Xeon Owners Club

[Dinnercore]

XFX nForce 780i SLI is my board, which is identical to the EVGA 780i one. My PSU is an EVGA Supernova 1600 T2. The VRM of the board never got hot, actually the X5470 with it´s x10 multi is already producing much more heat / drawing more power at stock speed.

I have switched the CPU now and will continue on the X5470.

I did try hwinfo64 and I like that my SPP temp is showing up, but I don´t trust the core temps that hwinfo reports. They are much lower then realtemp or hwmonitor, to the point where hwinfo64 tells me that my cores idle at 20°C with deactivated halt-state (CPU always full speed, no idle throttle). That is 3°C lower then the water temp in my loop which physically can´t be true. Be careful with core temps from this tool.
I rather trust real temp and hwmonitor with the core temps. It was still good to see that I can check the SPP from my chipset, it narrows down the mystery about the 'Motherboard/TEMPIN' sensor to either VRM or MCP. Since it does get much hotter during small FFT torture tests vs. blend test I will now assume that this is my VRM temp.

Initially the stock speeds were unstable with the X5470 on my board, it did boot but I had to set voltages manually, especially GTL values since on auto the board did absolute nonsense and raised only a single lane to +65mV...
I overshot my Vcore a bit, but eh I will tune that down later. My initial settings that I grabbed from thin air:
1.343 Vcore (in Bios results in ~ 1.29V idle and ~1.256V in torture test)
1.2V FSB / VTT
1.9V Mem
1.3V SPP
1.5V MCP
1.2V HT

GTL offset values (I can only adjust positive): +40mV lane 1 / +40mV lane 2 / +40mV lane 3 / +40mV lane 4

This took me here:

1 error on Core 1 after 15 minutes blend test (the point it reaches small FFT).
After this I decided to really read into the forbidden black magic around GTL, calculating for my 1.2V VTT I thought I got pretty close. I found that for quad cores lane 1 is data for die 1, lane 2 is address for die 1, lane 3 is data for die 2 and lane 4 address for die 2. Address lanes are said to be less strained so they can be set a tiny bit lower. Since I saw in prime95 that die 1 threw the error I decided to adjust that by one step.
So I went for +45mV / +40mV / +40mV / +35mV and went straight for a small FFT only test:

I´d call this a success. With the background I learned in the past hour I know that these values are my current lower end border, I can now raise the values in tiny steps until it becomes unstable once again, which will be my upper border. The sweetspot should be right between the two. Once I get there I will try and raise FSB. Vcore should be enough for more frequency, I don´t want to go higher there. I think my VRM or whatever TMPIN0 is might not like that.
Looking back at the X5482 with the knowledge I gained today I could have done a lot better. I might have just hit a hole in the margin range, plus the 1.3 VTT introduced heat and noise that may not have been necessary until I got to 450 / 1800 FSB.

I learned a lot today and can´t wait to try and move on from here. This is so much more fun then OC with modern hardware!

 

[unclewebb]

the xeon 5400 series tjmax is 85c

There is no Intel documentation that supports that. If you have some, please post it. I know Intel released some BS numbers at a conference once but that is not a processor datasheet or proper documentation. It was just a way for Intel to claim that their Xeons run so much cooler. I call BS.

Have a look at the screenshot that @Dinnercore posted of his Xeon X5482. While Prime95 is running, the peak core temperature is 46°C. If TJ Max was only 85°C then his reported core temperature would only be 31°C. Of course that is not right.

Under the Temperatures category, CPUID HWMonitor reports a CPU temperature of 45°C which is coming from a completely separate CPU sensor. I would leave TJ Max set to 100°C.

 

[Edwired]

Here the link based on the tj max
https://forums.overclockers.co.uk/threads/official-tjmax-figures.17936945/

Got 4ghz stable here the settings the cpu just touch 50c on the hottest core 1

JumperFree Configuration Settings
AI Overclock tuner: manual
CPU Ratio Setting: auto
FSB Strap to North Bridge: auto
FSB Frequency: 445
PCI-E Frequency: 105
DRAM Frequency: 1069
DRAM CLK Skew on Channel A1: normal
DRAM CLK Skew on Channel A2: normal
DRAM CLK Skew on Channel B1: normal
DRAM CLK Skew on Channel B2: normal
DRAM Timing Control:

1st Information :

CAS# Latency: 5
DRAM RAS# to CAS# Delay: 5
DRAM RAS# Precharge: 5
DRAM RAS# Activate to Precharge: 15
RAS# to RAS# Delay : auto
Row Refresh Cycle Time: 70
Write Recovery Time: auto
Read to Precharge Time: auto

2nd Information :

READ to WRITE Delay (S/D): auto
Write to Read Delay (S): auto
WRITE to READ Delay (D): auto
READ to READ Delay (S): auto
READ to READ Delay (D): auto
WRITE to WRITE Delay (S): auto
WRITE to WRITE Delay (D): auto

3rd Information :

WRITE to PRE Delay: auto
READ to PRE Delay: auto
PRE to PRE Delay: auto
ALL PRE to ACT Delay: auto
ALL PRE to REF Delay: auto
DRAM Static Read Control: disabled
DRAM Read Training: disabled
MEM. OC Charger: auto
AI Clock Twister: auto
AI Transaction Booster: auto
Common Performance Level: 10
Pull-In of CHA PH1:
Pull-In of CHA PH2:
Pull-In of CHA PH3:
Pull-In of CHA PH4:
Pull-In of CHB PH1:
Pull-In of CHB PH2:
Pull-In of CHB PH3:
Pull-In of CHB PH4:

CPU Voltage: 1.25000 (max in os 1.240v)
CPU GTL Voltage Reference (0/2): 0.660
CPU GTL Voltage Reference (1/3): 0.700
CPU PLL Voltage: 1.52v
FSB Termination Voltage: 1.16v
DRAM Voltage: 2.10v
NB Voltage: 1.22v
NB GTL Reference: 0.650
SBridge Voltage: 1.20
PCIE SATA Voltage: 1.60v

Load Line Calibration: enabled
CPU Spread Spectrum: disabled
PCIE Spread Spectrum: disabled
CPU Clock Skew : 200ps
NB Clock Skew : 200ps

Advance CPU Settings
CPU Ratio Setting: auto
CPU VID: 1.113v
C1E Suppport: disabled
Max CPUID Value Limit: disabled
Intel® Virtualization Tech: disabled
Vanderpool Technology: disabled
CPU TM Function: disabled
Execute Disable Bit: disabled

 

[unclewebb]

Here the link based on the tj max
https://forums.overclockers.co.uk/threads/official-tjmax-figures.17936945/

The information posted in that article is not contained in any Intel Datasheet.

What really happened way back then is Intel released some extremely low ball TJ Max numbers. Myself and a couple of other programmers complained. I sent information to Gavin Steacy, the writer of that Tom's Hardware article, that showed that Intel's TJ Max numbers were not correct. He forwarded that information to Intel.

A couple of weeks went by, users in forums continued to complain, and then suddenly, Intel came back and released a completely new set of TJ Max numbers. New and improved. They actually blamed the secretary and said that she did not copy the information they gave her correctly. Total BS.

Some TJ Max values from their revised release are correct but unfortunately, the majority are not. A news release with some numbers from the back of a bar napkin are not the same as a proper datasheet. Users had complained for years that they needed this information and this was their best attempt at quieting the crowd.

I am probably one of the few people left that kept a copy of the ridiculous TJ Max numbers that Intel originally released at a trade show. I also have their updated document but both of these are by NDA so I cannot release any further info.

People can choose to believe whatever they want. Just scroll up to the X5470 pictures posted above. Intel's new and improved numbers are simply not possible.

After this debacle, Intel released their first gen Core i CPUs that finally had the correct TJ Max value written to each core. Any TJ Max numbers released before this are not worth the paper they were written on.

 

[Edwired]

Ok that strange then how is it that intel and amd providing the cpu to everyone and still get the tjmax wrong. My question is who is it that we suppose to trust for monitoring the temperature of the cpu as for example intel, amd or software as i debated about it before how come two different monitoring softwares gives two completely different readings.
as for the image i have checked the hwmonitor has the default tjmax as 100c and the hwinfo has the tjmax as 85c. as you can see in the core area shows the difference of 15c.
I dont think that hwmonitor even have the 12v reading correct as i have confirmed hwinfo is doing a better job as the 12v reading matches the bios reading in the asus premium. Have a look at this link http://lenry.atw.hu/tjmax.pdf to back up the post you mentioned about the xwon x5470 which suppose to be 85c not 100c tj max. Here the pdf stating what tjmax is suppose to be for the listed cpu

 

[unclewebb]

This problem has nothing to do with AMD. Intel created this problem because TJ Max information for their Core 2 Duo and similar Xeon desktop CPUs was never published in any Intel datasheet. For years software developers all had to guess at what the correct TJ Max value should be and for years, users in forums argued about this subject. Intel avoided answering this question until the very end of Core 2 Duo development. When the 1st Gen Core i was being released, then they went to their Intel Developers Forum (IDF) with their highly anticipated news release.

Here the pdf stating what tjmax is suppose to be for the listed cpu

10 years after that IDF news release, everyone has forgotten that the document that you posted is not the original document. There are two different versions of that PDF document. The first version that Intel released at the Intel Developers Forum (IDF) was full of errors because their secretary was not very good at copying and typing numbers. Users in forums continued to complain and then two weeks later, that is when Intel released a second version of that document. Neither document is 100% correct. The information in the PDF you posted does not come from any datasheet. Intel never published one.

how come two different monitoring softwares gives two completely different readings.

These Intel CPUs do not have a traditional temperature sensor. Instead of a temperature sensor that counts up as the CPU gets hotter, Intel CPUs use a sensor that counts down to 0 as the CPU gets hotter. When this sensor reaches 0, that is when thermal throttling will begin. TJ Max is the temperature the CPU will be when this sensor reads 0. All monitoring software uses this information and then tries to work backward to determine what the actual temperature is.

Reported Temperature = TJ Max - Digital Thermal Sensor (DTS) reading

Since TJ Max is not defined in any datasheet, for the Core 2 Duo desktop CPUs, monitoring software has always just guessed at this value. As an example, let's say that the DTS sensor is reporting a value of 25. If the first monitoring program thinks that TJ Max is 85°C and the second monitoring program thinks TJ Max is 100°C then there will be two completely different reported temperatures for the same CPU.

Reported Temperature = 85°C - 25
Reported Temperature = 60°C

Reported Temperature = 100°C - 25
Reported Temperature = 75°C

Which program is right? No one knows. It is possible that both programs are wrong. The real TJ Max might be 90°C, 95°C, 105°C or any number between 85°C and 105°C. That document you posted is basically saying that Intel does not have any idea. Instead of using the term TJ Max, Intel created a new term called TJ Target. In other words, TJ Max is somewhere around this TJ Target value but actual TJ Max is likely higher. Read that document closely. It was written by their marketing department and the lawyers made sure to cover their butts.

The values listed for TJ Target are not specifications

And that is the whole problem. If software does not know what TJ Max really is, it is impossible to determine what the core temperature really is. Intel never released TJ Max specifications for any of their Core 2 Duo desktop CPUs so all monitoring software is just guessing.

Based on that PDF document, I would not advise any X5470 owner to change their monitoring software's TJ Max value from 100°C to 85°C. You can endlessly argue about this but ultimately it really does no matter. As long as the DTS value is greater than 0, the CPU is running within spec and it will not be thermal throttling. That means it will be able to run at its full Intel rated speed. The core temperature reported by software is just a meaningless number.

 

[Edwired]

What you are saying that we have to run the cpu to the point where it become so hot that the distance to tjmax hit 0c it start to back off but that still dont give a real reading so how we suppose to know where the temperature is at. The part i cant get my head around is the tcase for example one cpu will have a max temp on tcase is 72.5c and another have 62c does that mean the cpu package rated at the stated spec or is that the overall cpu?

 

[lexluthermiester]

They actually blamed the secretary and said that she did not copy the information they gave her correctly. Total BS.

