Eight cores with an i7 processor ?!?!?

T

ToddAndMargo

Hi All,

I just installed XP-Pro-SP3 on an Intel Quad core i7
based motherboard.

XP's Task Manager show it has *eight* cores. (I presume
there is something like the old Pentium Hyperthread going
on here.)

Question: is eight cores a little much? Would it be better
to tell the BIOS to shut off the Hyperthread (or whatever)
and just go four raw cores? Any one have any words of
wisdom to share here?

Many thanks,
-T
 
J

JS

4, 8, 16, 32 or 64 cores.
It's the software that make it all work.
Poor software = poor results.

Besides all computers are limited by using
the "Von Neumann architecture" and the FSB.
 
T

Tim Meddick

I think it would probably be a good idea to do some tests of your own, as to
whether or not your PC will perform better.

Try turning off HT in your BIOS (under CPU Management or something similar)
for a while and take notice of how your PC performs.

If you prefer how it performs with HT switched off - then I'd leave it that
way.


==

Cheers, Tim Meddick, Peckham, London. :)
 
I

Ian D

Tim Meddick said:
I think it would probably be a good idea to do some tests of your own, as
to whether or not your PC will perform better.

Try turning off HT in your BIOS (under CPU Management or something
similar) for a while and take notice of how your PC performs.

If you prefer how it performs with HT switched off - then I'd leave it
that way.


==

Cheers, Tim Meddick, Peckham, London. :)

In order to use more than one core, an application has to be
multiprocessor capable, and multithreaded, however most
apps are multithreaded. Just check the processes tab in
Task Manager. MS Flight Simulator FSX is one of the most
CPU bound apps around, and with its SP1 and SP2 updates
it's optimized to utilize upto 16 or more cores. I did a lot of
testing with my i7 920 and FSX, and found that it does run
smoother with HT enabled. The difference was more noticeable
with 64 bit Vista Ultimate than with XP Pro.

With SiSoft Sandra 2009, I ran the multicore efficiency benchmark
with HT off, and the i7 performance was in the basement. I
remembered HT was off, so I enabled it, and re-ran the test.
This time the results were excellent. The i7 920 was close behind
the i7 940 and i7 Extreme, way ahead of the Core 2 quads. The
result is that I now leave HT on.

One thing of interest is that with Task Manager in XP, the
HT 0, 2, 4, and 6 units are displayed first, then units 1, 3,
5, and 7. In Vista's Task Manager, the HT units are
displayed in the proper order as 0, 1, 2, 3, 4, 5, 6, and 7,
so each core's HT pairs are together.
 
P

Paul

ToddAndMargo said:
Hi Tim,

At some point too many cores start to slow things down.
I thought it was 7 where things started to bog down.

I was wondering if it was better to run four real cores
or eight fake cores.

-T

The Core i7 doesn't have a conventional bottleneck.

If you look at my motherboard, it works like this. It is one
of the slowest ones you can buy.

Core2_Duo_LGA775
|
| FSB800 * 8 bytes = 6.4GB/sec
|
Northbridge -- 2 channels of DDR2

That is the "old" architecture. The FSB is the bottleneck.

In the case of Core2 Quad processors, they use two silicon
die, and cache coherency traffic would have to travel over
the FSB. A Quad for LGA775 looks like this. (In a
"perfect scaling" test, the Core2 Quad goes 3.5x as fast,
when using 4 cores, so does not scale perfectly. That is
about 88% performance.)

Core1,2 Core3,4 Core2 Quad - two silicon die
| |
+----+----+ LGA775
|
| FSB
|
v
to Northbridge

Now, compare that to Core i7. It uses one silicon die.
It adds L3 cache to the design (one more level of caching
than on Core2).

Core_i7_LGA1366 -- 3 channels of DDR3, up to six sticks
|
| QPI
|
Northbridge X58

What travels over QPI, is peripheral traffic. On an SLI
motherboard, that would be gigabytes per second of graphics
traffic. But since the memory is connected directly to the
processor, the memory path is no longer a bottleneck for
the processor. So there is some improvement, similar to
what AMD did years ago.

I haven't seen a scaling test result for Core i7, but
I presume it does a little better than Core2 designs.

Hyperthreading is used, when a core is blocked on some
operation, like towards memory. The second (virtual) core,
continues its thread of execution in the interim. On
older processors, there were cases where this caused
more problems than it solved.

Part of the fun of owning new hardware, is testing
this stuff :) Switch it on, switch it off, run
some benchmarks.

Good luck,
Paul
 
T

ToddAndMargo

Ian said:
In order to use more than one core, an application has to be
multiprocessor capable, and multithreaded, however most
apps are multithreaded. Just check the processes tab in
Task Manager. MS Flight Simulator FSX is one of the most
CPU bound apps around, and with its SP1 and SP2 updates
it's optimized to utilize upto 16 or more cores. I did a lot of
testing with my i7 920 and FSX, and found that it does run
smoother with HT enabled. The difference was more noticeable
with 64 bit Vista Ultimate than with XP Pro.

With SiSoft Sandra 2009, I ran the multicore efficiency benchmark
with HT off, and the i7 performance was in the basement. I
remembered HT was off, so I enabled it, and re-ran the test.
This time the results were excellent. The i7 920 was close behind
the i7 940 and i7 Extreme, way ahead of the Core 2 quads. The
result is that I now leave HT on.

One thing of interest is that with Task Manager in XP, the
HT 0, 2, 4, and 6 units are displayed first, then units 1, 3,
5, and 7. In Vista's Task Manager, the HT units are
displayed in the proper order as 0, 1, 2, 3, 4, 5, 6, and 7,
so each core's HT pairs are together.

Cool! Great feedback. Thank you.

-T
 
T

ToddAndMargo

Paul said:
The Core i7 doesn't have a conventional bottleneck.

If you look at my motherboard, it works like this. It is one
of the slowest ones you can buy.

Core2_Duo_LGA775
|
| FSB800 * 8 bytes = 6.4GB/sec
|
Northbridge -- 2 channels of DDR2

That is the "old" architecture. The FSB is the bottleneck.

In the case of Core2 Quad processors, they use two silicon
die, and cache coherency traffic would have to travel over
the FSB. A Quad for LGA775 looks like this. (In a
"perfect scaling" test, the Core2 Quad goes 3.5x as fast,
when using 4 cores, so does not scale perfectly. That is
about 88% performance.)

Core1,2 Core3,4 Core2 Quad - two silicon die
| |
+----+----+ LGA775
|
| FSB
|
v
to Northbridge

Now, compare that to Core i7. It uses one silicon die.
It adds L3 cache to the design (one more level of caching
than on Core2).

Core_i7_LGA1366 -- 3 channels of DDR3, up to six sticks
|
| QPI
|
Northbridge X58

What travels over QPI, is peripheral traffic. On an SLI
motherboard, that would be gigabytes per second of graphics
traffic. But since the memory is connected directly to the
processor, the memory path is no longer a bottleneck for
the processor. So there is some improvement, similar to
what AMD did years ago.

I haven't seen a scaling test result for Core i7, but
I presume it does a little better than Core2 designs.

Hyperthreading is used, when a core is blocked on some
operation, like towards memory. The second (virtual) core,
continues its thread of execution in the interim. On
older processors, there were cases where this caused
more problems than it solved.

Part of the fun of owning new hardware, is testing
this stuff :) Switch it on, switch it off, run
some benchmarks.

Good luck,
Paul

Thank you!
-T
 

Ask a Question

Want to reply to this thread or ask your own question?

You'll need to choose a username for the site, which only take a couple of moments. After that, you can post your question and our members will help you out.

Ask a Question

Top