TR Forums

Here:

http://www.fudzilla.com/content/view/18474/1/

Topic moved to OC/Cooling forum.

Asrock board ?? HUH ?

I've only owned one board from asrock but it did turn out to be as solid as a rock(hehe).Currently it's around 4 years old and still chugging along smoothly in a hand me down system i gave to a friend.Paired with a Athlon 64 X2 4600+ and a 4850 it still works nicely as a decent all around system.

Gah. Asrock.

I had DOA boards like peanuts, and those that weren't DOA were soon dead with no explanation. Few survived.

MrBojangles wrote:

I've only owned one board from asrock but it did turn out to be as solid as a rock(hehe).Currently it's around 4 years old and still chugging along smoothly in a hand me down system i gave to a friend.Paired with a Athlon 64 X2 4600+ and a 4850 it still works nicely as a decent all around system.

I also have to confess, i have owned a Asrock board which had P4 1.7 GHZ CPU on it, it's still working nice as server for a school.

The 1GB stick of memory was holding it back! Try a 512MB stick and push that baby!

Asrock has always been a hit or miss, though they were always the company doing interesting things. I believe they had a socket939/am2 board out at one point....

So many comments on the Asrock brand. Is no one surprised at the 6.29 GHz number? I think this is a good sign for AMD, and that we'll be seeing higher clocks from AMD soon.

I still want HyperThreading on AMD chips. How about a CPU that can handle 24 threads? Anyone?

JdL wrote:

So many comments on the Asrock brand. Is no one surprised at the 6.29 GHz number? I think this is a good sign for AMD, and that we'll be seeing higher clocks from AMD soon.

I've never used Asrock. The 6.29 GHz is indeed impressive, but keep in mind that it was done with liquid nitrogen cooling!

I still want HyperThreading on AMD chips. How about a CPU that can handle 24 threads? Anyone?

To me, Hyperthreading has always seemed like a workaround for other shortcomings of the core design, i.e. inability to keep all the pipelines busy. If you design the core such that it can keep its piplelines properly fed, Hyperthreading doesn't gain you much.

just brew it! wrote:

JdL wrote:

So many comments on the Asrock brand. Is no one surprised at the 6.29 GHz number? I think this is a good sign for AMD, and that we'll be seeing higher clocks from AMD soon.

I've never used Asrock. The 6.29 GHz is indeed impressive, but keep in mind that it was done with liquid nitrogen cooling!

I still want HyperThreading on AMD chips. How about a CPU that can handle 24 threads? Anyone?

To me, Hyperthreading has always seemed like a workaround for other shortcomings of the core design, i.e. inability to keep all the pipelines busy. If you design the core such that it can keep its piplelines properly fed, Hyperthreading doesn't gain you much.

The alternative is to use single-issue cores. Then all pipelines will be busy all the time.

Shining Arcanine wrote:

just brew it! wrote:

To me, Hyperthreading has always seemed like a workaround for other shortcomings of the core design, i.e. inability to keep all the pipelines busy. If you design the core such that it can keep its piplelines properly fed, Hyperthreading doesn't gain you much.

The alternative is to use single-issue cores. Then all pipelines will be busy all the time.

That's kind of like throwing out the baby with the bathwater. Yes, you make more efficient use of your execution resources; but the core as a whole does less work per clock cycle. Single-issue also does not help mitigate the effect of pipeline stalls caused by cache misses or branch mis-predictions.

P4 (especially Prescott) really needed Hyperthreading because of its ultra-deep pipeline. K8 would have probably benefited much less from Hyperthreading than the P4 did, since its pipeline was much shallower (therefore pipeline stalls/flushes were not as costly in terms of tying up resources or causing work already done to be thrown away).

just brew it! wrote:

I still want HyperThreading on AMD chips. How about a CPU that can handle 24 threads? Anyone?

To me, Hyperthreading has always seemed like a workaround for other shortcomings of the core design, i.e. inability to keep all the pipelines busy. If you design the core such that it can keep its piplelines properly fed, Hyperthreading doesn't gain you much.

I agree. Except I think that HT is a key component in keeping pipelines busier.

My understanding is that the OS / application sending instructions to the CPU understands how many logical CPU's there are, and attempts to schedule / distribute itself based on that known number. So if you have a single physical core / pipeline with HyperThreading enabled, the OS will see 2 CPU's, and it will "start" 2 operations. The CPU can then subschedule / subdistribute those operations within the pipeline, thus making the process as a whole more efficient.

Back during the days of P4, this was not always a performance-enhancing feature because very little software was built to multi-cored platforms. Today is a different story. I think this would provide some benefit even on the most efficent cores, especially in today's increasingly multi-threaded OS / application environments.

JdL wrote:

I agree. Except I think that HT is a key component in keeping pipelines busier.

My understanding is that the OS / application sending instructions to the CPU understands how many logical CPU's there are, and attempts to schedule / distribute itself based on that known number. So if you have a single physical core / pipeline with HyperThreading enabled, the OS will see 2 CPU's, and it will "start" 2 operations. The CPU can then subschedule / subdistribute those operations within the pipeline, thus making the process as a whole more efficient.

Back during the days of P4, this was not always a performance-enhancing feature because very little software was built to multi-cored platforms. Today is a different story. I think this would provide some benefit even on the most efficent cores, especially in today's increasingly multi-threaded OS / application environments.

I have always though that HT would effect multi-core performance when an application could take use of 2 cores effectively. As in, lets say we are looking at an i7. Logical Core 0 and Core 1 = Physical Core 0. If something can only take use of dual-cores, and chooses to use Core 0 and Core 1 when applying the two threads is this or is this not effecting performance? Would it not make more sense to have it run on two physical cores rather than two logical? Possibly I'm missing something....