Threw together a little chart based off the CPU compatibility list for my HP Z600. Should be valuable for anyone else looking at a Xeon 5600 series cruncher.
Assuming processors can not be overclocked and will run boost clocks 24/7, here's a little chart to compare Xeon 5600 series CPUs. Variables are for my specific region, but all formulas in my spreadsheet use those variables, so I should be able to easily give out a customized list if anyone's interested. All wattages do not include anything except processor wattage. I should be including some constant system wattage in the formulas since two systems with 50w CPUs will consume more than one system with a 100w CPU. I'll see about updating this later. (Edit: Done) (Anyone know how much wattage to add for mobo/mem/HDD/etc? 30w?)
Legend:
Code:
HT = Assuming 30% performance boost from having HT enabled. Total freq is multiplied by 30% in HT-enabled CPUs. Pulled this number from my dealings with HT, but open to suggestions on more realistic figures.
$$$ = Current eBay prices (non-ES chips)
Total freq = Boost freq * cores * HT (rough indication of BOINC performance, same architecture, so should be accurate)
Freq/100w = Total freq / TDP (very rough indication of power efficiency, probably inaccurate)
Freq/$$ = Total freq / ebay prices (performance per dollar)
1 yr cost = eBay prices + power bill (CPU wattage + system wattage / psu efficiency * math to convert to kWh * $/kWh)
2 yr cost = eBay prices + 2x power bill (CPU wattage + system wattage / psu efficiency * math to convert to kWh * $/kWh)
Freq/1yr $ = How much performance each dollar gives you after 1 year of work (more is better, this is the result I'm most interested in)
Freq/2yr $ = Same as above, but over a two year period instead. Procs that are more expensive to buy, but use less power will start taking the lead eventually.
Notes/Conclusions: E5640 wins price/performance after 1 year, E5649 wins price/performance after 2 years.
Updated to add 30w to all calculations (estimated power consumption of mobo/memory/hdd/etc). E5649 squeeks ahead for both the 1yr and 2yr price/perf. Should factor in eBay prices for rest of system too. Might change the rankings yet again.
I'm using TDPs, so these are probably all wrong. I might try an overclock calculator to get a better estimate on actual CPU wattage.
Seeking input/advice on:
Typical wattage of system parts (mobo, memory, hdd, etc)
Real-world wattage of listed CPUs
Performance increase with HyperThreading on vs off
Important cpus you forgot there are the L56xx xeons. They are 6 cores and rated ad 60W tdp. the L5640 is actually really cheap. @Knoxx29 is in the process of setting up a rig like this right now. I did a similar spreadsheet earlier when looking at CPUs with knoxx but not sure if i saved it but the conclusion was (i used cinebench to compare performance between different generations) if i remeber well that for 1 year running hardware going + initial cost was that the X5670 i think was the cheapest followed by the L5640 and then the e3-1240 v2. Ofcourse this depends on the initial cost of everything and the $/KWh
I've seen on ebay listing for a pair of "matched" x5650 CPU's. What does matched refer to exactly? Judging from what I'm looking at, I'm assuming it refers to perhaps batch numbers. If so, how important is that? Thanks for any help
Usually @manofthem it pretty much refers to a pair that were purchased or included in a server that are known to work together. I would imagine that the batch numbers are the same or very similar.
I have seen many use totally different chips purchased from different parties in a multi processor board and work just fine. However one never knows and finding a "matching" pair will insure they work good together.
Might be moving out in the next couple weeks if everything goes according to plan, though I might have an issue with second month's rent as my boss is taking a vacation and we don't work then.
Usually @manofthem it pretty much refers to a pair that were purchased or included in a server that are known to work together. I would imagine that the batch numbers are the same or very similar.
I have seen many use totally different chips purchased from different parties in a multi processor board and work just fine. However one never knows and finding a "matching" pair will insure they work good together.
One more, do those dual socket server boards require ecc memory, or will it work with regular? I was looking at a link t_ski posted and it says "192gb ecc registered dimm/48gb unbuffered dimm" so thought maybe it would work with either....
Sorry for my ignorant questions on these setups but I'm just ignorant about them
One more, do those dual socket server boards require ecc memory, or will it work with regular? I was looking at a link t_ski posted and it says "192gb ecc registered dimm/48gb unbuffered dimm" so thought maybe it would work with either....
Sorry for my ignorant questions on these setups but I'm just ignorant about them
Many use both but some boards will only recognize ECC registered memory. If your looking at a specific board, do research and find the factory link as it will state memory acceptance.
Many use both but some boards will only recognize ECC registered memory. If your looking at a specific board, do research and find the factory link as it will state memory acceptance.
On matched pairs, I've usually seen that on ES chips. It probably doesn't matter for full production versions since in theory they're standardized even if they have different steppings but it might matter for ES versions. Fortunately a lot of ES chips that show up on ebay seem to come from the same small group of batches. I would take that into consideration for ES versions but be much less concerned for production versions.
I've seen on ebay listing for a pair of "matched" x5650 CPU's. What does matched refer to exactly? Judging from what I'm looking at, I'm assuming it refers to perhaps batch numbers. If so, how important is that? Thanks for any help
Matched CPUs refer to the same steppings, as in both chips are SR1AM or whatever. That means they are most compatible with each other. Sometimes if you get different steppings you may have different features, TDP or ram compatibility (like V1 1155 chips and ECC).
One more, do those dual socket server boards require ecc memory, or will it work with regular? I was looking at a link t_ski posted and it says "192gb ecc registered dimm/48gb unbuffered dimm" so thought maybe it would work with either....
Sorry for my ignorant questions on these setups but I'm just ignorant about them
A server board will often support a high amount of ram but only if it's ECC. Otherwise, standard ram would be far less dense, as with the example you listed above. Like 4GB sticks if they are regular DDR3 or 16GB sticks if they are ECC DDR3.
Matched CPUs refer to the same steppings, as in both chips are SR1AM or whatever. That means they are most compatible with each other. Sometimes if you get different steppings you may have different features, TDP or ram compatibility (like V1 1155 chips and ECC).
A server board will often support a high amount of ram but only if it's ECC. Otherwise, standard ram would be far less dense, as with the example you listed above. Like 4GB sticks if they are regular DDR3 or 16GB sticks if they are ECC DDR3.
Yeah, I just looked at that article a little while ago. If they run at comparable clock speeds to the Haswells those will make the 14 and 18 core chips looks like Raspberry Pi's.
Yeah, I just looked at that article a little while ago. If they run at comparable clock speeds to the Haswells those will make the 14 and 18 core chips looks like Raspberry Pi's.
It is silvermont architecture so I wouldn't expect as much clock for clock as haswell, and probably low clock speed. Still though, just like amd opterons are great crunchers these should do pretty great numbers.
Just shut down 2 rigs because some nasty storm rolled in. It's been a long while since a bad thunderstorm showed its head round these parts, believe it or not, and since the thunder and lightning is pretty nuts (scaring the little girls ), I shut them down. The one rig is plugged into a protector thing so it's still running for now, unless it gets worse
Just shut down 2 rigs because some nasty storm rolled in. It's been a long while since a bad thunderstorm showed its head round these parts, believe it or not, and since the thunder and lightning is pretty nuts (scaring the little girls ), I shut them down. The one rig is plugged into a protector thing so it's still running for now, unless it gets worse
