1366 x58 Xeon Enthusiast overclocks club

[Zoson]

I thought that 1.45v max for 24/7 applied to the old 45nm CPU's. I would have assumed our max for 32 and 22nm is substantially lower. I understand temperature is a big player. But even under ideal conditions, I have to question 1.45v on these chips.

Intel didn't change the vCore limit much on Gulftown/Westmere, only the QPI/UCLK.

Intel Core i7 900 ee and desktop processor series datasheet:
http://www.intel.com/content/dam/ww...-desktop-processor-series-datasheet-vol-1.pdf

Chapter 2 Electrical Specifications

Section 2.10 "Absolute Maximum and Minimum Ratings"
See Table 2-6 for proof that Vtt should not exceed 1.35v as ABSOLUTE MAXIMUM voltage. Period, end of story.

Now, how we arrive at 1.45v max spec VID:

Section 2.11.1 "DC Voltage and Current Specification"
See Table 2-7 to see 'Max' vid is 1.375v

Section 2.11.2: "Vcc Overshoot Specification"
See Table 2-16 for overshoot tolerance of 50mv brings max vCore to 1.425v

And basically through testing the rest of the community found that 1.45v was fine as long as you had a high end air cooler. Easily verifiable if you google something like: gulftown vid max
This is actually the search I put into google to re-find this whitepaper.

I've also been running a 1.45v 4.66GHz overclock that's 100% stable for years now.

Now, lets return to the point of don't exceed 1.35v on QPI/UCLK (called Vtt by this document, also making it clear that ASUS is mislabeling this value).
If you look at Table 2-10, you can see that maximum Vtt is actually dictated by current VID. So by raising your VID, you're also increasing your Vtt tolerance. Interestingly enough, Intel does not recommend a maximum voltage on Vtt other than 'absolute max'. They just supply this table that shows max Vtt is based on current being drawn by the device, and VID. However, the absolute max is still the absolute max. This is why you can run up to 1.35v QPI/UCLK(Vtt) but *NOT EXCEED IT AT ALL*.

 

[GonzoP]

give us some pics of that super air cooled setup....and for those reading if your getting bsods or crashes.....it be a smart choice to back down the voltage and the overclock or end up like Nate7311 with a blown cpu

Ask and you shall receive.

Before the pics the qualifier: I literally just took these five minutes ago and spent a grand total of 30 seconds taking all the shots. So no 'you're no Ansel Adams' comments.

So the first three pics is with flash on but side panel off for better pic and taken from my Galaxy Note 2 camera.

I moved my case over a few inches to show the floor vent. I mean literally this will be nothing exciting.

Next pic is with case where it normally resides.

This pic is a landscape of the case. You can see bottom of the Air 540 has several vents. I can only put my case over the one closest to the rear of the case [see the gray color as opposed to my tan carpet over the other two holes in the bottom.] Ideally I'd love to get this cool air right in front of my 2 140 intakes so the cool air blows across all motherboard. But as it is it is hitting my 7970 directly and my SSD [not that it runs hot] and some of the air goes around the 7970 and hits the heatsink. You can see the wall is just a few inches from the back of the case. Its the reality I'm dealing with.

These last two are with flash off [so quality is poor], side panel on and kinda showing the NXZT Hue I have inside.

Here's my hardware monitor temps.

My 3-4 coldest cores are downright frosty
This is obviously with SpeedStep enabled as witnesed by the 20W cpu power draw as opposed to 110w when its running 4.4Ghz. To give an idea. When I turn off SpeedStep and the cpu idles at 4.4Ghz my temp range is about 29ºC-40ºC from coldest to hottest core when idling. That's still decent but this extra cooling boost is excellent.
I still am plagued by the hottest core being over 10ºC hotter than the coldest one. I've done three separate heatsink re-seats and TIM re-applications and everytime its a variance like this.
The thermostat is set to 70ºF but clearly the air coming out of the vent is cooler than that. My SSD gets down to 69ºF, my 7970 idles at 73ºF and my two coldest cores both are below 70ºF. When running 3DMark my gpu never exceeds 55ºC.

