Printable View
I dont have the time to run 20 loops, Its always been fine for me outside of stressing / benchmarking as long as I can get it stable for 15-30 minutes in these stress tests.
I stack prime, linx and super pi to load the system as much as possible in that amount of time, I dont get a single error anywhere else as long as it completes that.
I'd be willing to keep linx running if I could carry on using my browser in the mean time, but that just wont happen.
Just got two of these kits http://www.corsair.com/memory-by-pro...m3a1600c9.html
Anyone know if my GFlops are ok ;
http://img856.imageshack.us/img856/4861/24gblinx.jpg
That ram should work at 1T.
Ive been through two lots of 24 Gb triple channel kits now, and all that they both did was 1600 Mhz cas 9, or 1800 Mhz cas 10, so boring and slow compared to 2 Gb sticks.
bhavv, with 5120MB memory size in Linx and 4.5ghz, it should take less than an hour to do 20 loops. Do it when you're watching tv, or when you eat dinner or something. It's not long at all.Quote:
Originally Posted by bhavv
Also, what you're doing by running all those stress programs at once is overloading your cpu resources. Most notably this causes cache trashing and significant cache misses, which result in bus flushes. This lowers your overall system stress significantly, as the CPU wastes an enormous number of cycles flushing the bus over and over. So in short, running all of them together is WORSE stress testing than running a single one by itself.
At that problem size, your GFlops are fine. However, you're not fully stressing your CPU. Drop your problem size to 9216MB and re-test. You should be hitting 70+Gflops.Quote:
Originally Posted by zpaf
Quote:
Originally Posted by bhavv
I switch to 1T and run LinX with problem size 9216MB.Quote:
Originally Posted by zoson
Thanks.
http://img59.imageshack.us/img59/3029/9216s.jpg
Sorry, I said 9216MB as in 9216 Megabytes. Not problem size 9216.Quote:
Originally Posted by zpaf
Change the Memory (MiB) to 9216. Running 656MB problem size doesn't even fully text your first memory stick.
How many sticks are you running? and what is your uncore/IMC freq? 1.35v is honestly unnecessary, I only need that much with 6x4GB at 1700 with uncore at 3600, and I'm sure I can even drop that down a notch.Quote:
Originally Posted by bhavv
Not every chip/imc is the same cookies.Quote:
Originally Posted by Cookiesowns
I need 1.375v QPI to run 2x uncore(3668MHz) with 3x 4GB sticks and 3x 2GB sticks at 1834MHz cas 8-9-8-24-1T.
Yep, but also depending on board.Quote:
Originally Posted by zoson
What clock is you`re core on?
1,35-1,425 Qpi for 24Gb uncore X2+1, mem 1800 7-8-7. 1,65pll seems to help reduce the other voltages.
4,4 core + HT. Limit on air.
my rig details are in my signature.Quote:
Originally Posted by XII
my bios voltage settings:
1.4875v core
1.375v qpi
1.656v ddr
1.2525v pll
1.166v IOH and ICH
100% LLC
500KHz CPU PWM frequency
My memory kits are:
Mushkin Redline Ridgeback 3x2GB 1600MHz 6-8-6-24-72-1T(does 1600MHz 6-7-6-20-60-1T, 2000MHz 7-9-7-24-72-1T, and 2133MHz 9-10-9-27-110-1T)
Mushkin Blackline Ridgeback 3x4GB 2000MHz 10-10-10-27-160-1T (Does 1600MHz 7-8-7-24-82-1T and 2000MHz 9-10-9-27-110-1T)
Together the best I've got them to run is 1834MHz 8-9-8-24-110-1T.
Yeap You are right.Quote:
Originally Posted by zoson
My temps are 10c up ~
Thanks again.
http://img337.imageshack.us/img337/5179/24gblinxs.jpg
and a picture from aida64 with temps and volts.
http://img443.imageshack.us/img443/44/aida64.jpg
Guys,
I am on 1301 BIOS, and I have Core i7 920@4.25Ghz, should I update to 1502 ?
I know this post is useless now but I just had to point this out for the future.
