Intel E5-2687w Workstation Processors for Server?

[swalsh2333]

I was able to pick-up (2) INTEL XEON E5-2687W 8 Core LGA2011 3.1G 20MB L3 150W cpu's for a great deal. After I bought them I realized they are 150w processors and meant for a server or workstation. Most barebone servers only support 135w max. It's my understanding that I would need to buy a 4U workstation/server to give these processors ample space for cooling. I need a high powered server and I my question is:

Is it ok to use this approach (a workstation) for a server. Are there disadvantages in doing it this way opposed to just buying different processors and a traditional 2U barebone server.

This is the workstation:

http://www.supermicro.com/products/system/4U/7047/SYS-7047R-3RF4_.cfm

THANKS!!

 

[nforce4max]

It depends on a few factors. If the board has a safety margin on the power vrm you can uses these 150w models (think of it as overclocking situation but nothing is different or changed except 15w each) and the ability to keep them cool. If the quality of the board is good enough using 150w samples won't have much of a difference in life span unless they got cheap when they made the board. The cooling well you are going to have to deal with more noise than perhaps normal even for a server due to the coolers. Likely they are somewhat low mass even for the 135w models so dealing with these 150w instead will leave temps higher and fans will be close to max due to thermal profiles.

30w difference total means nothing to the psu so beyond that drop them in and forget.

EDIT: maybe you should look into doing a mod on the board if there is room for it using vram or mosfet coolers which ever fits best. A pack of eight only runs for about $10-15 a pack. Two packs total and use them on the power vrm on the board for each cpu and it will extend life by a few years. Don't worry only one side of the mosfets are conductive. The side facing out towards you when working on the build is a insulator (black plastic) so there won't be any shorts from that effort.

 

[swalsh2333]

It depends on a few factors. If the board has a safety margin on the power vrm you can uses these 150w models (think of it as overclocking situation but nothing is different or changed except 15w each) and the ability to keep them cool. If the quality of the board is good enough using 150w samples won't have much of a difference in life span unless they got cheap when they made the board. The cooling well you are going to have to deal with more noise than perhaps normal even for a server due to the coolers. Likely they are somewhat low mass even for the 135w models so dealing with these 150w instead will leave temps higher and fans will be close to max due to thermal profiles.

30w difference total means nothing to the psu so beyond that drop them in and forget.

EDIT: maybe you should look into doing a mod on the board if there is room for it using vram or mosfet coolers which ever fits best. A pack of eight only runs for about $10-15 a pack. Two packs total and use them on the power vrm on the board for each cpu and it will extend life by a few years. Don't worry only one side of the mosfets are conductive. The side facing out towards you when working on the build is a insulator (black plastic) so there won't be any shorts from that effort.

Thanks for the info. Appreciate it!

 

[ChicagoBud]

Thanks for the info. Appreciate it!

Did this work out for you?

 

[swalsh2333]

we just received the OCZ Talos 3.5 SAS drives and the LSI controller card so I'll know in a week or two. Thanks

 

[ChicagoBud]

Check your PM's

 

[swalsh2333]

Check your PM's

I don't see any PM's

 

[ChicagoBud]

Sort of a hidden feature I guess. Should be at the top of the right column which might be hidden. Othierwise, this link might work: http://www.tomshardware.com/forum/forum1.php?config=tomshardwareus.inc&cat=prive.

Anyway, here was what I was asking...

Are you going to run linux or windows?

I'm running a similar box (1U supermicro with dual e5-2687w) w/ linux. For the most part, everything is running fine with the exception of the TCS's not being sync'd. The server runs fine but falls back to using the PET for timing which kills my application. If you get a chance, can you see if you have the same issue:


$ dmesg | grep -i "tsc\|cpu0"
[ 0.000000] Fast TSC calibration using PIT
[ 0.010484] CPU0: Thermal monitoring enabled (TM1)
[ 0.073825] CPU0: Intel(R) Xeon(R) CPU E5-2687W 0 @ 3.10GHz stepping 07
[ 1.047492] TSC synchronization [CPU#0 -> CPU#8]:
[ 1.047495] Measured 9353543 cycles TSC warp between CPUs, turning off TSC clock.
[ 0.008000] Marking TSC unstable due to check_tsc_sync_source failed

Thanks!

