Questions regarding non-ECC/reg DIMMs for Supermicro X9DRL-iF server board

[Martin]

Hello everyone,

I am currently in the process of adding some RAM to a server board that is used for a workstation.
We want to use the fastest RAM possible, and most ECC/reg 1600 DIMMs are CAS 11 which is
very slow. Please note that there is no use for reg/ECC in this particular computer.

So we would like to use either 2 of this kit:
http://www.corsair.com/en/vengeance...l-ddr3-memory-kit-cml16gx3m4x1600c7.html#tab5

Or 2 of this kit:
http://www.gskill.com/products.php?index=469

Both kits are rated 7-8-8-24 at 1600. But the question is, will this speed be detected in the motherboard?

I am getting mixed answers from the manufacturers. Supermicro says it doesn't support XMP profiles.
However, it has a "Force 1600" speed function in the BIOS.

I would appreciate any input on this matter.

Thank you for your time.

 

[Jeggs101]

Most server motherboards do not use XMP and most vendors are doing 1600/C11. Would suggest it won't work.

Why are you looking for faster RAM like that? Might more RAM help instead? Your speed difference will be under 3% in anything other than a stream benchmark. Seems like you are trying to use low quantity of RAM in a dual socket mobo. 32GB for these is nothing. Most Os these days can use way more than that.

 

[Martin]

Most server motherboards do not use XMP and most vendors are doing 1600/C11. Would suggest it won't work.

Why are you looking for faster RAM like that? Might more RAM help instead? Your speed difference will be under 3% in anything other than a stream benchmark. Seems like you are trying to use low quantity of RAM in a dual socket mobo. 32GB for these is nothing. Most Os these days can use way more than that.

Hi Jeggs101,

Thank you for your reply.

Why are you looking for faster RAM like that?
I was kind of surprised that the "standard" for 1600 seems to be CL 11. There should be quite a difference moving to CL 7,
perhaps even more than your suggested 3%. Or am I way out here?

Might more RAM help instead?
G.skill also has a similar kit with 4 x 8 GB instead of 4 x 4 GB with the same specs. The amount of RAM doesn't really matter much as
the only thing this workstation will do is to calculate chess positions using fixed Hash-sizes for the programs used. Typically this will be
limited to 8 GB or 16 GB per program, and only one program will run at any time using all 16 cores.

Hope this clears things up a bit.

 

[Jeggs101]

Why are you looking for faster RAM like that?
I was kind of surprised that the "standard" for 1600 seems to be CL 11. There should be quite a difference moving to CL 7,
perhaps even more than your suggested 3%. Or am I way out here?

Very few applications actually see anything tangible from faster ram. Manufacturers use it as a marketing ploy to sell gamers more expensive memory. That's why most servers go for more ram rather than faster. Not many applications depend on timing that much.

Might more RAM help instead?
G.skill also has a similar kit with 4 x 8 GB instead of 4 x 4 GB with the same specs. The amount of RAM doesn't really matter much as
the only thing this workstation will do is to calculate chess positions using fixed Hash-sizes for the programs used. Typically this will be
limited to 8 GB or 16 GB per program, and only one program will run at any time using all 16 cores.

Got it. This might be an app where faster timings might help. May be better to use an asus z9pe-d8 ws or supermicro hyper-speed solution here. Maybe Patrick or others can weigh.

 

[Mike]

Im pretty sure it will just revert to SPD speeds, which is 1333mhz with 9ns latency for those modules.
Doesn't EVGA have a dual socket board that has all the tweaks and stuff on board? I'm pretty sure that board is marketed to enthusiasts, and about the only one for that matter.

 

[Patrick]

Im pretty sure it will just revert to SPD speeds, which is 1333mhz with 9ns latency for those modules.
Doesn't EVGA have a dual socket board that has all the tweaks and stuff on board? I'm pretty sure that board is marketed to enthusiasts, and about the only one for that matter.

My sense is that this will run at SPD speeds also. Would suggest going for the server memory. Beyond that, the EVGA board is a thought, although quite expensive.

