is hyperthreading gonna make a comeback?

[grimREEFER]

well, no core 2 cpu's that i have seen have it, is it gonna be a feature that remains only to northwoods and prescots?
are we ever gonna see it with conroe or kentsfield? 8 logical cpu's would be kickass.
just wondering is all

[uOpt]

I hate Hyperthreading passionately.

Sun's Niagara platform has it, too.

I certainly hope it never rises from the x86 grave again. Hopefully no chipmaker will be stupid enough to make a new chip that is as pipeline-challenged as Prescott again.

[GoThr3k]

I hate Hyperthreading passionately.

Sun's Niagara platform has it, too.

I certainly hope it never rises from the x86 grave again. Hopefully no chipmaker will be stupid enough to make a new chip that is as pipeline-challenged as Prescott again.
QFT

[nn_step]

Hopefully NEVER

[Pete]

I duno what the problem was with it to be honest but things have to move on in life

[Cooper]

HT has it`s good and bad sides. IMO Intel has alot of experience now with this tech - perhaps it just might return even more mature

[madgamer]

there was another thread where this was brought up that future extreme editions could have it. it might not even be called hypertransport (as it would be slightly different from HT 1.0), but fugger and a number of others backed it up. with shorter pipelines, conroe certainly doesn't need it like the old p4's, but performance is performance, so if they can get it to offer some gains, I dont see why not.

[nn_step]

there was another thread where this was brought up that future extreme editions could have it. it might not even be called hypertransport (as it would be slightly different from HT 1.0), but fugger and a number of others backed it up. with shorter pipelines, conroe certainly doesn't need it like the old p4's, but performance is performance, so if they can get it to offer some gains, I dont see why not.
lets see in Prescott is accounted for almost 60% of all the transistors (not counting Cache)
It offered less than Double performance in applications that were designed FOR IT.
Dual/quad cores are a superior design by nature.. it is inevitable for it to die off

[Jagz64]

I hope it does coz when i was using HT on my old socket 478 it made a hell of
a difference.

:) - I wonder how much HT will boost Core 2 Duo if intel implemented it.

[StealthyFish]

dual core will make more of a difference than HT. you have two physical cores, not one prioritizing tasks to make everything run more efficiently. =D

[accord99]

lets see in Prescott is accounted for almost 60% of all the transistors (not counting Cache)
It offered less than Double performance in applications that were designed FOR IT.
Dual/quad cores are a superior design by nature.. it is inevitable for it to die off
I doubt it was 60% of Prescott. It was less than 5% of Northwood which generally gained 20% with two CPU intensive threads/processes.

http://www.extremetech.com/article2/0,1697,1155017,00.asp

So while it may not return for desktop, where the nature of most user tasks rarely even use a second processor, it'll probably return for server-variants.



I hope it does coz when i was using HT on my old socket 478 it made a hell of
a difference.

:) - I wonder how much HT will boost Core 2 Duo if intel implemented it.
It's hard to say, it may be a bigger gain than on the P4, given the extra width of Conroe, additional cache and execution resources and fewer weaknesses.

[ted3]

HT is good for single core CPUs to improve multitasking responsiveness, but for a dualcore its a waste in my opinion, they are far more responsive anyway. Maybe they could use it on Celeron versions of Core 2 (Core 2 Solo?), at least it helps more than it does damage in a singlecore.

I remember the early tests of Pentium D and Pentium EE 840, the D840 outperformed the EE840 in some scenarios, and was close enough in most of the benchmarks that benefited from HT. Getting rid of the disadvantages and keep only the good things in a nextgen HT sounds somewhat impossible to me.

[cupholder2.0]

If they make single core Conroe they should use HT. HT really rocks on single core CPUs. On dual cores it doesnt make much difference tho :(

[DJ Krypplephite]

Well I suppose they could if they really tried, but the idea of HT was derived from the idea of Netburst - a long pipeline. HT is just stuffing another thread into the pipeline before the other one is done. Doubtful it'll be on a 14 stage, it wouldn't be the easiest thing in the world to engineer. However when you have over twice as many stages, yeah, I could see HT happening.

With the future of multicore processors having tens or hundreds of cores . . . ehh, I doubt HT's makin' a comeback.

[Fuji]

I say yes. Intel said that by the end of the decade we'll see cores that can crank out 8 threads at once. Anandtech was saying that given the vast resources that Conroe has, HT would make sense and would help boost performance and efficiency.

