Hyperthreading

[dwhore]

I'm trying to work out of the hyperthreading on the i7 is going to make a big difference to me.

I run 8 VMWare sessions simultaneously, mostly idle, but it's possible up to three could be doing hard-core processing at any given time. This is on top of usual desktop applications.

Will it benefit my VMs to have '8' logical processors?

 

[Cogman]

Yes, it will. Hyperthreading provides a speed increase in Multithreaded applications at the cost of single threading speed (at least, the old Hyperthreading did). If you are running many different applications, then you most likely will see a speed increase from hyper threading.

 

[aigomorla]

Originally posted by: dwhore

Will it benefit my VMs to have '8' logical processors?

i7 with hyperthread will almost = 2 yorkfields @ same clock.

Almost. its like 1.75x

So yes, you will benifit from 8 tasks.

 

[AlexWade]

Didn't hyperthreading have a huge negative effect on server applications? I have to wonder if that will carry over to VM.

 

[videogames101]

Originally posted by: AlexWade
Didn't hyperthreading have a huge negative effect on server applications? I have to wonder if that will carry over to VM.

no, that was just netburst epic failing....

 

[exar333]

It sounds like hyperthreading on the i7 woukld suit your needs nicely. You can already see this in some of the early benchmarks, but it should become more apparent as time goes on how great this processer will be when executing 4+ tasks simultanesouly versus a standard quad (Intel or AMD).

 

[dguy6789]

Originally posted by: ExarKun333
It sounds like hyperthreading on the i7 woukld suit your needs nicely. You can already see this in some of the early benchmarks, but it should become more apparent as time goes on how great this processer will be when executing 4+ tasks simultanesouly versus a standard quad (Intel or AMD).

Agreed. Pretty much, the more stuff you do at the same time, the more benefit you will see from Nehalem.

 

[Denithor]

Originally posted by: videogames101

Originally posted by: AlexWade
Didn't hyperthreading have a huge negative effect on server applications? I have to wonder if that will carry over to VM.

no, that was just netburst epic failing....

+1

That's exactly the point AT made in their review, hyperthreading on P4 was hobbled by the low throughput of the processor (single core, remember?). CPUs have gotten much "smarter" since then and can juggle the tasks more effectively among different cores. And the 8MB L3 cache doesn't hurt, either!

 

[soccerballtux]

Originally posted by: aigomorla

Originally posted by: dwhore

Will it benefit my VMs to have '8' logical processors?

i7 with hyperthread will almost = 2 yorkfields @ same clock.

Almost. its like 1.75x

So yes, you will benifit from 8 tasks.

Lol, no. Citation needed please.

 

[Denithor]

Originally posted by: soccerballtux

Originally posted by: aigomorla
i7 with hyperthread will almost = 2 yorkfields @ same clock.

Almost. its like 1.75x

So yes, you will benifit from 8 tasks.

Lol, no. Citation needed please.

Is 1.6x close enough?

Take a look at the first benchmark there...i7 965 vs QX9770. Both chips at 3.2GHz, i7 blows past the QX without even noticing the speedbump in the road. Now, granted, most of the benchmarks don't look like that. But i7 manages a nearly 1.5x performance in the x264 encoding also. Anywhere the app is seriously optimized for multi-threaded computing the i7 just crushes the older generation.

Now, I wonder how those apps would run on GPGPU?
EDIT: Oh, wait...we already know!

 

[myocardia]

Originally posted by: Cogman
Yes, it will. Hyperthreading provides a speed increase in Multithreaded applications at the cost of single threading speed (at least, the old Hyperthreading did).

Yeah, it still works that way with the Nehalems, although with HT 2.0 (not Intel's official name, BTW), single threaded speed loses less performance than it did with the P4's. I'll see if I can find the review I was reading that showed this.

Originally posted by: soccerballtux
Lol, no. Citation needed please.

I like this reply. Mind if I borrow it sometime, as long as I don't happen to be disagreeing with you?


edit: BTW, I'm not really wanting to borrow that. It was definitely original, though.

 

[waffleironhead]

Originally posted by: Denithor

Originally posted by: soccerballtux

Originally posted by: aigomorla
i7 with hyperthread will almost = 2 yorkfields @ same clock.

Almost. its like 1.75x

So yes, you will benifit from 8 tasks.

Lol, no. Citation needed please.

Is 1.6x close enough?

Take a look at the first benchmark there...i7 965 vs QX9770. Both chips at 3.2GHz, i7 blows past the QX without even noticing the speedbump in the road. Now, granted, most of the benchmarks don't look like that. But i7 manages a nearly 1.5x performance in the x264 encoding also. Anywhere the app is seriously optimized for multi-threaded computing the i7 just crushes the older generation.

Now, I wonder how those apps would run on GPGPU?
EDIT: Oh, wait...we already know!

The problem here is you are pointing out the exception and using it as the rule.
the 1.6X is the high end of performance gains not the norm. (as far as I have seen).

 

[dguy6789]

Originally posted by: waffleironhead

Originally posted by: Denithor

Originally posted by: soccerballtux

Originally posted by: aigomorla
i7 with hyperthread will almost = 2 yorkfields @ same clock.

Almost. its like 1.75x

So yes, you will benifit from 8 tasks.

Lol, no. Citation needed please.

Is 1.6x close enough?

Take a look at the first benchmark there...i7 965 vs QX9770. Both chips at 3.2GHz, i7 blows past the QX without even noticing the speedbump in the road. Now, granted, most of the benchmarks don't look like that. But i7 manages a nearly 1.5x performance in the x264 encoding also. Anywhere the app is seriously optimized for multi-threaded computing the i7 just crushes the older generation.

Now, I wonder how those apps would run on GPGPU?
EDIT: Oh, wait...we already know!

The problem here is you are pointing out the exception and using it as the rule.
the 1.6X is the high end of performance gains not the norm. (as far as I have seen).

QFT. Majority of the time, you will never get anywhere near that kind of performance gain.

 

[palladium]

Wonder if there is a way for the OS to know which core is the 'physical' one and which is the 'virtual' one...

 

[myocardia]

Originally posted by: dguy6789
QFT. Majority of the time, you will never get anywhere near that kind of performance gain.

Actually, it would depend on what type of software you primarily run on your system. If you're running alot of scientific calculations (including folding proteins), 3D rendering (CAD), or serious video rendering, you'll get huge gains that will make it worth every penny. For the majority of us gamers and general usage users around here, though, it definitely won't be.

 

[Denithor]

Originally posted by: myocardia
Actually, it would depend on what type of software you primarily run on your system. If you're running alot of scientific calculations (including folding proteins), 3D rendering (CAD), or serious video rendering, you'll get huge gains that will make it worth every penny. For the majority of us gamers and general usage users around here, though, it definitely won't be.

Wonder how ppd will stack up against C2Q running SMP client? If significantly higher will it begin to approach the points seen when folding with GPGPU? Obviously the cost will still be out of line as you can get ~4500 ppd from a $50 8800GS/9600GSO (and up to 8-9k ppd from an overclocked GTX 280).

 

[magreen]

Come on, don't pile-on aigo so much. He probably meant 1.75x in the types of apps the OP is talking about, not across the board.

I still can't tell if anyone here really considered the OP's question fully. He only has up to 3 intensive threads going at a time. Maybe Hyperthreading wouldn't improve his performance since he still always has at least one idle physical core (well, idle except for system/internet tasks)

 

[Idontcare]

Originally posted by: magreen
Come on, don't pile-on aigo so much. He probably meant 1.75x in the types of apps the OP is talking about, not across the board.

I still can't tell if anyone here really considered the OP's question fully. He only has up to 3 intensive threads going at a time. Maybe Hyperthreading wouldn't improve his performance since he still always has at least one idle physical core (well, idle except for system/internet tasks)

If he's strictly running single-threaded apps within his VM's then the additional thread processing capability of i7 over Yorkfield won't help any more than going to skulltrail would help.

If he's running multi-threaded apps within the VM then he stands reasonable chance of his three active VM sessions utilizating more than 4 logical cores and he would benefit from skulltrail/i7 type system.

 

[aigomorla]

i was assuming the op had all 8 threads in use on VM.

 

[aka1nas]

Originally posted by: palladium
Wonder if there is a way for the OS to know which core is the 'physical' one and which is the 'virtual' one...

IIRC, all modern versions of Windows are aware of logical and physical cores now.

 

[Comdrpopnfresh]

Yes. Hyperthreading makes sure the processor's pipelines are logically partitioned so that if a set of instructions doesn't saturate the pipeline, portions of another instruction set and be queued along with the first, so the pipeline is used more efficiently clock for clock

 

[Comdrpopnfresh]

Originally posted by: Idontcare

Originally posted by: magreen
Come on, don't pile-on aigo so much. He probably meant 1.75x in the types of apps the OP is talking about, not across the board.

I still can't tell if anyone here really considered the OP's question fully. He only has up to 3 intensive threads going at a time. Maybe Hyperthreading wouldn't improve his performance since he still always has at least one idle physical core (well, idle except for system/internet tasks)

If he's strictly running single-threaded apps within his VM's then the additional thread processing capability of i7 over Yorkfield won't help any more than going to skulltrail would help.

If he's running multi-threaded apps within the VM then he stands reasonable chance of his three active VM sessions utilizating more than 4 logical cores and he would benefit from skulltrail/i7 type system.

wouldn't someone see a performance gain w/ HT with a multi-threaded program, or multiple simultaneous single threaded apps? Like multi-core optimized programs + user multitasking gains from more than 1 core?

 

[Idontcare]

Originally posted by: Comdrpopnfresh

Originally posted by: Idontcare

Originally posted by: magreen
Come on, don't pile-on aigo so much. He probably meant 1.75x in the types of apps the OP is talking about, not across the board.

I still can't tell if anyone here really considered the OP's question fully. He only has up to 3 intensive threads going at a time. Maybe Hyperthreading wouldn't improve his performance since he still always has at least one idle physical core (well, idle except for system/internet tasks)

If he's strictly running single-threaded apps within his VM's then the additional thread processing capability of i7 over Yorkfield won't help any more than going to skulltrail would help.

If he's running multi-threaded apps within the VM then he stands reasonable chance of his three active VM sessions utilizating more than 4 logical cores and he would benefit from skulltrail/i7 type system.

wouldn't someone see a performance gain w/ HT with a multi-threaded program, or multiple simultaneous single threaded apps? Like multi-core optimized programs + user multitasking gains from more than 1 core?

Of course what you say is true, but my comments in this thread are being intentionally restricted to the OP's stated boundary condition that at most he has 3 VM sessions which would be actively doing something with the processor.

3 being less than 4, and 4 being the number of cores on a yorkfield, makes or breaks the argument for or against an i7 in this case.

 

[Zstream]

Hyperthreading in VM's provides mixed results. Most of the results are generally opionion and based on the workload.

Going off of basic test any software that sets affinity you will notice weird, sometimes buggy results. If you are using quite a bit of the CPU on any of the VM's it is best to turn off HT. If you are just launching 6-8 VM's with very low cpu activity then HT is a plus.

 

[sunnn]

Originally posted by: Denithor

Originally posted by: soccerballtux

Originally posted by: aigomorla
i7 with hyperthread will almost = 2 yorkfields @ same clock.

Almost. its like 1.75x

So yes, you will benifit from 8 tasks.

Lol, no. Citation needed please.

Is 1.6x close enough?

Take a look at the first benchmark there...i7 965 vs QX9770. Both chips at 3.2GHz, i7 blows past the QX without even noticing the speedbump in the road. Now, granted, most of the benchmarks don't look like that. But i7 manages a nearly 1.5x performance in the x264 encoding also. Anywhere the app is seriously optimized for multi-threaded computing the i7 just crushes the older generation.

Now, I wonder how those apps would run on GPGPU?
EDIT: Oh, wait...we already know!

are these figures with turbo mode on or off?
also, the way things re shaping up, cpu becomes less and less important, proven by the emergence of atom-based products.
games are gpu bound and hpc, video encoding/editing, 3d rendering leaning towards gpgpu.
soon we will have a chipset with integrated cpu or gpu with integrated cpu.
the future of nvidia hinges on CUDA, amd on fusion and intel on larabee. a new battleground looms.
or maybe that future is already here? does that mean nehalem is late to the party?
amd is in a very unique position. nvidia is moving heaven and earth to create market for CUDA while intel's larabee isnt even out.