As someone who works in a science/technical environment, I can't tell you how many times something like that has happened. There's a good chance it actually happened that way.

 

[unclewebb]

so how we suppose to know where the temperature is at

Read the last line I posted.

The core temperature reported by software is just a meaningless number.

The only accurate thing that monitoring software can tell you is how far the CPU is away from thermal throttling. No software can tell you what the actual core temperature is with any degree of accuracy. With Core 2 Duo based CPUs, it is simply not possible so stop worrying about core temperatures.

The TCase rating is only intended for system builders. This temperature can only be measured accurately by drilling into the top of the heat spreader (IHS) and embedding a thermocouple at the geometric center of the heat spreader. The purpose of the TCase rating is so system builders can design a heatsink and fan and select case fans for airflow management purposes. If during testing, if the temperature measured at the center of the heat spreader remains below the Intel specified TCase temperature, in theory, during normal use, the peak CPU core temperature should rarely reach the thermal throttling temperature. In other words, the cooling is adequate so the CPU can run at its full rated speed during normal use.

Intel does not want individuals hacking up their new CPUs with a Dremel or similar tool to mount a temperature sensor. That is why they mounted temperature sensors on the hottest spots on the individual cores. Thermal throttling is controlled by these sensors. As long as the peak core temperature as measured by the DTS sensors is within spec (DTS > 0), the CPU is allowed to run at full speed. As soon as the DTS sensor reaches 0, thermal throttling begins to keep the peak CPU core temperature in check.

For individual users, the TCase rating is another meaningless number. It cannot be measured by users and it cannot be converted into a peak core temperature. The difference between the peak core temperature and the TCase temperature varies depending on the load and what type of CPU instructions are being executed. The difference is also going to vary significantly between Core 2 Duo and Core 2 Quad CPUs due to the layout of the cores.

 

[Dinnercore]

I already posted the result for overclocking my X5470 in my project log but thought I might share it here too.

The 780i chipset is not the best performer for FSB overclocking but if you fiddle with it long enough you can get something out of it and in my opinion it is not as bad as some people claim.

As you can see I got to 4.44GHz on 1.35V CPU. It had a hole spanning from ~427MHz FSB to ~433MHz FSB where it would not boot and get stuck in memory testing despite higher Vcore. I do not want to push much further for the sake of keeping the chipset within a reasonable temp.

If anyone else is on a 780i board or considers using one for the xeon mod, these are my settings. Important: I did pencil-mod the board to fix vdroop! The voltage I set for my OC will be much to low for a stock board with high vdroop.

Bios settings:
1.36125V on CPU
1.2V on FSB/VTT
2.0V on Memory
1.45V on SPP
1.5V on MCP
1.25V on HT-Link

GTLref Offsets: +75mV / +60mV / +75mV / +60mV

 

[dalekdukesboy]

Read the last line I posted.


The only accurate thing that monitoring software can tell you is how far the CPU is away from thermal throttling. No software can tell you what the actual core temperature is with any degree of accuracy. With Core 2 Duo based CPUs, it is simply not possible so stop worrying about core temperatures.

The TCase rating is only intended for system builders. This temperature can only be measured accurately by drilling into the top of the heat spreader (IHS) and embedding a thermocouple at the geometric center of the heat spreader. The purpose of the TCase rating is so system builders can design a heatsink and fan and select case fans for airflow management purposes. If during testing, if the temperature measured at the center of the heat spreader remains below the Intel specified TCase temperature, in theory, during normal use, the peak CPU core temperature should rarely reach the thermal throttling temperature. In other words, the cooling is adequate so the CPU can run at its full rated speed during normal use.

Intel does not want individuals hacking up their new CPUs with a Dremel or similar tool to mount a temperature sensor. That is why they mounted temperature sensors on the hottest spots on the individual cores. Thermal throttling is controlled by these sensors. As long as the peak core temperature as measured by the DTS sensors is within spec (DTS > 0), the CPU is allowed to run at full speed. As soon as the DTS sensor reaches 0, thermal throttling begins to keep the peak CPU core temperature in check.

