Desktop CPU VS server CPU performance - is passmark relevant?
I am replacing my SOHO server (desktop-grade components) with a used server. A quad-core i5 is being replaced with dual quad-core Xeons X5355, here are the details:
i5-2300, passmark score 5276:
http://www.cpubenchmark.net/cpu.php?cpu=Intel+Core+i5-2300+%40+2.80GHz
Dual Xeon X5355, passmark score 5880:
http://cpubenchmark.net/cpu.php?cpu=Intel+Xeon+X5355+%40+2.66GHz&id=1294&cpuCount=2
My question is, with the difference between passmark scores being only ~600, is it correct to assume that the performance would be similar, or are there any other advantages of the server board? My main reason is for the move is ECC-RAM and dual-PSU reliability and other monitoring advantages.
In general server-grade hardware will perform better than desktop-grade equivalents when under load from multiple processes (e.g. running several services serving requests, hosting multiple users, etc). For day to day desktop usage server hardware won't necessarily perform any better, and can underperform. However, server hardware typically is less error and fault prone, and is often found in "workstation-grade" hardware--effectively high-end desktops used for professional purposes such as CAD, animation, or running local services like databases.
For a typical desktop user it's better to focus on the hardware metrics that reduce latency. More L2 or L3 cache on a CPU can be more beneficial than many more cores--a faster FSB frequency can be better than extra clock speed, etc.
The typical tradeoffs are that server hardware is more expensive, can (in some cases) consume more electricity, and isn't fully utilized in a desktop environment. In terms of performance per dollar for desktop usage it's usually best to get high quality desktop components--a motherboard with a high-end chipset, quality RAM, and quality SSD(s). Many-core server CPUs, ECC memory, and SAS drives don't provide the gains to a desktop that they do to a server.
All else being roughly equal, a processor is a processor. "Server" processors tend to have larger caches which is useful for switching among many tasks, but does less with single-threaded processing. What processor is going to give an edge is going to be highly dependent on the load you're placing on the processor. The Passmark score is going to be a good general indication of performance in general, so in general your new server should be slightly faster than the old one, and in some specific applications will be noticeable faster.