-- Bud

 

[swalsh2333]

Sort of a hidden feature I guess. Should be at the top of the right column which might be hidden. Othierwise, this link might work: http://www.tomshardware.com/forum/forum1.php?config=tomshardwareus.inc&cat=prive.

Anyway, here was what I was asking...

Are you going to run linux or windows?

I'm running a similar box (1U supermicro with dual e5-2687w) w/ linux. For the most part, everything is running fine with the exception of the TCS's not being sync'd. The server runs fine but falls back to using the PET for timing which kills my application. If you get a chance, can you see if you have the same issue:


$ dmesg | grep -i "tsc\|cpu0"
[ 0.000000] Fast TSC calibration using PIT
[ 0.010484] CPU0: Thermal monitoring enabled (TM1)
[ 0.073825] CPU0: Intel(R) Xeon(R) CPU E5-2687W 0 @ 3.10GHz stepping 07
[ 1.047492] TSC synchronization [CPU#0 -> CPU#8]:
[ 1.047495] Measured 9353543 cycles TSC warp between CPUs, turning off TSC clock.
[ 0.008000] Marking TSC unstable due to check_tsc_sync_source failed

Thanks!

-- Bud

Hey Bud,

I will let you know. I'm not an IT guy but I was told that I couldn't run these processors in a 1U or 2U chassis because they are 150w. I don't know if that helps you but you may want to check to see if this has anything to do with your issue.

Thanks,
Scott

 

[ChicagoBud]

I'm not an IT guy but I was told that I couldn't run these processors in a 1U or 2U chassis because they are 150w.

Yeah, that would be the conservative approach. However, they work fine with this exception and I think is is more likely a bug based on some test utilities I've run. There are no heat issues - at least with the load I am applying:


$ sensors
coretemp-isa-0000
Adapter: ISA adapter
Physical id 0: +47.0°C (high = +82.0°C, crit = +92.0°C)
Core 0: +44.0°C (high = +82.0°C, crit = +92.0°C)
Core 1: +43.0°C (high = +82.0°C, crit = +92.0°C)
Core 2: +47.0°C (high = +82.0°C, crit = +92.0°C)
Core 3: +47.0°C (high = +82.0°C, crit = +92.0°C)
Core 4: +43.0°C (high = +82.0°C, crit = +92.0°C)
Core 5: +39.0°C (high = +82.0°C, crit = +92.0°C)
Core 6: +46.0°C (high = +82.0°C, crit = +92.0°C)
Core 7: +42.0°C (high = +82.0°C, crit = +92.0°C)

coretemp-isa-0001
Adapter: ISA adapter
Physical id 1: +43.0°C (high = +82.0°C, crit = +92.0°C)
Core 0: +42.0°C (high = +82.0°C, crit = +92.0°C)
Core 1: +43.0°C (high = +82.0°C, crit = +92.0°C)
Core 2: +43.0°C (high = +82.0°C, crit = +92.0°C)
Core 3: +43.0°C (high = +82.0°C, crit = +92.0°C)
Core 4: +40.0°C (high = +82.0°C, crit = +92.0°C)
Core 5: +43.0°C (high = +82.0°C, crit = +92.0°C)
Core 6: +44.0°C (high = +82.0°C, crit = +92.0°C)
Core 7: +40.0°C (high = +82.0°C, crit = +92.0°C)


-- Bud

 

[KoSoVaR]

Yeah, that would be the conservative approach. However, they work fine with this exception and I think is is more likely a bug based on some test utilities I've run. There are no heat issues - at least with the load I am applying:


$ sensors
coretemp-isa-0000
Adapter: ISA adapter
Physical id 0: +47.0°C (high = +82.0°C, crit = +92.0°C)
Core 0: +44.0°C (high = +82.0°C, crit = +92.0°C)
Core 1: +43.0°C (high = +82.0°C, crit = +92.0°C)
Core 2: +47.0°C (high = +82.0°C, crit = +92.0°C)
Core 3: +47.0°C (high = +82.0°C, crit = +92.0°C)
Core 4: +43.0°C (high = +82.0°C, crit = +92.0°C)
Core 5: +39.0°C (high = +82.0°C, crit = +92.0°C)
Core 6: +46.0°C (high = +82.0°C, crit = +92.0°C)
Core 7: +42.0°C (high = +82.0°C, crit = +92.0°C)

coretemp-isa-0001
Adapter: ISA adapter
Physical id 1: +43.0°C (high = +82.0°C, crit = +92.0°C)
Core 0: +42.0°C (high = +82.0°C, crit = +92.0°C)
Core 1: +43.0°C (high = +82.0°C, crit = +92.0°C)
Core 2: +43.0°C (high = +82.0°C, crit = +92.0°C)
Core 3: +43.0°C (high = +82.0°C, crit = +92.0°C)
Core 4: +40.0°C (high = +82.0°C, crit = +92.0°C)
Core 5: +43.0°C (high = +82.0°C, crit = +92.0°C)
Core 6: +44.0°C (high = +82.0°C, crit = +92.0°C)
Core 7: +40.0°C (high = +82.0°C, crit = +92.0°C)


-- Bud

Bud,

Did you ever figure out why the clocks weren't syncing?

I'm also getting this on an i7 3970x running kernel 3.2.2.

 

[ChicagoBud]

Bud,

Did you ever figure out why the clocks weren't syncing?

I'm also getting this on an i7 3970x running kernel 3.2.2.

No. I was working with supermicro on this recently but we never really got it sorted. They sort of gave up after they couldn't reproduce the issue with a pair of E5-2643 quad core chips. I had the issue with dual e5-2687w and e5-2690 chips - not syncing between the two CPUs.

I ended up using an intel chassis with e5-2690's. That combo didn't have the TSC sync issue. In general though, I've had issues dealing with the overly aggressive power management capabilities of these chips.


-- Bud

 

[KoSoVaR]

No. I was working with supermicro on this recently but we never really got it sorted. They sort of gave up after they couldn't reproduce the issue with a pair of E5-2643 quad core chips. I had the issue with dual e5-2687w and e5-2690 chips - not syncing between the two CPUs.

I ended up using an intel chassis with e5-2690's. That combo didn't have the TSC sync issue. In general though, I've had issues dealing with the overly aggressive power management capabilities of these chips.


-- Bud

You've been working on this for quite awhile. I'm overly frustrated with ASUS response thus far and I tried a few things that I found quite interesting. I'm overclocking my 3970x to 5GHz, I'm certain it's 100% rock stable by using Prime95 for 15hrs+ to bench it.

I get the tsc error over and over again. I have two of the same exact machines with different CPUs (one 3970x and one 3930k). The motherboard I believe is the culpirit, just as in your case. It's the ASUS Rampage IV Extreme.

So I tried on my home PC which has an ASUS Z77 Sabertooth and a 3770k. It worked, TSC was sync'd correctly with Ivy Bridge.

I'm certain it's the motherboard in my case and I'm going to keep pushing ASUS for help. The overclocking abilities of this board are insane.

I ended up picking up three new boards which I'll test today. I have a ASUS P9X79 Pro and a ASUS X79 Sabertooth. I also have the ASRock X79 Extreme11 coming in tomorrow.

And lastly, I have tried several different kernels and several different PCs. It always leads back to the motherboard - but I have yet to get a SB-E X79 board with one of my CPUs to test this out, and this is why I have the boards. I'll even put a 2690 in the machine and see if I still get the error. What do I got to lose? Then I'll really point all fingers @ ASUS.

 

[KoSoVaR]

Sabertooth X79 reacts the same way. No TSC sync.