[wwwww]

well, no core 2 cpu's that i have seen have it, is it gonna be a feature that remains only to northwoods and prescots?
are we ever gonna see it with conroe or kentsfield? 8 logical cpu's would be kickass.
just wondering is all

My 955EE (ES) had hyperthreading. 2 physical cores each with 2 logical cores

[ethernal]

God, will the myth that HT is about long pipelines die? It has little to do with that. Dealing with long pipeline issues would be superthreading, not hyperthreading. Hyperthreading is executing two independent threads in a single clock cycle in multiple execution units to try to make full use of the execution units. Hyperthreading may help with long pipelines a bit, but it's far from the only reason why it exists.

Either way, I'm sure Intel has decided that HT shouldn't be on Conroe at this time, and I'm sure there's a good reason for it. Someone made a good post explaining possibilities as to why not in another thread, which I'm too lazy to find at this point in time.

[metro.cl]

HT isnt that bad :)

[FunkyRider]

HT not appear in Conroe is a problem of lowering performance/watt, as intel officially said

[death metal]

I'd like to see HT make a come back again, it made a significant improvement with my computing experience...

[grimREEFER]

yea, quake 4 gives a huge boost to northwoods with hyperthreading on as oppossed to off.
it also helps in a bunch of benchmarks....

[ea6gka]

ΗΤ was useful back in the days in 478 socket , and it made a hell of a difference compared to AXP's and early 754 socket cpus. I dont really understand why you guys hate it.

[JoeBar]

HT has a meaning only for single cores. In dual cores there's no need of it... ;)

[savantu]

So much missinformation at the start of the thread that it hurts.And I bet those people also didn't own a HT enabled CPU.

With SMT ( or HT for Intel ) you can better use the core's resources since the chances for a single thread to use all execution resources are very slim.As a result for a small die size increase ( 5-20% depending on implementation ) you can gain anywhere from 10 to 40% better performance and the most needed "system responsivness"

Some associated it with Netburst and the fact that in some few cases it offers no to slightly less ( -1-3% ) performance than a non-SMT enabled chip when actually it was SW's fault because it wasn't MT aware.
Look now : a Northwood or Prescott gives the boot to a single core A64 with similar rating in most new games. ( In Quake 4 a P4 540 acts like a A64 3800+ )

SMT has nothing to do with Netburst.Netburst has little execution resources and SMT was used more to hide latency when a thread stalled waiting for data.

On a CPU like Conroe , with lots of execution units , SMT should have a far bigger impact , problem is it also adds complexity.For the P4 it added 5% more die area but accounted for 95% of time at debugging.

My impression is that SMT didn't make it into Merom familly because of time to market restraints and because SMT adds extra power consumption ( since you use constantly more resources of the chip )

I would also bet we will see a comeback of SMT with the Nehalem generation of cores ( where the SMT maniack Andy Glew works for ) in 2008.

[grimREEFER]

good ole Andy

[Cooper]

Electronic News: Let’s shift subjects. How many cores will you have in your chips, how will they be managed and what will they be used for?
Pal Gelsinger: I don’t think we know how many there will be. That’s a topic of both research and product and market understanding. We’re on track. We have lots of duals out already, and we have quads and octs under development. Each of those cores can support multiple threads of execution. You can have 16 or 32 threads each.
Source (http://www.reed-electronics.com/electronicnews/index.asp?layout=articlePrint&articleID=CA6329160)

Pat Gelsinger, senior VP and general manager of Intel’s Digital Enterprise Group

So what do you think ? :cool:

[Shadowmage]

So much missinformation at the start of the thread that it hurts.And I bet those people also didn't own a HT enabled CPU.

With SMT ( or HT for Intel ) you can better use the core's resources since the chances for a single thread to use all execution resources are very slim.As a result for a small die size increase ( 5-20% depending on implementation ) you can gain anywhere from 10 to 40% better performance and the most needed "system responsivness"

Some associated it with Netburst and the fact that in some few cases it offers no to slightly less ( -1-3% ) performance than a non-SMT enabled chip when actually it was SW's fault because it wasn't MT aware.
Look now : a Northwood or Prescott gives the boot to a single core A64 with similar rating in most new games. ( In Quake 4 a P4 540 acts like a A64 3800+ )

SMT has nothing to do with Netburst.Netburst has little execution resources and SMT was used more to hide latency when a thread stalled waiting for data.

On a CPU like Conroe , with lots of execution units , SMT should have a far bigger impact , problem is it also adds complexity.For the P4 it added 5% more die area but accounted for 95% of time at debugging.

My impression is that SMT didn't make it into Merom familly because of time to market restraints and because SMT adds extra power consumption ( since you use constantly more resources of the chip )

I would also bet we will see a comeback of SMT with the Nehalem generation of cores ( where the SMT maniack Andy Glew works for ) in 2008.


Awesome post. Saved me some trouble typing it out myself. This guy knows what he's talking about :toast: