| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Completed Research
Forum: The Clean Energy Project - Phase 2 Forum
Thread: GPU Units for CEP??? |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 13
|
|
| Author |
|
|
ryan222h
Senior Cruncher Joined: Sep 4, 2006 Post Count: 425 Status: Offline |
I was running an older 32 gb SSD with wcg when i decided to let it loose on some CEP2 wu. Within a few minutes the computer was unresponsive and eventually the whole O/S became useless! The drive is pretty much done for, and i had to purchase a replacement. It was probably on its last leg, anyway.
---------------------------------------- |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Wow. That's amazing. Very surprised that CEP coupled with GPU tasks can knock a SSD RAID down. Although the RAID sets I lost were mechanical, I would personally blame it on the particular Marvell 9123 controller chip/firmware/driver.I joined a long list of consumers who found out that buying boards with Marvell 91XX chips and running RAID on them at SATA 3.0/6 Gb/s "works"...up until you run 'em hard. Eight (8) simultaneous CEP2 tasks on an i7 950 + 1 Collatz Conjecture task on an ATI GPU appears to meet or exceed the definition of "hard". Whether or not the hard drive is SSD or mechanical shouldn't matter except in terms of efficiency; any hard drive/RAID combination should take the heat - the driver should generate an I/O wait (to the O/S to the application) if the I/O queue becomes full. Unless, that is, the driver, firmware, or the I/O hardware itself doesn't have its act together. The thing is I was pushing the performance envelope to its absolute max - 100% CPU load on all threads/cores, 100% GPU load with the GPU task intermittently stealing cycles (0.1% with Collatz, IIRC) from the CEP2 tasks. Under those conditions, you're not just "running hard"; in the vernacular of materials testing, you're destructive testing. I assume the Marvel drivers/firmware didn't correctly signal/coordinate its state with with the chip or whatever...which was a bummer. (But I did did heat 4 upstairs rooms all winter just with computers.) Since that experience (repeated twice), I stick to running the CPUs flat-out iif. its an Intel controller and let the GPUs goof off. |
||
|
|
Jim1348
Veteran Cruncher USA Joined: Jul 13, 2009 Post Count: 1066 Status: Offline Project Badges: 
|
I was running an older 32 gb SSD with wcg when i decided to let it loose on some CEP2 wu. Within a few minutes the computer was unresponsive and eventually the whole O/S became useless! The drive is pretty much done for, and i had to purchase a replacement. It was probably on its last leg, anyway. I have two SSDs with good write performance (a Samsung Series 470 and an OCZ Vertex 2) on two different PCs. But I still don't trust them with CEP2, so I use a RAMDisk on one of them (Dataram) to hold the BOINC data folder, and a DRAM cache on the other (FancyCache). The Dataram RAMDisk is cheap, and even free if you don't use more than 4 GB, and the FancyCache is presently in Beta with a 180 day free trial period. The main restriction is that you have to protect against data loss in case of a power outage, and so I use backup power supplies (APC Back-UPS ES 750 and BX1000G). And of course you need enough DRAM. The ramdisk is more demanding, since CEP2 requires 2 GB free disk space before it will even download, so I use at least 4GB. The cache only requires around 1 or 2 GB memory space if you set it to write-only, depending on the latency you set. Also, if your PC is not stable to begin with and tends to crash, then don't touch either a ramdisk or a cache with a ten foot pole. Otherwise, they are great. All the memory writes go to the DRAM, and so are very fast, and don't touch your SSD. [Edit 1 times, last edit by Jim1348 at May 16, 2012 6:55:49 PM] |
||
|
|
|