Its not rocket science I was running my AC long before I got this case and did this move; now I'm just reaping what amounts basically to a 10ºC reduction benefit cross all system components.
I'm no cable management guru either but with this case it wasn't hard to get it looking rather nice and tidy with minimal effort

 

[GonzoP]

Remember kids, if you care about your cpu, do not exceed the following voltages:
vCore: 1.45v
QPI/vTT/UCLK: 1.35v
PLL: 1.9v
vDDR: 1.65v

I thought that 1.45v max for 24/7 applied to the old 45nm CPU's. I would have assumed our max for 32 and 22nm is substantially lower. I understand temperature is a big player. But even under ideal conditions, I have to question 1.45v on these chips.

Intel didn't change the vCore limit much on Gulftown/Westmere, only the QPI/UCLK.

Intel Core i7 900 ee and desktop processor series datasheet:
http://www.intel.com/content/dam/ww...-desktop-processor-series-datasheet-vol-1.pdf

Chapter 2 Electrical Specifications

Section 2.10 "Absolute Maximum and Minimum Ratings"
See Table 2-6 for proof that Vtt should not exceed 1.35v as ABSOLUTE MAXIMUM voltage. Period, end of story.

Now, how we arrive at 1.45v max spec VID:

Section 2.11.1 "DC Voltage and Current Specification"
See Table 2-7 to see 'Max' vid is 1.375v

Section 2.11.2: "Vcc Overshoot Specification"
See Table 2-16 for overshoot tolerance of 50mv brings max vCore to 1.425v

And basically through testing the rest of the community found that 1.45v was fine as long as you had a high end air cooler. Easily verifiable if you google something like: gulftown vid max
This is actually the search I put into google to re-find this whitepaper.

I've also been running a 1.45v 4.66GHz overclock that's 100% stable for years now.

Now, lets return to the point of don't exceed 1.35v on QPI/UCLK (called Vtt by this document, also making it clear that ASUS is mislabeling this value).
If you look at Table 2-10, you can see that maximum Vtt is actually dictated by current VID. So by raising your VID, you're also increasing your Vtt tolerance. Interestingly enough, Intel does not recommend a maximum voltage on Vtt other than 'absolute max'. They just supply this table that shows max Vtt is based on current being drawn by the device, and VID. However, the absolute max is still the absolute max. This is why you can run up to 1.35v QPI/UCLK(Vtt) but *NOT EXCEED IT AT ALL*.

Nice link there Zoson, much appreciated. I realize every cpu is different but if you've been at 4.66 for years on 1.45v I suppose that says a lot. I'll assume you also use SpeedStep so your cpu hasn't been per se running 4.66Ghz on 1.45v 24/7 for years?
I finally turned my on after over a week trying to dial this in right. No need for me to run hot and high voltage when its just sitting idle right.


Side note: yesterday after turning on SpeedStep and letting her idle for like 8 hours I came back and she wouldn't wake up. I know this is not an uncommon problem so I tinkered with a few things and we'll see what happens. I'd prefer to keep SpeepStep enabled since I'm saving 10ºC plus the electricity but if I cannot get it right I suppose I'll go without. Or maybe actually turn the sucker off when I walk away or especially when I go to work or something.

 

[GonzoP]

http://valid.x86.fr/6uyfuj
Again not 100% stable but nevertheless here it is. If I had the balls for more vcore I likely could stabilize it or maybe even run 4.8

 

[GonzoP]

We're all doing it wrong guys

In addition to this club I did read this thread before I bought mine and joined the club http://www.overclock.net/t/1489955/official-x58-xeon-club/1270

A few of these guys ignore the voltage warnings clearly. With my cooling enhancement I'd be tempted to try the 4.8 but beyond that I just don't know.

Looking at ebay the 5670 seems to be a better value than the 5660 as of today. MEI based out of Texas has the 5670 $160 OBO which means $155. Whereas they sell the 5660 for $145 which means $140. For $15 I'd clearly get the 5670 if I decide to get another Xeon to play with.

