first foray into dualcore

Don't know exaclty how Intel will impliment the cache, but it might be better if there was 2MB of dedicated cache, and each processor took what it needed based on the workload it was assigned. So if core 1 needed 1.75MB of cache and core 2 needed only 128K cache, then I would think that would be more efficient. Assuming of course the processor and the OS were capable of such proper balancing.

 

[lithium726]

Don't know exaclty how Intel will impliment the cache, but it might be better if there was 2MB of dedicated cache, and each processor took what it needed based on the workload it was assigned. So if core 1 needed 1.75MB of cache and core 2 needed only 128K cache, then I would think that would be more efficient. Assuming of course the processor and the OS were capable of such proper balancing.

but if they both wanted 1.75mb they would have to fight for it, and that *might* increase latency.

of course, we are talking about two 3.73ghz cores, i dont think that would hurt much at all in real world performance...

 

[Redefined]

wait im a little confused, will the dual core P4 EE be 4mb of L3 cache? or just same old 2MB. and what do u mean you can run 4 things of seti@home.

 

[lithium726]

wait im a little confused, will the dual core P4 EE be 4mb of L3 cache? or just same old 2MB. and what do u mean you can run 4 things of seti@home.

2mb, we think. i highly doubt itll have 4mb, that would skyrocket the price of the chip, even higher than it already is

the 4 instances of f@h referes to the fact that with dual hyperthreading cores, you will have 4 logical processors. you run one instance of the client for each processor, so thats 4 instances of f@h

 

[Redefined]

2mb, we think. i highly doubt itll have 4mb, that would skyrocket the price of the chip, even higher than it already is

the 4 instances of f@h referes to the fact that with dual hyperthreading cores, you will have 4 logical processors. you run one instance of the client for each processor, so thats 4 instances of f@h

but i thought dual cores means that the cpu will have 2 in 1. not 4 in 1?

 

[tornadotsunamilife]

but i thought dual cores means that the cpu will have 2 in 1. not 4 in 1?

if you have a P4 and use HTT you theoretically have 2 processor's (your computer will list 2 processor's if you check your system information). so with with 2 cores (basically 2 processors on one piece of silicon) and HTT enabled you will theoretically have 4 processors.

 

[lithium726]

if you have a P4 and use HT you theoretically have 2 processor's (your computer will list 2 processor's if you check your system information). so with with 2 cores (basically 2 processors on one piece of silicon) and HT enabled you will theoretically have 4 processors.

fixed. HTT is hyper transport

its not 4 in 1, which is why they are called logical processors. the silicon physically has two cores in it, which is, yes, two processors. if both of these cores have hyperthreading technology, its going to emulate two processors on each core. viola, 2 physical cores and 2 logical processors.

 

[tornadotsunamilife]

i always thought HTT stood for HyperThreading Technology, well at least that's the way i've used it

 

[Jonsey]

I seem to think this processor will rock. It will allow for you to not skip a beat when it comes to multitasking.

Seems like I remember you being very excited before the prescott came out. Hopefully this won't disappoint like the prescott did.

 

[lithium726]

i always thought HTT stood for HyperThreading Technology, well at least that's the way i've used it

ive always thought that HTT stood for hyper transport tech, since intel was the first to the "HT" moniker with hyperthreading tech, they got stuck with that, at least thats how ive always seen it abbriviated.. and amd got HTT for hypertransport tech

 

[tornadotsunamilife]

ive always thought that HTT stood for hyper transport tech, since intel was the first to the "HT" moniker with hyperthreading tech, they got stuck with that, at least thats how ive always seen it abbriviated.. and amd got HTT for hypertransport tech

not to worry. now i know for next time

 

[xonik]

Okay, neither abbreviation is official, so this discussion is pointless. Just think of the context of the usage and everything will make sense, or just don't abbreviate.

 

[hardwarephreak]

Dual cores....require a dual CPU compatible OS? 2000/Server/XP Pro/Linux...now with four virtual CPUs...

