Romex Software Forum

I've been doing some tests with E3 and E5 (dual 6-core CPU) servers and been getting poor results on the E5.

I benchmarked the RAM using STREAM (http://www.cs.virginia.edu/stream/ref.html) and the E5 seems to need more than 8 threads to get good results while the E3 performs very well even with a single thread. Of course the E5 obliterates the E3 with 16 threads and more.

My question is how many threads does FancyCache use? I suspect that it is single threaded since the difference in performance in CrystalDiskMark (and others) is pretty much the same with the difference in STREAM benchmarks.

What are you trying to accomplish? Or is this just an academic question?

dustyny wrote:What are you trying to accomplish? Or is this just an academic question?

I'm afraid it's not academic at all, I wish it was actually...

The problem is that on the E5 I get 1/5 of the performance I get from the E3 when using FancyCache. I've been troubleshooting the issue but if FancyCache is using just one thread then that's it. No way I can get the same performance from the E5...

In the end using more then one thread (at least two) may return higher performance on high-end, too, academically

I assume according to the support questionising your simple yes/no-question with if it is academically instead of simply answering, that means "yes, it is single threaded, sry", they just forget to mention that. To say something in their defence, it is pretty hard to transfer the code of a "single threaded block-layer system-wide r/w-cache-application" to multi-threaded layout Basically that may mean a complete re-write of ALL program source-code.

I personally think you can start doing that when the product is final and all major-bugs are fixed.

I'm curious as to what you are doing. I've tested F.C on a few E3's & E5's (HP & Dell's), Core i7 and some older processors such a Core duo & a Core2 duo, I never really saw any real hit to the CPUs.