 

[Zoson]

Nice link there Zoson, much appreciated. I realize every cpu is different but if you've been at 4.66 for years on 1.45v I suppose that says a lot. I'll assume you also use SpeedStep so your cpu hasn't been per se running 4.66Ghz on 1.45v 24/7 for years?

My OC has EIST and Turbo enabled. So I get clocking up and down but no C1E, so no voltage transitions with the clock changes. It does net power savings, but no, my CPU is _always_ at 1.45v. I found that if I enabled voltage state transitions, I would get lockups attempting to come out of sleep, and random lockups at idle.

We're all doing it wrong guys

In addition to this club I did read this thread before I bought mine and joined the club http://www.overclock.net/t/1489955/official-x58-xeon-club/1270

A few of these guys ignore the voltage warnings clearly. With my cooling enhancement I'd be tempted to try the 4.8 but beyond that I just don't know.

Looking at ebay the 5670 seems to be a better value than the 5660 as of today. MEI based out of Texas has the 5670 $160 OBO which means $155. Whereas they sell the 5660 for $145 which means $140. For $15 I'd clearly get the 5670 if I decide to get another Xeon to play with.

As intel says in the doc, you can actually go up to about 1.55vCore as long as you keep things cool(absolute max VID value).
So, I see your unstable CPUID from OCN and raise a 100% stable 5GHz overclock:


Oh, and then there's this:
http://valid.canardpc.com/show_oc.php?id=2329925

These chips *can* survive short jaunts into high voltage/high temp land, but the key word here is 'short.' I've certainly abused the hell out of mine on more than one occasion and it came out unscathed.

 

[Rock&Roll]

Intel didn't change the vCore limit much on Gulftown/Westmere, only the QPI/UCLK.
..

Awesome knowledge drop. Thank you.

 

[GonzoP]

Great stuff Zoson. Yeah the 5.3 CPU-Z shots isn't mine its over from OCN.
Yeah I had a would not wake up from sleep yesterday myself. If I get another problem I might try your method to stabilize this coming out of sleep. I mean the first week and a half I had all the power saving stuff disabled and it was reliable coming out of sleep.
I'm somewhat reassured to see you have nearly the same range on your 990x from coldest to hottest core as I have on my Xeon. I guess its just how the hexa's are.
Pretty bad ass H20 set up as well.

 

[Zoson]

Great stuff Zoson. Yeah the 5.3 CPU-Z shots isn't mine its over from OCN.
Yeah I had a would not wake up from sleep yesterday myself. If I get another problem I might try your method to stabilize this coming out of sleep. I mean the first week and a half I had all the power saving stuff disabled and it was reliable coming out of sleep.
I'm somewhat reassured to see you have nearly the same range on your 990x from coldest to hottest core as I have on my Xeon. I guess its just how the hexa's are.
Pretty bad ass H20 set up as well.

It's a common thing for 1 or 2 of the cores on a cpu to be either abnormally hot or cold. So don't think anything of it.

And thanks! I'm actually really proud of the setup! It may not be the best looking, but I like to say it's 'engineered' since several pieces had ~1mm clearance and it took a lot of planning and measuring to get it all shoehorned in there!

And I just realized I totally had a brain fart moment, I have EIST and Turbo on, and C1E disabled. EIST gives dynamic multiplier, Turbo enables the 'turbo' multipliers(also always appears on for unlocked cpus using multipliers above stock), and C1E enables voltage transitions.

Sorry about that mixup. Corrected the other post also.

 

[GonzoP]

Remember kids, if you care about your cpu, do not exceed the following voltages:
vCore: 1.45v
QPI/vTT/UCLK: 1.35v
PLL: 1.9v
vDDR: 1.65v

I thought that 1.45v max for 24/7 applied to the old 45nm CPU's. I would have assumed our max for 32 and 22nm is substantially lower. I understand temperature is a big player. But even under ideal conditions, I have to question 1.45v on these chips.

