I picked up a IBM eServer x235 with Dual Xeon 2.66 processors and 1 GB RAM for crunching. I am pretty disappointed in the results so far. Here is a link to the results. Link. Anyone else out there running a similiar rig getting the same results? I have HT enabled so it is crunching 4 WUs at a time. There is nothing else running on the server except for BOINC. Any tips for getting some better results?

TIA,

Eric

---

link does not seem to work!

---

i think he means this and these results

link does not seem to work!

Sorry about the bad link. The above post is correct.

---

Is that 1GB Ram in total or per core?

---

Is that 1GB Ram in total or per core?

That is 1GB of RAM. Are you refering to the cache memory? If so. Each CPU has 512 KB L2 cache.

---

---

The benchmarks are very low:

Measured floating point speed 1139.23 million ops/sec
Measured integer speed 1059.3 million ops/sec

Is it running at 2.66GHz (i.e. any speedstep or something similar running?)

---

The benchmarks are very low:

Measured floating point speed 1139.23 million ops/sec
Measured integer speed 1059.3 million ops/sec

Is it running at 2.66GHz (i.e. any speedstep or something similar running?)

Nothing like speedstep running (unless there is something in the IMB BIOS that I have overlooked) and this is a straight Win 2K3 R2 with nothing else besides BOINC installed. I will take a look at the BIOS again just to make sure. Are you thinking of a dual core Xeon? These Xeons are single core with HT. I ran CPUid and they show the correct clock. Thanks for the input.

---

yeah - HT is probably responsible for the low benchmarks. This is a celeron 2.66GHz:

Measured floating point speed 1340.57 million ops/sec
Measured integer speed 2277.39 million ops/sec

which should get the same as the xeon as (i believe) they're both based on the same netburst core and cache doesn't affect the benchmarks...

Seeing as it's running 4 tasks at once, I think it's doing fine. If you want more then you could always oc it ;)

---

Since you are low on memory (1GB) for that setup, I would restrict BOINC to running only two processes. Luckily the benchmark means little here at Rosetta.

the reason is that most of the time one or the other of the rosetta will be sat in idle due to memory restrictions.
512MB per instance is recommended.
Though a 2.66GHz Xeon Prestonia CPU is not as fast as you might think since they are still basically a Pentium 4. Arrived on the market in 2002 (the 2666MHz Xeon) I think, so 5 years old now.

But you just cannot run 4 Rosetta's in 1GB efficiently.

---

Team mauisun.org

DCDC, you speak about cache not affecting the benchmarks, but infer that it does affect the way rosetta runs. I have heard this elsewhere too. I have a AMD Athlon 64 3000+ with 512kb cache, and a 3.00GHz P4 with 1MB cache, and it runs slightly better (more credit per WU) on the AMD. What effect does cache have on performance with rosetta?

---

DCDC, you speak about cache not affecting the benchmarks, but infer that it does affect the way rosetta runs. I have heard this elsewhere too. I have a AMD Athlon 64 3000+ with 512kb cache, and a 3.00GHz P4 with 1MB cache, and it runs slightly better (more credit per WU) on the AMD. What effect does cache have on performance with rosetta?

Cache doesn't affect the benchmarks as the benchmark code is tiny and so fits into any size L2 cache. The Rosetta application doesn't though - there can be tens of megabytes of active code being called so the more that can fit into the cache the better. Although your P4 has a bigger cache, the Athlon is a more efficient architecture (the P4's netburst architecture was designed to be less efficient on this type of code, but hit very high clock speeds and be optimised for multimedia), especially when it comes to FPU intensive code which Rosetta is, so it makes up for the lack of cache.

The general rule is run it on whatever you can though ;)

---