Oops - meant to write Phenom 955

Hi,
Is anybody able to compare the core i7 arch versus the !Phenom 955! arch? I'll be running fc11 or centos5 and the machine will be a workhorse for Mathematical modeling. I'll be running 4 inistances of a model and expect to have all four cores running at 100% for long peroids of time.

I have a core i7 machine now running RHEL5 (recognizing 8 logical processors) and am noticing what i think might be some overhead due to having hyperthreading enabeled when i have 4 threads running at 100%.

When I would expect the cpu to be @100%, it often will drop to 0% usage for a moment then back up to 100%.

I've run the same models on a dual athalon MP 2000+ architecture, and had straight 100% usage on both processors. The i7 is my first multi-core system, so im not too familiar with them.

Thanks.

I can't directly give you the answer that you seem to be asking for, but your question is a bit like (not quite as extreme as this...I'm trying to emphasise a point here)

Tell me, the differences between a Model T Ford and a spce shuttle.

I can't see any circumstance in which an exact (as opposed to slightly general, rather handwaving-ish) answer to that is useful and I'm not sure about your question either. Additionally, it doesn't seem relevant to the comparison to a to compare a .

There are various flavours of sempron, all of which are slower versions of some contempory or obsolescent AMD architecure, available in a higher performance version. I don't know of -I don't need to know- about the lower performance parts, but a sempron 955? There is a Phenom 955, which would have quite a different architecture from any yet-released Sempron, and be yet another variable.

What I can tell you is that some apps are faster with hyperthreading off. It is not the majority of apps though and I don't know why this happens, but, in corner cases, it can be pronounced. People who find this for an app that they run predominantly consider turning hyperthreading off. It may even depend on the kernel, although I haven't seen any testing on that.

Hyperthreading isn't really comparable with anything on the AMD side of the wall. In part, this is because at a deeper level of the architecture, it isn't (actually, hasn't been) as useful in the context of 'the AMD arch' as it was for 'the Intel arch'. (As you may have guessed neither the phrase 'the intel Arch' nor 'the AMD arch' are strictly meaningful phrases, without further definition, but they will have to do, for now.)

How are you measuring this? What tools are you using and what exactly are the measurements?

Arguably, a better benchmark is usually how much work you get done per unit time, rather than how long the cpus spend idle.

I think you'd be better off building your own system from scratch or use Gentoo. Might even want to build a custom kernel just for the system you will use.

Then you'd have to re-compile the application for this system.

Anything less would be useless.

I would disagree on the Gentoo and building from scratch. I have known a lot of people who have bickered about building from source, and weather or not it is worth it... Most of the time the performance you are going to get by building from scratch is minimal, or so I have seen.

Yeah you're absolutely right- I didnt sleep last night and I've been working for about 12 hours. I meant Phenom 955.

<Qoute>
I can't directly give you the answer that you seem to be asking for, but your question is a bit like (not quite as extreme as this...I'm trying to emphasise a point here)

Tell me, the differences between a Model T Ford and a spce shuttle.,
</Quote>

Yeah thanks for the subtlety! heh heh.. Not really. More like choosing to fly on the test flight of a 1st generation Jet or a routine High Speed Prop plane flight- or whatever... I'm comparing 2 chips/chipsets.

I guess I need to be a bit more specific.

Im Comparing two similarly priced quad core chips (systems) running at similar clock speed ~3GHz.

The AMD phenom 955 is 3.2GHz or so quad core chip.
The Intel core i7 940 is a 64 bit quad core chip that runs around 2.9GHz With no Front side bus, and an integrated memory controller.- a bit more expensive

<Qoute>
Additionally, it doesn't seem relevant to the comparison to a
Quote:
dual thalon MP 2000+
</Quote>

what im saying here is that I've run the models (divide and conquer which easily scale from 2 to 8 processors) that i plan on buying the system for on an actual Dual processor SMP system, with no hiccups in the CPU usages. Both with red hat linux -though the old (athlon mp) system was running an older version of redhat with an older kernel so this could be an issue.

I'm now running the same models on a i7 system, and had seen CPUs running for some time at 100% then dropping to 0% then back to 100%. You're right its not always that the processors are maxed out that matters, but I'm not sure whats happening here, and if its a memory thing...

I'm buying a new system specifically to run these models on, and i'm wondering if the performance has anything to do with the i7's new architecture- like memory access via the i7's new memory controller, or if this is something that i can expect to see in any Quad core chip. From what I understand, There are some other issues with quad core processors that keep them from getting a linear speedup/core, ex. I believe that the quad cores have the same cache as Dual cores. so two cores are sharing a cache. this has got to be a bottle neck I don't know if this is still true. I'm not an expert.

So I guess what it comes down how would people rate a core i7 without hyperthreading against a phenom 955. Thread for thread on a memory intensive application? Would you expect to see similar preformance?

do you think that its a kernel issue that i'm seeing?