If anyone is interested, I am benchmarking HT/CMT on the new Nehalem CPU's. I have passed what I typically call burn-in for 4 new servers. Usually I stop accepting new work, let my cache run it's course, and start putting these machine to work on their intended function. I am now interested in comparing these machines to others running R@H. Additionally I am comparing them against each other.

My Computers

The servers be1-4 are my test cases. They have 2 different distro's on them. Additionally be1 and b3 have HT/CMT enabled and be2/be4 do not have HT/CMT enabled. I plan to run these for another couple weeks if I can hold off on putting to my real work.

If you have anything to share about the new HT please do so here. Otherwise thanks for the read!

---

Hi Rob

We can't see the computer names but I can see from the number of CPUs that:

1078449 = HT
1079102 = HT
1078450 = No HT
1079103 = No HT

(although it's also evident from the RACs!)

If anyone is interested, I am benchmarking HT/CMT on the new Nehalem CPU's. I have passed what I typically call burn-in for 4 new servers. Usually I stop accepting new work, let my cache run it's course, and start putting these machine to work on their intended function. I am now interested in comparing these machines to others running R@H. Additionally I am comparing them against each other.

My Computers

The servers be1-4 are my test cases. They have 2 different distro's on them. Additionally be1 and b3 have HT/CMT enabled and be2/be4 do not have HT/CMT enabled. I plan to run these for another couple weeks if I can hold off on putting to my real work.

If you have anything to share about the new HT please do so here. Otherwise thanks for the read!

---

Thanks, and you are correct.

1078449 = HT
1078450 = No HT

1079102 = HT
1079103 = No HT

The second set of machines were initiated on BOINC/R@H about a day after the first pair due to a wonderful driver limitation in nearly every Debian & it's variants installers. I finally found something that worked the next day. Price you pay for using "new" hardware with "free" software =)

---

That's an impressive RAC!

---

thx

---

That's a impressive RCA along with a great contribution to Rosetta. What is the main purpose of your servers, is your power bill high with all those PCs?

---

Have a crunching good day!!

They are not PC's in my home, but rather machines in a data center. I typically use R@H for burning in new machines, giving retired machines something to do, or using up some cycles on machines with a light work load.

Thanks For info. Maybe you could add a I7 server computer to the fold, if would make your RCA even higher and it would speed things up in the data center. This is just a idea.

---

Have a crunching good day!!

Thanks For info. Maybe you could add a I7 server computer to the fold, if would make your RCA even higher and it would speed things up in the data center. This is just a idea.

? He's running four!

---

Ok the fastest cpu types i see is 2 Xeon, I don't see any I7's in his list. If I have miss understood anything I stand corrected.

---

Have a crunching good day!!

E5520 = nehalem / i7 ;)

---

I stand corrected I apologize for my misunderstanding. Thank you dcdc for your correction.

---

Have a crunching good day!!

If you have anything to share about the new HT please do so here. Otherwise thanks for the read!

HT is obviously improved from the Pentium 4 days. But the performance is still workload-dependent. For example, on the SETI and ClimatePrediction forums, we've found that having HT off actually increases RAC in most cases. SETI is about break-even, but CPDN performance is definitely hurt by running 8 of the same tasks per physical CPU.

I personally get the best workstation-crunching trade-off by having HT on but limiting BOINC to 4 cores. My applications and such don't scale beyond 4 threads anyway, so everything works smoothly this way. Playing games makes HT a waste, but I'm too lazy to change BIOS every time I want to play a game.

If you have anything to share about the new HT please do so here. Otherwise thanks for the read!

HT is obviously improved from the Pentium 4 days. But the performance is still workload-dependent. For example, on the SETI and ClimatePrediction forums, we've found that having HT off actually increases RAC in most cases. SETI is about break-even, but CPDN performance is definitely hurt by running 8 of the same tasks per physical CPU.

I personally get the best workstation-crunching trade-off by having HT on but limiting BOINC to 4 cores. My applications and such don't scale beyond 4 threads anyway, so everything works smoothly this way. Playing games makes HT a waste, but I'm too lazy to change BIOS every time I want to play a game.

WOW...I have an HT from the old P4 days and it SCREAMS, it is now running under Linux and puts some AMD quad core 9850 machines to shame most days in the RAC. Same project, obviously different hardware and the P4 is the only one running Linux, but it flat screams!!!
http://abcathome.com/show_host_detail.php?hostid=65041
If the new ones are much better I can't wait for them to come down in price to my price neighborhood!!

---

If you have anything to share about the new HT please do so here. Otherwise thanks for the read!

HT is obviously improved from the Pentium 4 days. But the performance is still workload-dependent. For example, on the SETI and ClimatePrediction forums, we've found that having HT off actually increases RAC in most cases. SETI is about break-even, but CPDN performance is definitely hurt by running 8 of the same tasks per physical CPU.

I personally get the best workstation-crunching trade-off by having HT on but limiting BOINC to 4 cores. My applications and such don't scale beyond 4 threads anyway, so everything works smoothly this way. Playing games makes HT a waste, but I'm too lazy to change BIOS every time I want to play a game.

Agreed. In our case HT was causing our P4's to crack under our production work load (non-Boinc). "No world" issues and stability forced us to disable HT. It appears that R@H is integer based and that HT on the Nehalems is providing additional Integer computational headroom.

These machines will become high-performance mail servers. When I benchmark them with SPECmail09 I'll try to post basic results here for those interested. Since mail is usually I/O bound and these won't ne doing AV/AS I doubt HT will make much difference, but time will tell.

Agreed. In our case HT was causing our P4's to crack under our production work load (non-Boinc). "No world" issues and stability forced us to disable HT. It appears that R@H is integer based and that HT on the Nehalems is providing additional Integer computational headroom.

Rosetta is integer heavy? Gosh, maybe I could get more RAC by using more processors. My computer's workload is divided over Climate, SETI, Rosetta, and Einstein. I thought all of them were floating-point heavy.