Intel didn't change the vCore limit much on Gulftown/Westmere, only the QPI/UCLK.

Intel Core i7 900 ee and desktop processor series datasheet:
http://www.intel.com/content/dam/ww...-desktop-processor-series-datasheet-vol-1.pdf

Chapter 2 Electrical Specifications

Section 2.10 "Absolute Maximum and Minimum Ratings"
See Table 2-6 for proof that Vtt should not exceed 1.35v as ABSOLUTE MAXIMUM voltage. Period, end of story.

Now, how we arrive at 1.45v max spec VID:

Section 2.11.1 "DC Voltage and Current Specification"
See Table 2-7 to see 'Max' vid is 1.375v

Section 2.11.2: "Vcc Overshoot Specification"
See Table 2-16 for overshoot tolerance of 50mv brings max vCore to 1.425v

And basically through testing the rest of the community found that 1.45v was fine as long as you had a high end air cooler. Easily verifiable if you google something like: gulftown vid max
This is actually the search I put into google to re-find this whitepaper.

I've also been running a 1.45v 4.66GHz overclock that's 100% stable for years now.

Now, lets return to the point of don't exceed 1.35v on QPI/UCLK (called Vtt by this document, also making it clear that ASUS is mislabeling this value).
If you look at Table 2-10, you can see that maximum Vtt is actually dictated by current VID. So by raising your VID, you're also increasing your Vtt tolerance. Interestingly enough, Intel does not recommend a maximum voltage on Vtt other than 'absolute max'. They just supply this table that shows max Vtt is based on current being drawn by the device, and VID. However, the absolute max is still the absolute max. This is why you can run up to 1.35v QPI/UCLK(Vtt) but *NOT EXCEED IT AT ALL*.

It's a common thing for 1 or 2 of the cores on a cpu to be either abnormally hot or cold. So don't think anything of it.

And thanks! I'm actually really proud of the setup! It may not be the best looking, but I like to say it's 'engineered' since several pieces had ~1mm clearance and it took a lot of planning and measuring to get it all shoehorned in there!

And I just realized I totally had a brain fart moment, I have EIST and Turbo on, and C1E disabled. EIST gives dynamic multiplier, Turbo enables the 'turbo' multipliers(also always appears on for unlocked cpus using multipliers above stock), and C1E enables voltage transitions.

Sorry about that mixup. Corrected the other post also.

OK good to note. If I have another sleep issue I'll start with C1E. Thanks.

 

[Sturmangriff]

Well I got my modified board back from EVGA and started playing with my x5670. My initial OC is stable so now I have a good point to work from http://valid.x86.fr/2u75mi . I didn't touch my my i7 920 overclock for years so its been a bit of a refresher course on x58 overclocking for me. Fun stuff

 

[GonzoP]

Well I got my modified board back from EVGA and started playing with my x5670. My initial OC is stable so now I have a good point to work from http://valid.x86.fr/2u75mi . I didn't touch my my i7 920 overclock for years so its been a bit of a refresher course on x58 overclocking for me. Fun stuff

Same boat I'm in here. Thankful for the prime's, Deimos's and Zoson's out there to give us a reminder on what and how to do. It appears you have a good OC'er there. You're at 4.3 with less vcore than I needed. If you're willing to pump 1.45v I'd guess 4.8Ghz is possible for you. If you limit to 1.35v 4.4 seems like a sure thing based on how these babies scale

 

[Ultra-m-a-n]

Hmm all of these Bios options on the x58 platform are kind of boggling my mind...

anyways I just got my ex58-ud3r rev 1 working with the x5650 that I picked up!

 

[GonzoP]

Hmm all of these Bios options on the x58 platform are kind of boggling my mind...

anyways I just got my ex58-ud3r rev 1 working with the x5650 that I picked up!