Microcenter was selling the R3E for $191.99 and R3BE for $367.99. Did anyone know about this? Of course they are ALL sold out now. FMLLLL :mad:
I've made a rather significant discovery and thought I might share my findings. Seeing how many people have memory overclocking issues I'm hoping this may help.
I have recently began fiddling with DRAM Ref Voltages. I have actually always set them to +0mV without completely knowing what it is. After learning what they are, and after learning that R3E sets the proper DRAM REF Voltage on Auto (1/2 of VDIMM voltage), I was using the AUTO setting. After being unable to stabilize certain overclocks, and playing with and exhausting every voltage, memory timing and clock skew settings, I decided to give some thought to DRAM REF Voltages.
Okay so basically my R3E overvolts all parameters by about .012v-.013v. CPU PLL set at 1.788v will run around 1.8v. VDIMM at 1.616v will run around 1.628v. IOH at 1.126v will be around 1.139v. So I'm assuming DDR REF Voltage is being overvolted by about the same amount. So if my VDIMM is set at 1.616v and Ref voltages at Auto, then according to the way the R3E Bios sets Ref voltages, it will be 'half' of VDIMM voltage. In this case, it will theoretically be ~.808v.
If my board overvolts every parameter by about .012v, then my Ref voltages are being set at ~.821v. If Rev voltage should be half of VDIMM, then .814v is required for 1.628v VDIMM. I have to decrease the Auto Ref Voltage of .821v by .007v. DRAM REF Voltages have smaller increments when going negative (-2.5mV) than when going positive (+12.5mV). So setting Ref voltages to -2.5mV or -.0025v makes my hypothetical Ref voltage .8185v (.8185v*2 = 1.632v). Still too much for the voltage my memory runs at. So I set my Ref voltages to -7.5mV or .0075v, making my hypothetical Ref voltage .8135v (.8135v*2 = 1.627v). This made a day and night difference for me. More so than adjusting any other parameter in the BIOS.
In regards to VCore, with my Ref voltages at AUTO and running 4.187GHz, I would pass 5 loops of LinX at 1.25v (~83GFlops), 20 loops at 1.25625v (~82GFlops), then only 3 loops at 1.2625v (~82.8GFlops). Only after finally finding the right DRAM REF Voltage setting, it passed all 25 loops at 1.25v (~84GFlops). I might try to take VCore, VTT, and VDIMM even lower now :p:.
It's late and I'm tired so hopefully what I just said makes sense. I would suggest R3E owners to get a hold of a voltmeter and test how much their boards overvolt. Play with DRAM REF Voltages to see if you gain stability.
Here is a screenshot and my BIOS template.
Bios 1208 167x25 4.187GHz 2010MHz CL7 (12x) 3685MHz Uncore (22x)
VCore 1.214v-1.215v (real_load)
VTT 1.245v-1.246v (real_load)
VDIMM 1.628v-1.629v (real_load)
http://img854.imageshack.us/img854/9...5418ghzcl7.jpg
Code:
Target CPU Frequency 4175MHz
Target DRAM Frequency 2009MHz
LN2 Mode [Disabled]***
QPI Loadline Calibration [Disabled]***
AI Overclock Tuner [Manual]
OC From CPU level Up [Auto]
2D Performance Booster [Disabled]
CPU Ratio Setting [25.0]
CPU Turbo Power Limit [Disabled]
>CPU Configuration----------------------------
CPU Ratio Setting [25.0]
C1E Support [Disabled]
Hardware Prefetcher [Enabled]
Adjacent Cache Line Prefetcher [Enabled]
MPS and ACPI MADT ordering [Modern ordering]
Intel(R) Virtualiyation Tech [Enabled]
CPU TM Function [Disabled]
Execute Disable Bit [Enabled]
