Xeon Owners Club

[biffzinker]

X5677 makes most sense but i really want a 5.0ghz cpu.

That's going to be difficult to obtain on a older X56xx Westmere chip.

 

[FilipM]

It's not that can't me done, it's how long will it last...you are looking at 1.45V+ for those speeds.

 

[FireFox]

The X5677 makes most sense but i really want a 5.0ghz cpu.

X5687 - stock 3.60 GHz turbo 3.86, and because i know you are good Overclocking Xeons i bet you get it at 5.0GHz

Second option: X5698 - stock 4.40GHz but i don't think 2 cores will be enough for you

That's going to be difficult to obtain on a older X56xx Westmere chip.

Not if you pair 2 x X5698 and i bet it could be done with a single X5687, of course for that you need a good 1366 socket board.

 

[CAPSLOCKSTUCK]

Cinebench R 15 record is 4.9ghz for both of them ( at the moment)

http://hwbot.org/submission/3357350_12ax7_cinebench___r15_xeon_x5677_777_cb


http://hwbot.org/submission/2853503_aleslammer_cinebench___r15_xeon_x5687_754_cb/

 

[FireFox]

Cinebench R 15 record is 4.9ghz for both of them ( at the moment)

But you are you and i trust you

 

[FilipM]

X5660 record is at 5.1...just sayin

 

[FireFox]

X5660 record is at 5.1...just sayin

24/7?

 

[agent_x007]

View attachment 85550
This is what I settled on. Somehow still slower than my old ram @ 1600 (9-9-9-24). I can't find which combination of sub-timings is making all the difference.

Your North Bridge Clock sucks (which is IMC/L3 clock FYI).

PS.
RAM capacity 18GB : http://valid.x86.fr/aiu6lt
tRFC = 100.

 

[biffzinker]

X5660 record is at 5.1...just sayin

I don't believe it.

Your North Bridge Clock sucks (which is IMC/L3 clock FYI).

Does it have to run in sync with the CPU cores clockspeed or is it on a separate clock domain?

 

[FilipM]

IT is a separate domain. I was getting over 28K memory copy / over 26K memory read with northbridge speed on 3200 mate. What QPI voltage you running for that speed?

 

[agent_x007]

Does it have to run in sync with the CPU cores clockspeed or is it on a separate clock domain?

On 45nm Bloomfield (I'm using), Uncore must be synced to DRAM Frequency with minimum 2:1 speed (over effective DRAM), on 32nm Westmere and LGA 1156 it's 3:2 minimum (if I remember correcly).

PS. My QPI/DRAM = 1,475V (it is C0/C1 revision 45nm chip and 3x2GB + 3x4GB RAM configuration after all )

 

[FilipM]

Holy mother of QPI voltage...may i also ask why you running CR2?

 

[agent_x007]

Because IMC working with 6 memory sticks I'm using "doesn't like" 1T with that kind of NB clock
1T for stable operation requires NB=3200/3333MHz, and from my tests that 400-500MHz faster NB was actually better in AIDA64 RAM tests vs. CR1T with slower NB clock.

Kinda high QPI, but it is Prime95 1h stable tho :
http://i.imgur.com/oIvyoPU.png

PS. In past (2009-2012) I used i7 920 D0 with NB Clock of 4GHz+ and ~1,52V QPI voltage (8GB 2GHz CL8 RAM)

 

[FilipM]

My TRFC by SPD is at 232 for 2100MHz on the ram. In the bios. I cant go over 160, and it gives me problems (cant boot)... hmm

 

[agent_x007]

Higher frequncy makes high tRFC a must.
You are clocked ~233MHz higher after all + memory I'm using isn't exactly generic as well (6x 2000MHz CL9, all 2000MHz CL9 CR1T capable @ 1,65V).

 

[FilipM]

Off to the bios I go

 

[FilipM]

Ok.

Cpu at 21x210
NB(QPI) at 3565
RAM at 2100 (10-12-11-1T)

25500 read / 21700 write / 26600 copy

I plug in the old ram (3x2GB), dont change anything and i get this:

29200 read / 22350 write / 34500 copy

Ps. tRFC on the "old" ram is 98 at 2100MHz.

Pps. Why the F*** is there such a big difference?

 

[agent_x007]

Meanwhile I tested my 12GB RAM @ 1T :

Pretty much this is the end of my chip (NB clock @ 4GHz is a "NO GO" zone for C0/C1 revision).
Above result was done with QPI/DRAM = 1,5V
tRFC = 143

Ps. tRFC on the "old" ram is 98 at 2100MHz.

Pps. Why the F*** is there such a big difference?

Because some DRAM chips simply suck.
Also 2GB/module is easier to drive than 4GB/module.

 

[FilipM]

Impressive scores indeed! No degradation issues at 1.5V QPI? Or is it just a quick test run

I will definately look into this issue, I might end up buying new ram completely.

 

[agent_x007]

Quick run, I don't use 1,5V QPI 24/7 (it's silly, since actual gain is small and possible problems quite big).
From what I used, ~1,45V should be fine within CPU life span of few years (at least on 45nm CPU's).
Much depends on what settings you will be running RAM/Uncore (you may not need that much QPI voltage for just a bit less MHz on Uncore).

Valid of my 3,93GHz OC : http://valid.x86.fr/b3db0g

PS. I found this table :

It's for Core i7 900 series (32nm).
Source : http://www.intel.com/content/www/us...op-processor-series-32nm-datasheet-vol-1.html

And this is for 45nm Xeon 5500 series :
http://www.intel.com/content/dam/www/public/us/en/documents/datasheets/xeon-5500-vol-1-datasheet.pdf

and Core i7 900 series (45nm) one :
http://www.intel.com/content/www/us...desktop-processor-series-datasheet-vol-1.html

Interesting stuff

But most interesting is Xeon 5600 series dataspec, max voltage table is on page 31 (had to use my google drive, since Intel doesn't have it [?]) :
https://drive.google.com/file/d/0BzaczsN8lPFQNEJlVWoyWDM1RE0/view?usp=sharing

 

[FilipM]

I have tried all the timing solutions, no luck...however I tried something else: I put the 6GB kit and 12GB kit together, so running 18GB in total.

2100 is a no-go now but at 1680 with 9-9-9-24-1T now I get:

26000 read / 22500 write / 29000 copy...Interestink, very interestinK

 

[agent_x007]

Well, since Uncore is kinda slow, very high DRAM speed is pointless (at least that's how I see it).
How's memory latency ?

 

[FilipM]

At 1680 its 53.0ns...2100 is at 51.6ns. Uncore is at 3570MHz with 1.35V on QPI

 

[FilipM]

I think I may have found why the RAM is slower: The old ram DIMMs were Dual Rank, the new ones are Single Rank. I think that explains it.

 

[agent_x007]

That might be it.
All my DIMMs are Dual Rank.