For individual users, the TCase rating is another meaningless number. It cannot be measured by users and it cannot be converted into a peak core temperature. The difference between the peak core temperature and the TCase temperature varies depending on the load and what type of CPU instructions are being executed. The difference is also going to vary significantly between Core 2 Duo and Core 2 Quad CPUs due to the layout of the cores.

That's useful information, and I love your cpu temp utility by the way been using it for years. So if I hear you correctly my age old assumption on my cpu's including my current 8 core xeon is the reason the cores all have a fairly wide range of temps when at load is because they are really not all that accurate and only start to get "somewhat" accurate when put under load and particularly when closest to the set throttling temp.

 

[Zyll Goliat]

Can`t wait to see first performance tests!!!

 

[Vycyous]

I just bought a motherboard, CPU, RAM combo for less than US $200 on Ebay with the following:

Intel Xeon E5-1650
MSI X79A-GD45 Plus
G.SKILL RipjawsX DDR3-1600 CL9 8GB (2 x 4GB) x2

This will be the first Xeon processor I've owned or messed around with. I'm assuming there can't be much different from most of Intel's more modern processors (I consider roughly Sandy Bridge and newer to be "modern"). From what I can tell, the processor (which is supposed to be unlocked) is essentially the same as an i7-3930K. It would've been nice if it had v2 of that same processor (22nm) or, better yet, a Xeon E5-1680 v2, but I couldn't pass up what I think was a pretty good deal. The motherboard certainly wouldn't have been my first choice, but I suppose there are probably worse X79 motherboards. I'm excited to see what I can get out of the CPU and quad-channel memory.

Anyway, does anyone have any suggestions for this particular generation of Xeon processors or anything regarding the X79/LGA2011 platform?

 

[natr0n]

https://www.guru3d.com/news-story/d...dition-unofficial-mod-for-hedt-platforms.html

 

[lexluthermiester]

Anyway, does anyone have any suggestions for this particular generation of Xeon processors or anything regarding the X79/LGA2011 platform?

Looks like you've got a great setup, have fun!

https://www.guru3d.com/news-story/d...dition-unofficial-mod-for-hedt-platforms.html

Thanks for that! Gonna give it a try.

 

[phill]

Can`t wait to see first performance tests!!!

I love mine

http://imgur.com/iueBFVR

http://imgur.com/T9A9sxx

http://imgur.com/IvvbgDh

http://imgur.com/ZAGqC8M

Some of my most favourite setups of all time

 

[Deleted member 178884]

My clevo x7200 paired with a x5650 (running stock in this benchmark, highest it can go is 3ghz, *cough no vcore adjustment, thanks clevo cough*) got some nice 24gb ddr3 sodimm 1600mhz samsung memory modules for just £30 for all 24gb! (passed memtest fine etc and torture loads), all stock currently but I'll have to try overclocking

 

[Mr.Scott]

Mine. X5675's in this time.

 

[DR4G00N]

Mine. X5675's in this time.

Always wanted one of those but it'd probably just end up like my D5400XS, ie. in a box untouched for months.
Now if I had a pair of QX9775's then it'd probably be a different story.

 

[phill]

Mine. X5675's in this time.

View attachment 115904

I've been tempted with X5690's for ages but just never pulled the plug.. The X5675's where the next in line I'm using it as a crunching machine at the moment, 24 threads purring away at 2.67Ghz is ok for the moment

 

[Mr.Scott]

I've been tempted with X5690's for ages but just never pulled the plug.. The X5675's where the next in line I'm using it as a crunching machine at the moment, 24 threads purring away at 2.67Ghz is ok for the moment

 

[dalekdukesboy]

My latest attempts with single sided high speed RAM pushing both the RAM speed as well as FSB, didn't think I could get this far but almost at 140 fsb and close to 3000 mhz on memory.

 

[phill]

View attachment 115993

Your really not helping me with trying to save money are you??

 

[dalekdukesboy]

Still tinkering with multipliers/Fsb/memory. This is best combination I have found so far.

 

[Mr.Scott]

Your really not helping me with trying to save money are you??

You'll puke when I tell you I got that entire setup for *cough* 75 bucks.
Some people don't know crap about hardware. lol