Intel(R) HT Technology [Enabled]
Active Processor Cores [All]
A20M [Disabled]
Intel(R) SpeedStep(TM) Tech [Disabled]
Intel(R) C-STATE Tech [Disabled]
----------------------------------------------
BCLK Frequency [167]
PCIE Frequency [100]
DRAM Frequency [DDR3-2009MHz]
UCLK Frequency [3683MHz]
QPI Link Data Rate [7366MT/s]
Memory Configuration Protect [Disabled]
>DRAM Timing Control--------------------------
CAS# Latency [ 7 DRAM Clock]
RAS# to CAS# Delay [ 7 DRAM Clock]
RAS# PRE Time [ 7 DRAM Clock]
RAS# ACT Time [20 DRAM Clock]
RAS# to RAS# Delay [Auto]
REF Cycle Time [60 DRAM Clock]
WRITE Recovery Time [Auto]
READ to PRE Time [Auto]
FOUR ACT WIN Time [Auto]
Back-To-BackCAS# Delay [Auto]
Timing Mode [1N]
Round Trip Latency on CHA [51]
Round Trip Latency on CHB [53]
Round Trip Latency on CHC [55]
WRITE To READ Delay(DD) [Auto]
WRITE To READ Delay(DR) [Auto]
WRITE To READ Delay(SR) [Auto]
READ To WRITE Delay(DD) [Auto]
READ To WRITE Delay(DR) [Auto]
READ To WRITE Delay(SR) [Auto]
READ To READ Delay(DD) [Auto]
READ To READ Delay(DR) [Auto]
READ To READ Delay(SR) [Auto]
WRITE To WRITE Delay(DD) [Auto]
WRITE To WRITE Delay(DR) [Auto]
WRITE To WRITE Delay(SR) [Auto]
----------------------------------------------
CPU Differential Amplitude [Auto]
CPU Clock Skew [Normal]
IOH Clock Skew [Normal]
------------ Extreme Engine Digi+ ------------
Digi+ PWR Mode [T-Balanced]
PWR Volt. Control [Auto]
Load-Line Calibration [50% Calibration]
CPU Voltege OCP [Enabled]
CPU PWM Frequency [500KHz]
Extreme OV [Disabled]
Extreme OC [Auto]
CPU Voltage [1.25000]
CPU PLL Voltage [1.78875]
QPI/DRAM Core Voltage [1.25625]
DRAM Bus Voltage [1.61650]
>DRAM REF Voltages----------------------------
DRAM DATA REF Voltage on CHA [-7.5mV]
DRAM CTRL REF Voltage on CHA [-7.5mV]
DRAM DATA REF Voltage on CHB [-7.5mV]
DRAM CTRL REF Voltage on CHB [-7.5mV]
DRAM DATA REF Voltage on CHC [-7.5mV]
DRAM CTRL REF Voltage on CHC [-7.5mV]
----------------------------------------------
IOH Voltage [1.12625]
IOH PCIE Voltage [1.51050]
ICH Voltage [1.11300]
ICH PCIE Voltage [1.51050]
---------- Spread Spectrum Control -----------
CPU Spread Spectrum [Disabled]
DRAM Spread Spectrum [Disabled]
***on-board jumper dependant
VERY interesting even if the "sweetspot" could be different from memory to memory. ;)
Hi, this is the ram I use:Quote:
Originally Posted by Cookiesowns
http://www.xtremesystems.org/forums/...-17000-2133MHz
They are crazy clockers at CL 7 in dual channel / P55 / 1.65v, but not so good on X58. But they are still better than any triple channel kit I've seen (in the UK at least).
At 158 BCLK and 951 Mhz (1902 DDR), it needs 1.65v DRAM and 1.35v QPI. If I set the QPI to 1.3375v, I get immediate BSODs within 5-10 minutes of stress testing.
I just managed to get it working stable with no errors at 190 BCLK and 952.5 Mhz (only 1 Mhz higher, but previously it wouldnt work, it seems to gotten burned in). This takes 1.3875v QPI, and 1.67v Dram, which I have read is fine to go just one setting above 1.65v on the DRAM on X58.
Loosening the timings above 7-8-7-24 does nothing to improve stability over 1900 Mhz. Going just 1 BCLK higher than 1905 Mhz ram = only 8 Gb ram detected and errors.
I managed to boot with 12 Gb ram detected and complete super PI 32m with 1925 Mhz cas 7 ram, but it took 1.7v DRAM which is too much extra for too little gain. 1900 Mhz is the max it will do at 1.65v on X58.
