Some of You may be interested, how much boost - if any - we achieve when hyperthreading is ON. Is it 100%, 50%, 25%? Here are my results.

[b]		Avg WU Time [s]		Avg WU per core / day	Avg Claimed Credit	Avg Granted Credit	Avg Credit per day	HT Diff[/b]
[b]i7 4770K HT OFF[/b]	10600,0006666667	8,1509428836		123,812			144,2326666667		4702,5289117906		0,00%
[b]i7 4770K HT ON[/b]	10311,2933333333	8,3791622648		120,4386111111		89,5136111111		6000,3925792691		[b][color=green]+27,60%[/color][/b]

As You can see, there is ~28% improvement when using HT. To be honest, I was hoping for 50-75% performance increase, so I am little dissapointed. I wonder is it HT limitation or Rosetta non-HT optimized client?

---

There can be significant variation between WUs. How many days did you run each trial? In fact, if one configuration is able to crunch another model in the same exact WU (if you had copied identical data for the two trials), then it would try. This extra model would get you extra granted credit, but perhaps alter the seconds of runtime per WU.

HT is not a magic CPU that you get the use of, but do not pay for. It's just another method of dividing the resources you have. If you were doing lots of disk IO or network traffic, where keeping the pipe full is important for throughput, then you might see more significant improvement. In the case of R@h, the primary bottleneck is floating point calculations per second. As you can see by your claimed credit figure, the BOINC benchmark of your machine (which is used for computing the credit claim) is roughly the same either way.

I am a little disoriented by your figures. Is your avg credit per day based on the credit claim? Or the granted credit? Since your credit claimed is roughly the same, it would seem to be based upon granted credit. But your granted for HT OFF is might higher than with HT on, and the avg per day is the opposite.

---

Rosetta Moderator: Mod.Sense

I calculated average credit per day as a formula :=

Avg WU per core/day * number of cores(threads) * Avg Granted credit

HT OFF = 8,1509428836 * [b]4[/b] * 144,2326666667
HT ON = 8,3791622648 * [b]8[/b] * 89,5136111111

It was very short period of testing (2 days for each one)
I'll try some long term statistic, and then i will comapre my previous results.

Of course there is some measure error as it was not compared to exactly the same WU's.

---

I calculated average credit per day as a formula :=

Avg WU per core/day * number of cores(threads) * Avg Granted credit

HT OFF = 8,1509428836 * [b]4[/b] * 144,2326666667
HT ON = 8,3791622648 * [b]8[/b] * 89,5136111111

It was very short period of testing (2 days for each one)
I'll try some long term statistic, and then i will comapre my previous results.

Of course there is some measure error as it was not compared to exactly the same WU's.

I don't have a HT capable CPU myself, but 25% to 30% is a figure I have heard elsewhere, several times.

Incidentally, I think the price difference for a HT vs non-HT capable Intel CPU is somewhere in that area, when you account for other differences.

Incidentally, I think the price difference for a HT vs non-HT capable Intel CPU is somewhere in that area, when you account for other differences.

i7 4770K is ~45% more expensive than i5 4670K, but you gain 100 MHZ extra at stock clock. Top CPU's has always worse price/performance ratio.

---

Premier, interesting test that you did. I have one question was the task length set on default? (3 hours)

---

Have a crunching good day!!

The benefit of HT to Rosetta on an i7 is much more significant than on an i3 because of the additional L2 cache. My first gen i3 (Nehalem) gets slightly more credit running 4 threads than 2, but 3 comes out slightly ahead (although that test was around 18 months ago). The crystalwell CPUs will probably do the best because of the L4 cache.

---

Premier, interesting test that you did. I have one question was the task length set on default? (3 hours)

Yes, default task lenght was 3 hours.

Curently I am making another test to see how it's done on the same WU's (I have little problems with it).
I am also checking the impact of OC - seems to me that there is no linear gain @i7 processor (20% MHz boost is not giving 20% performance increase like in pure synthetic benchmarks, but I need more testing samples).

---

just an old HT example a 3ghz HT intel was equal to 2- 2.2 celeronds [sp] so I think I got maybe 25-30 % higher output in general