The ASUS Z9PE-D8 is a good alternative as is the Supermicro Hyper-Speed line. Here is a link to the Hyper-Speed BIOS memory settings.

 

[Martin]

Hello Patrick,

Thank your for your input. The EVGA SR-X has been discontinued though.

The already mentioned Supermicro board has already been ordered, but I guess I could sell it as new and buy a Hyper-Speed variant or the Asus board.
The case (Cooler Master HAF X) is limited to EATX though; the Supermicro X9DAX-iF is bigger with EEATX (could probably fit) and the Asus board is SSI EEB form factor which I don't believe has the same mounting holes.

The question is whether it would be beneficial, I mean a 5% or so increase would be very nice and worth it on my end. I have yet to find a memory bandwidth/latency
calculator where one can input the latencies and speed and get an automatic calculation of the theoretical throughput. Does anyone know of such a program/website?

My sense is that this will run at SPD speeds also. Would suggest going for the server memory. Beyond that, the EVGA board is a thought, although quite expensive.

The ASUS Z9PE-D8 is a good alternative as is the Supermicro Hyper-Speed line. Here is a link to the Hyper-Speed BIOS memory settings.

 

[Patrick]

Hello Patrick,

Thank your for your input. The EVGA SR-X has been discontinued though.

The already mentioned Supermicro board has already been ordered, but I guess I could sell it as new and buy a Hyper-Speed variant or the Asus board.
The case (Cooler Master HAF X) is limited to EATX though; the Supermicro X9DAX-iF is bigger with EEATX (could probably fit) and the Asus board is SSI EEB form factor which I don't believe has the same mounting holes.

The question is whether it would be beneficial, I mean a 5% or so increase would be very nice and worth it on my end. I have yet to find a memory bandwidth/latency
calculator where one can input the latencies and speed and get an automatic calculation of the theoretical throughput. Does anyone know of such a program/website?

I don't have that... if you get me the equations, I'd be willing to pay to have it developed.

On the SR-X, I can understand why. No IPMI even.

On the 5% increase, depends on what you are doing. If it is academic and finishes in a few minutes anyway, then not a big deal to me. If it is something that runs for multiple days, that would push me towards getting more performance (although CPU speed may also be a factor there.) Also, if you are just trying to hit chess benchmark records, then I want to see your config as I have been looking for a great chess benchmark application.

 

[Martin]

I would say that each engine uses about 10 hours of time, at max, meaning x 16 in CPU time as I will be using dual 2689's.
The whole event runs for about 2 months, which is separated by different stages.
Right now I only use a 3770k that is overclocked, but I have already begun buying the parts for the dedicated Xeon box.

You can use several benchmark tools, Fritz Chess Benchmark is one that is often used.
http://www.sedatcanbaz.com/chess/?page_id=105
http://sedatchess.110mb.com/index.php?p=1_72

Anyway, this computer is not for benchmarking.

I don't have that... if you get me the equations, I'd be willing to pay to have it developed.

On the SR-X, I can understand why. No IPMI even.

On the 5% increase, depends on what you are doing. If it is academic and finishes in a few minutes anyway, then not a big deal to me. If it is something that runs for multiple days, that would push me towards getting more performance (although CPU speed may also be a factor there.) Also, if you are just trying to hit chess benchmark records, then I want to see your config as I have been looking for a great chess benchmark application.

 

[MiniKnight]

If you already have those cpus - overclock even 2% for cpu and many percent for memory would bebe nice. 600min 2% = 12 min faster.

 

[renderfarmer]

You're application is unfamiliar to me, but if it's anything like Ray Tracing my testing has shown less than 0.2% performance variance between CAS8 and CAS11 RAM. Statistically insignificant.

But if this is important enough to you I'd just buy two sets from Amazon and test both out. Return the ones you don't like for a full refund minus $6 shipping.

 

[renderfarmer]

I should add that so far, all of the 1600MHz UDIMMs I've tried in each of the X9DRL boards I have default to 1333MHz @ 8-8-8-20 timings.