That's true. The endless need for staggering when starting/restarting threads, is extremely tiresome and irritating. At the same time the constant tinkering is a bit addictive, in a pachinko kind of way.


Edit.

Pachinko on Wikipedia.




A modern pachinko parlor. Listen to the deafening sound.

Here is a silent instruction video aimed at foreigners that visit Japan and who would like to know how to play pachinko.

Apparently there are also videogames that mimic mechanical pachinko.

The 2% error rate is caused almost entirely by CEP2 wus. Other projects hardly ever end up in error. I must also point out that it seems as if the new, undervolted computers with standard memory are more prone to errors, than the old, stockvolted computers with ECC memory. Perhaps 0,8Vcore is too low, but the ppd/W seem to be topnotch.

RAMdisk is superior, but also expensive. Maybe I'll upgrade if and when memory prices come down. I'll probably try the cache on my new machines after Patch Tuesday, but first I'll spread the CEP2 threads so that I have 6 on every HDD, 12 per computer. This approach will also mitigate the intra crunch staggering problem with CEP2 wus.


I'm now test running 18 CEP2 threads out of 24 on one of my old machines. The rest are OET.

@symbolic link: this is geting slightly off topic, but very interesting! So they're using "custom xml linking". Taking a look into my current slots (WCG, Einstein, GPU-Grid) shows lot's of real files, but also quite a few files with just 100 - 120 bytes. Taking a look into these reveals the xml links.

Side note, quote from that wiki: "These links aren't symbolic or hard links, because Windows doesn't have them"
Yeah.. that's why Microsoft has instructions on how to use it in their Dev Center. It's available like this since Vista, and with "Link Shell Extension" it also works in XP. Anyway, that wiki article is certainly not your fault and recently I got fed up with Windows bashing.

@TBMS: OK, 12 concurrent tasks will load any drive much more than what I'm seeing. Yet I'm still not convinced CEP2 would create such a load. Consider the corresponding throughput curve:

The SSDs write at 35k - 62k writes/s for 20 - 50 minutes before performance starts to drop seriously. Let's assume the worst case, 35k/s for 20 minutes.. that's 42 mega-writes in total. HDDs usually deliver about 100 4k writes per second (200 for a Raptor and 300 to 400 for 10k or 15k rpm drives). Hence a raptor would need to write for 58 hours non-stop to transfer the same volume. Your HDD copes with the load, so it must be far less stressful, right? Which would let the SSD take short brakes, which it could use for garbage collection.

@SekeRob & topic: so far I used Ressource Monitor, which provides instantaneous read and write rates. And Task Manager, which in Win 8 has a column "disk transfer" in the advanced view, process tab. It doesn't distinguish between reads and writes, though. I'll take a closer look using the other tools within the next few days, once I've got time and catch a running CEP2.

MrS