Honestly there are only 5-10 line items in your BIOS you really need to concern yourself with. Once you get familiar with your BIOS and your mobo you'll almost assuredly join the rest of us well over 4Ghz on that Xeon. Welcome to the club

 

[GonzoP]

Hey Guys 2 questions:
So in Hardware Monitor I'm clearly using less power at idle. I go down to like 20w from 110w at full song. CPUZ shows the clock speed drop to 2400 [i.e. the multi drops to 12 from 22 for me]. But the vcore never changes in CPUZ. Is this just an error in CPUZ reporting. I cannot imagine merely reducing clockspeed without reducing voltage would drop temps and power that much. Or would it?

Also ever since turning on SpeedStep et al I cannot complete a 3DMark run without the program kicking me out for some error. All other benchmarks and real games run fine. Thoughts?

 

[Zoson]

Hey Guys 2 questions:
So in Hardware Monitor I'm clearly using less power at idle. I go down to like 20w from 110w at full song. CPUZ shows the clock speed drop to 2400 [i.e. the multi drops to 12 from 22 for me]. But the vcore never changes in CPUZ. Is this just an error in CPUZ reporting. I cannot imagine merely reducing clockspeed without reducing voltage would drop temps and power that much. Or would it?

Also ever since turning on SpeedStep et al I cannot complete a 3DMark run without the program kicking me out for some error. All other benchmarks and real games run fine. Thoughts?

Remember, all these cpus feature power gating separately from clock and vid control, so even if your vid is pinned at the highest level, your cpu will still scale very well as far as consumption is concerned by simply turning off parts of the cpu not in use. C1E controls the voltage transitions, so i'd expect it to work if you have that enabled.

And yeah my cpu scales from 1866 (12*155.5) to 4665 (30*155.5). Interestingly it seems to sit at 4.5GHz most of the time (29*155.5) though. All of this is at 1.45v (1.4675v setting in bios on my R3E) with Full LLC. My coolest core idles all the way down to 19C!

 

[Deimos]

You also have to remember that wattage is a function of both voltage AND current. Electricity is interesting in that a circuit will only draw the current it needs to operate. load = transistor switching, no load means no switching = less current = less wattage.

 

[ZorchThatCPU]

Looking at ebay the 5670 seems to be a better value than the 5660 as of today. MEI based out of Texas has the 5670 $160 OBO which means $155. Whereas they sell the 5660 for $145 which means $140. For $15 I'd clearly get the 5670 if I decide to get another Xeon to play with.

What for? You must have amazing watercooling, or live in Antarctica. An X5650 (of which I have two) should be able to max out any reasonable cooling well before reaching the limits of the CPU or mobo, even with HyperThreading off. Anything higher than an X5650 is purely wasting money unless you're maybe going to run it daily under phase-change.

One of my X5650s is actually running almost stock, since it's basically a media server.

But the other one is tweaked up a bit, idling at 4.6GHz and staying there unless I need more than two cores, at which point it drops down to 4.4GHz. It's kind of perfect, and I finally found a use for allowing VDroop to occur.

That's with HT off since ambients are terrible here in a third-floor room during the summer. Basically when autumn gets here, it'll be the same settings except with HT on. Either way, I'll be running these settings without even exceeding Intel's recommended VCore limit of 1.4, so that's nice.

But I'm only at 200 BClk (x22 or x23, depending), so it's temps, not board or CPU that are holding me back at all. This is under average-quality custom water, better than a Corsair H-series cooler but nothing special among custom kits.

Anyway, I'd strongly recommend the X5650 as the fastest chip needed to max out almost anyone's cooling. The whole point of going back to a five-year-old platform is because you can get awesome performance for dirt cheap. When you start paying the same prices that new CPUs cost (plus more than likely overpaying for an X58 board right now), it doesn't make much sense, unless it's just for amusement value. Both my X5650s were $85 apiece, and at that price point, I'm thrilled (even though they're even cheaper now). Running a CPU that cost twice as much or more and then finding that I can't even run it any faster wouldn't leave me nearly as happy.

Anyway, that's my two cents.

