Xeon vs. Quad CPU (775)

[EdH]

Sorry if I missed or misunderstood this somewhere else, or if I should already know better...

I'm looking over quad replacements for some dual core CPUs. Probably not really cost effective, but I've done some Q8400 for E8400 machines and am now looking specifically at some QX9770 and QX9650 CPUs on Ebay.

I keep seeing Xeons being offered with modifications and modification strips to swap a couple pins, plus notched edges to fit the 775 format.

Do these Xeons actually work and are they what they boast in performance comparisons? Don't the Xeons want ECC memory? (Not sure where I picked up the idea.)

What are the pros and cons?

Thanks!

[Mark Rose]

Is it really worth spending any money to upgrade to a 9 year old processor?

[VictordeHolland]

Core2Duo and Core2Quad don't support AVX, so in terms of performance and power efficiency they are not really useful for prime testing.

[EdH]

Originally Posted by Mark Rose

Is it really worth spending any money to upgrade to a 9 year old processor?

I guess that falls into the , "I should already know better." portion.

In reality, even if I buy something new, I would still run many of the old machines, although I have been able to bring myself to retire the 32 bit systems.

Originally Posted by VictordeHolland

Core2Duo and Core2Quad don't support AVX, so in terms of performance and power efficiency they are not really useful for prime testing.

I'm currently involved with Primo certificates and Aliquot sequences, mainly. Not sure how much advantage AVX would give me. Is AVX helpful for ECM/NFS runs?

Thanks much for the replies.

[henryzz]

It would highly surprise me if you couldn't save money in a year on power by getting the same performance for less power even with Primo and ECM/NFS.


If you run some benchmarks I can run them on my 6700K and we can compare speed.
I do know that hyperthreading is a massive help in NFS sieving on modern processors providing a 1.5x speed boost(probably due to the asm being core 2 era)

[EdH]

Originally Posted by henryzz

It would highly surprise me if you couldn't save money in a year on power by getting the same performance for less power even with Primo and ECM/NFS.


If you run some benchmarks I can run them on my 6700K and we can compare speed.
I do know that hyperthreading is a massive help in NFS sieving on modern processors providing a 1.5x speed boost(probably due to the asm being core 2 era)

Thanks for the offer, but I'm starting to be tied up with some projects here and may not have any time to do much with my machines. I'm not really sure how to benchmark. Perhaps a little later you can educate me.

[VBCurtis]

"benchmark", in this context, is as simple as you wish to make it. For instance, report the time per curve on ECM on a specific number and bound, and David will do the same on his machine; likewise for a small Primo proof, etc. You can get an idea of just how much faster a modern CPU is, and perhaps demonstrate to yourself that electricity savings pays for the cost of silicon (I bet it's not within 1 year, but is within two).

[EdH]

Would this work?

On one of my faster machines, I used Primo to certify a 3712 dd number with the following input:

;This is an input file for PRIMO
;Number expressed to the base 16
;The database-id of this number is 1100000000294458182, it has 3712 digits.
;Short form is 8^4110+7

[Candidate]
N$=40000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000007

It took just over 4 hours and produced the following certificate (only the first part is shown):

[PRIMO - Primality Certificate]
Version=4.2.1 - LX64
WebSite=http://www.ellipsa.eu/
Format=4
ID=B3CE504C4F4A9
Created=06-05-2017 10:13:40 PM
TestCount=477
Status=Candidate certified prime

[Comments]
Put here any comment...

[Running Times (Wall-Clock)]
1stPhase=10666s
2ndPhase=4381s
Total=15047s

[Running Times (Processes)]
1stPhase=41619s
2ndPhase=17493s
Total=59113s
...

Would this be too long for a benchmark?

For my machines that have hyper-threading, I have not noticed it to be an advantage with Primo.

[EdH]

Here's a much shorter certificate I just ran:

;This is an input file for PRIMO
;Number expressed to the base 16
;The database-id of this number is 1100000000936417096, it has 1087 digits.

[Candidate]
N$=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

with the result:

[PRIMO - Primality Certificate]
Version=4.2.1 - LX64
WebSite=http://www.ellipsa.eu/
Format=4
ID=B3CE60364936B
Created=06-06-2017 03:48:43 PM
TestCount=160
Status=Candidate certified prime

[Comments]
Put here any comment...

[Running Times (Wall-Clock)]
1stPhase=54.08s
2ndPhase=11.42s
Total=66s

[Running Times (Processes)]
1stPhase=177s
2ndPhase=43.15s
Total=220s

I suppose this one is possibly too short to be meaningful, but I will try to have a more moderate one a little later.

Thanks everyone!

[Dubslow]

Hyperthreading is *definitely* a benefit for NFS sieving and ECM, and I think linear algebra too (up to a point). 50% gains vs no hyperthreads, on the exact same processor/core speed, is very reasonable to expect. I couldn't say why Primo doesn't seem to benefit as such, it almost certainly should?

[GP2]

Originally Posted by Dubslow

Hyperthreading is *definitely* a benefit for NFS sieving and ECM, and I think linear algebra too (up to a point). 50% gains vs no hyperthreads, on the exact same processor/core speed, is very reasonable to expect. I couldn't say why Primo doesn't seem to benefit as such, it almost certainly should?

In mprime 29.1, hyperthreading is on by default for factoring but can be turned off, and is off by default for LL testing but can be turned on, and ... I presume for ECM or PRP or P−1 or anything else, it's always on and is unaffected by any HyperthreadLL=0 or any other settings in local.txt ?