 

[lithium726]

yes, it will require a dual capable OS. however, microsoft has already stated that licensing of its OS will be done on the physical processor level, so if you have dual cores that have HT, you can still use XP Pro.

Dual cores....require a dual CPU compatible OS? 2000/Server/XP Pro/Linux...now with four virtual CPUs...

 

[freddiepm61]

the dual core desktop (as opposed to server parts) have as much cache as their single cored brethren, and they just share it.
double the cache would add a huge amount of die size and cost

 

[tsuehpsyde]

wait....I remember hearing that in the first stages of dual-cores, each core would run half the speed of a single core now, equaling the GHz of a processor out now...since the heat output of 2 cores would be a bit much? So....is it say 2x1.7GHz cores for 3.4, or 2x3.4 for a 6.8GHz (combined speeds, of course)?

 

[Sc4freak]

Intel's dual core CPUs will come in 2.8ghz, 3.0ghz and 3.2ghz versions. Each core runs at 2.8/3.0/3.4.

 

[bountyhunter]

yes, it will require a dual capable OS. however, microsoft has already stated that licensing of its OS will be done on the physical processor level, so if you have dual cores that have HT, you can still use XP Pro.

since everyone is paying to licesne it in the beginning.

If you have 4 cpu's then, IMO there's no reason you should be running windows.

 

[USMC2Hard4U]

Intel's dual core CPUs will come in 2.8ghz, 3.0ghz and 3.2ghz versions. Each core runs at 2.8/3.0/3.4.

This is correct...

The Low end dual core upon release is 2.8Ghz, Per core... The High end upon release is 3.2Ghz Per Core

 

[lithium726]

since everyone is paying to licesne it in the beginning.

If you have 4 cpu's then, IMO there's no reason you should be running windows.

what does number of CPU's have anything to do with what OS you are running? if i had a dual core EE youd better believe id be using XP, just the same as i would if i had a 1.7ghz willamette or 1.4ghz tbird

 

[xonik]

Memory management and thread management, I would imagine.

 

[FreiDOg]

what does number of CPU's have anything to do with what OS you are running? if i had a dual core EE youd better believe id be using XP, just the same as i would if i had a 1.7ghz willamette or 1.4ghz tbird

Well, schedualling is the big one,
but WinXP (unlike the default install of the Linux Kernel right now) is actually an HT aware OS, so in theory it should do proper load balancing between the cores.

in a non HT aware enviroment, the OS would simply see 4 cpus, CPU:0 and 1 would be on the two logical CPUs on core:0, and CPU:2 and 3 on core:1.
If the schedualler isn't HT aware, it would load balance the logical CPUs, not the physcial cores (or CPUs if you're on a dual Xeon platform now), so if CPU:0 were at 100%, and CPU:1 were at 100% (1 thread each), with CPU:2 and 3 being idle a non HT aware OS would see that as balanced, even though in reality Core:0 is sharing 2 high priority threads, while Core:1 sits completely idle. Which is of course a waste, the threads should be spread equally across physical chips (or cores), not logically ones.

It's a bigger issue when you get to multiple physical chips like 2P Xeons, where you need to balance and account for I/O per chip, not just load. But with dual core, I/O is still the same as with a single CPU, all the extra balancing should be artbitrated internally.

There are other things, like how strong the schedular actually is, can it properly manage 4 simultaneous threads, that's not a workload a home OS has traditionally had to keep up,

how well it manages memory, and paging, 4 times the active threads, 4 times as much space in ram (on average) is needed. That means the OS probably should be more agressvie in pushing idle / blocked threads to the page file.

 

[Big Worm]

wait....I remember hearing that in the first stages of dual-cores, each core would run half the speed of a single core now, equaling the GHz of a processor out now...since the heat output of 2 cores would be a bit much? So....is it say 2x1.7GHz cores for 3.4, or 2x3.4 for a 6.8GHz (combined speeds, of course)?

Pretty sure they wont be cut in half each speed wise...

But even if they are running the full 3.4 each, like always since the beginning of computer time, you dont add the speeds together.