190 BCLK and 1905 Mhz 7-8-7-24 ram is the current best I can get everything to. I have the CPU running at stock volts and a 21x multi for 4 Ghz now for everyday use, plus the CO voltage and clock reducing thing active which reduces the multi down to 12x while I'm not playing games :)
My 6x4 Gb sticks ran fine at 1800 Mhz cas 10 with just 1.275v QPI, but those crappy modules arent comparable to these geil ultras. I've abandoned hope on looking for performance 4 Gb modules now, I just cant find any and have been through two lots of 6x4 Gb (Corsair Vengeance and G Skill ripjaws, both had identical lame ram chips - Cas 9 1600 or Cas 10 1800).
Also, I was consumed by madness, and bought a new SSD to replace my Raid 0 gaming drive:
http://img46.imageshack.us/img46/9716/dsc00269vc.jpg
Thats my graphics card budget gone oopsie, now I have to wait until 2013 and the 28 nm refresh gfx cards before more upgrades. I think I'll save towards a second 512 Gb M4 and new graphics cards in 18 months time.
Nice cooler you took in you`re madness. :ROTF:
What fans are you using on them?
Just the fans it came with, plus a Xigmatec red led fan as a third:
http://www.overclockers.co.uk/showpr...odid=FG-002-XG
I have 5 of those fans in total, they push a lot of air and are completely silent (2x intakes, 2x exhausts, 1 on the Phanteks cooler plus the two Phanteks fans).
Ok, after reading that thread about my ram again, I decided to copy the timings that others were using:
http://img545.imageshack.us/img545/6996/burnedin.png
The ram is beginning to get more and more 'burned in', prior to running it at 1866 Mhz Cas 7, it would NEVER run stable at this current setting, even with 1.4v QPI and more DRAM voltage!
Right now its going up to 1933 Mhz but requires 1.696v for that, this is too much for X58 right? I tested 1933 Mhz once and it completed super pi 32m with all 12 Gb detected, this was never possible before!
... My ram isnt going to die is it?
Using DMM, my R3BE gives, using 24Gb:
Vdimm set: 1,65625V vs. DMM Vdimm 1,644
Vdimm set: 1,68275V vs. DMM 1,673V (Didn`t give any gain, though)
Vcore set: 1,275V 100% LLC vs. DMM Vcore: 1,268~1,281V during LinX, or Prime95. Spike up is seen, when running LinX, at the end of a run, going to the next.
CPU PLL set to 1,65625 :D
12V Silverstone Idle: 12,04V vs Load: 11,99
Same variation on the other lines.
New core, first 4500 run:D
http://valid.canardpc.com/show_oc.php?id=2156358
Linx and 3DVantage/11 stable
Nice run, very good chip you got there! What Batch is it? So it was just a weak IMC on the last one huh? Very impressive 24GB run :up:
Batch I have to look up.:shrug:
IMC seems to be a little less than the Xeons, both can do +1800, both core`s limited to ~4,2. (On air that is)
The X990 has a better core, 4,5@1,3V-100%LLC (6C-6T)- DMM 1,32V - 4,3@1,3V-100%LLC (6C-12T)- DMM1,34V :rolleyes:
Still tuning, and hoping for more.......
That's not bad at least the Core is better on the 990X.
I would suggest using 50% LLC on CPU and no LLC on QPI. From my knowledge and experience it seems Intel architecture is designed to have fluctuations in voltage. You may be more successful with your overclock so give it a try. Also since you are running 24GB, I would mess with DRAM Ref voltages. I recently learned increasing it helps also, so try +12.5mV on all channels data & ctrl.
Every single one of my chips has done better with LLC at 100% and QPI LLC enabled.Quote:
Originally Posted by spoolindsm127
That's three i7 970's, two i7 980x's, one i7 920, four i7 930's and two i7 950's.
Sure if you are setting the same voltage in BIOS that makes sense as the 'real' voltage with no droop will be higher hence higher clocks. I think Spoolin was saying that if you compensate for that droop in BIOS but setting the voltage higher and have 50% LLC and Vtt LLC off is that the result is less electrical noise on each load line and the result of this cleaner power plane is often slightly higher clocks at the same real voltage.
My new 980x
bench in watter (T° room 20° )
http://img69.imageshack.us/img69/8690/56839016.png
http://img97.imageshack.us/img97/7219/4700.png