E5-2600 v1,v2,v3 article

[pgh5278]

Patrick, Thank You for the article and the power graphs, have been looking for some details like this for a while, to make a decision. The V2 information was invaluable, real power savings over v1, and not much more power than V3. In addition the availability of motherboards and CPUS for V1/V2 being similarly priced at the lower end where i was looking...
Thank YOU

 

[Patrick]

Happy to help when we can.

 

[Joel]

I agree that the charts are very enlightening. I'm in the process of building a home server/NAS and I'd love to see the entry level 6 core options included from LGA1366 (X5650 and L5640). The 1366 processors are stupid cheap now and I'm debating if the lower platform power is worth the cost delta.

I did see another article but I wasn't sure if it was mounted in a server chassis with lots of high watt fans:

http://www.servethehome.com/intel-xeon-l5640-60w-dual-core-processor-benchmarks-review-power/

​

 

[chinesestunna]

I agree that the charts are very enlightening. I'm in the process of building a home server/NAS and I'd love to see the entry level 6 core options included from LGA1366 (X5650 and L5640). The 1366 processors are stupid cheap now and I'm debating if the lower platform power is worth the cost delta.

I did see another article but I wasn't sure if it was mounted in a server chassis with lots of high watt fans:

http://www.servethehome.com/intel-xeon-l5640-60w-dual-core-processor-benchmarks-review-power/

​

In the exact same boat, been scratching my head for past 2 weeks on this myself. I did some very crude math, assuming $.10 per kWh of power, each 10W from the wall running 24x7 costs about $8.50 per year, so the 90W difference between a dual L5640 vs. E3-1220 1280 at idle (which is most of the time for us homebrew lab users), would equal $80 per year.

Of course keep in mind the motherboard Patrick used for the L5620s was loaded with 2x 5520 chipsets for all those PCI-e lanes, quad GBe NIC and onboard SAS so the base platform power consumption would probably be 40W IMO as a baseline as 5520 chipset alone is about 24W.

Regarding fans, I don't think we'll ever get an apples to apples comparison as newer chips are lower power and require less/lower power fans. You can't apply the same setup to older systems that has higher TDP and expect to be thermal safe. So in my current models/plans I count fans as a necessary piece that contributes to the overall power use, which actually widens the advantage even more towards newer Xeons.

 

[chinesestunna]

Happy to help when we can.

Also a big thank you to Patrick for providing all these insights with a good mix of data points with consideration for "ServeTheHome" type of users. The idle load power is especially key IMO when comparing platforms as one considers the typical home/lab use scenario.

 

[Joel]

Based on some quick calculations I did using common eBay prices, your analysis of idle power results in:

1366: $150 (mobo) + $150 (2x L5640s) = $300
2011: $300 (mobo) + $650 (2x e5-2620 v1) = $950

Cost delta = $650 / $80 per yr = 7.8 years (based on idle power alone).

I tried to compare the benchmarks to get a handle of the performance differences, but the only common one across the two articles was the 7zip compression. Both processors had similar performance on that metric, so I couldn't draw any meaningful conclusion.

 

[PigLover]

The story changes quite a lot when you live in an expensive power jurisdiction. PG&E country in Central/North CA averages $0.23/kwh and is tiered so that heavy users get penalized with up to $0.51/kwh. The way the tiering works it raises the cost of ALL your power consumed above the tiers - so if your server kicks you from a $0.23/kwh tier into a $0.40/kwh tier it effectively raises the price of ALL the power you consume in the household to that higher rate. I've done some spreadsheet modeling that showed my rack had an effective marginal rate of over $1.00/kwh due to power consumed by the system and adding in the increased cost of the rest of the power.

Curiously there is an out. IF you have an electric vehicle you can shift to the only rate plan that is not "tiered" this way. So I drive an EV - not because I am ecologically responsible but because it saves my money running my computers . My effective rate on this plan is about $0.21/kwh (time of date rated so its not flat) - but I can consume as much as I want without the rate going up.

 

[Patrick]

I agree that the charts are very enlightening. I'm in the process of building a home server/NAS and I'd love to see the entry level 6 core options included from LGA1366 (X5650 and L5640). The 1366 processors are stupid cheap now and I'm debating if the lower platform power is worth the cost delta.

The biggest issue I have right now is room in the lab. LGA1366 is now 4-5 generations old and that is just one line. I still have a few active, but those are not in the lab so not near the Extech power unit. Going forward, I am trying to track to standard setups so we can do this properly. STH in the LGA1366 era was just starting out so the methodologies were not there. Just looking to improve in the future.

 

[Patrick]

Based on some quick calculations I did using common eBay prices, your analysis of idle power results in:

1366: $150 (mobo) + $150 (2x L5640s) = $300
2011: $300 (mobo) + $650 (2x e5-2620 v1) = $950

Cost delta = $650 / $80 per yr = 7.8 years (based on idle power alone).

I tried to compare the benchmarks to get a handle of the performance differences, but the only common one across the two articles was the 7zip compression. Both processors had similar performance on that metric, so I couldn't draw any meaningful conclusion.

Joel check the Linux-Bench forum for the latest dev site link. There is still a bit of work to go until we flip the switch but you should be able to get the data you need now and make your own comparison graphs.

 

[chinesestunna]

Based on some quick calculations I did using common eBay prices, your analysis of idle power results in:

1366: $150 (mobo) + $150 (2x L5640s) = $300
2011: $300 (mobo) + $650 (2x e5-2620 v1) = $950

Cost delta = $650 / $80 per yr = 7.8 years (based on idle power alone).

I tried to compare the benchmarks to get a handle of the performance differences, but the only common one across the two articles was the 7zip compression. Both processors had similar performance on that metric, so I couldn't draw any meaningful conclusion.

**we totally hijacked this thread by the way, maybe piglover can move these posts to a different thread **

My math looks a bit different:
1366: $150 (mobo) + $60 (2x L5630s) = $210 (I'm not going for dual hex core, I don't need the speed as much as the power savings)
1155: $150 mobo + $130-200 e3-2xxx Xeon = $280+ (this platform is better at power but less computation power than 1366, single CPU)
2011: about same as your assessment, $950

One thing I'm considering is just running a single chip in the 1366 platform, my current setup is a i7 920 + 24GB of RAM + 12 HDDs and it's about 165W idle with fans/drives/etc. It drops to 120W when drives are spun down, compared to Xeon W3540 from Patricks review of the L5640 which is basically same chip, his idle is 48W (not sure about other components but I think it's inline with what I'm seeing with a single 1366)

Another major variation which we didn't touch on is memory capacity, 1366 dual cpu boards can easily get 12 slots of dirty cheap ECC Reg DDR3 (current eBay rates is about 6x4GB for $120 shipped). 2011 v1/v2 boards has same benefit, 1155 uses ECC UnReg which is very pricey.

I think a lot of what we go with will come down to use case, for me I'm not running anything intensive, main just a NAS VM driving my 10x3TB array which gets spun down when not in use and some other VMs for testing and exploration.

 

[chinesestunna]

The story changes quite a lot when you live in an expensive power jurisdiction. PG&E country in Central/North CA averages $0.23/kwh and is tiered so that heavy users get penalized with up to $0.51/kwh. The way the tiering works it raises the cost of ALL your power consumed above the tiers - so if your server kicks you from a $0.23/kwh tier into a $0.40/kwh tier it effectively raises the price of ALL the power you consume in the household to that higher rate. I've done some spreadsheet modeling that showed my rack had an effective marginal rate of over $1.00/kwh due to power consumed by the system and adding in the increased cost of the rest of the power.

Curiously there is an out. IF you have an electric vehicle you can shift to the only rate plan that is not "tiered" this way. So I drive an EV - not because I am ecologically responsible but because it saves my money running my computers . My effective rate on this plan is about $0.21/kwh (time of date rated so its not flat) - but I can consume as much as I want without the rate going up.

Definitely true, this may really shift the impact one way or another. I actually drive an electric as well, Nissan Leaf and love it

 

[chinesestunna]

The biggest issue I have right now is room in the lab. LGA1366 is now 4-5 generations old and that is just one line. I still have a few active, but those are not in the lab so not near the Extech power unit. Going forward, I am trying to track to standard setups so we can do this properly. STH in the LGA1366 era was just starting out so the methodologies were not there. Just looking to improve in the future.

Patrick, the site and content has come a long way and there's been ton's of improvements. I think ultimately trying to shoot for "common platform" might still be difficult because we can't really isolate the components themselves so apples to apples would probably not happen. Esp. when required cooling and other components come into play