Triple XXX: Gigabyte GA-X58A-UD3R, Xeon X5650 (Westmere-EP / B1) @ 4.6GHz daily, 3 x 4GB Crucial Ballistix Tactical Tracer DDR3-2000 (8-8-8-24 2T)
Mainframe: Asus Sabertooth X58, Xeon X5650 (Westmere-EP / B1) @ 3.0GHz daily, 3 x 4GB Crucial Ballistix Tactical Tracer DDR3-1600 (8-8-8-24 2T)

 

[Rock&Roll]

For a minute I was thinking "4.6Ghz on a 20 multi; wow the bclk". Then I realized...Turbo. I really don't know what the general consensus is. But I never felt that my turbo clocks were worth advertising. Maybe I'm wrong in that way of thinking.

 

[ZorchThatCPU]

For a minute I was thinking "4.6Ghz on a 20 multi; wow the bclk". Then I realized...Turbo. I really don't know what the general consensus is. But I never felt that my turbo clocks were worth advertising. Maybe I'm wrong in that way of thinking.

Depends on how you use it. In my case, it runs at 4.6 (x23) and stays at 4.6 unless it needs more than two cores (four threads if HT is on). If it needs more cores/threads than that, it will drop down to 4.4 (x22). So it's nothing like Turbo in the sense it was meant to be used. It never drops down to x20 under any circumstances, never anything below x22. 4.4 is the floor, but almost all the time it's 4.6. I could run 4.6 100% of the time perfectly fine if my cooling and/or ambients were more favorable.

I had been running this at 4.4 all the time, with that never changing, but only because that's all my cooling can handle this time of year with all six cores under extended load. (I've gone as high as 5013MHz, but I can't remotely cool that for daily usage.) Then I realized, why not take advantage of the x23 multiplier when all cores aren't loaded (which is almost all the time)?

To get this result, you must have Turbo, SpeedStep, and all C-States enabled. (On ASUS boards, you must have the Multiplier set to Auto.) Then in the operating system, go to Power settings in the Control Panel and set System Cooling Policy to Passive and Minimum Processor State to 100%. Bam, my daily speed is now 200 x 23 for 4.6GHz all the time, unless I need to do some transcoding or something.

If I were Turbo-ing up to 4.6 I wouldn't mention it either. But that's not the situation.

 

[Cecil]

Depends on how you use it. In my case, it runs at 4.6 (x23) and stays at 4.6 unless it needs more than two cores (four threads if HT is on). If it needs more cores/threads than that, it will drop down to 4.4 (x22). So it's nothing like Turbo in the sense it was meant to be used. It never drops down to x20 under any circumstances, never anything below x22. 4.4 is the floor, but almost all the time it's 4.6. I could run 4.6 100% of the time perfectly fine if my cooling and/or ambients were more favorable.

I had been running this at 4.4 all the time, with that never changing, but only because that's all my cooling can handle this time of year with all six cores under extended load. (I've gone as high as 5013MHz, but I can't remotely cool that for daily usage.) Then I realized, why not take advantage of the x23 multiplier when all cores aren't loaded (which is almost all the time)?

To get this result, you must have Turbo, SpeedStep, and all C-States enabled. (On ASUS boards, you must have the Multiplier set to Auto.) Then in the operating system, go to Power settings in the Control Panel and set System Cooling Policy to Passive and Minimum Processor State to 100%. Bam, my daily speed is now 200 x 23 for 4.6GHz all the time, unless I need to do some transcoding or something.

If I were Turbo-ing up to 4.6 I wouldn't mention it either. But that's not the situation.

Heyhey, didnt know you were in this thread.



Going to be taking my M500 and a GTX 660 out of my 2500K rig, and use in Z's above this Friday, to compare some games and SATA benches. If someone has a request let me know.

 

[ZorchThatCPU]

Heyhey, didnt know you were in this thread.

Yeah, after you mentioned on the phone that people in this thread were throwing away money on the most expensive Westmere-EP chips when their results for half the CPU cost or less would most likely be the same, I figured I owed it to the community to stop by and offer a personal testimonial about the X5650 (for $75 or less) being plenty to max out almost everyone's cooling. I don't even know if the chips that are faster at stock could do better under better cooling, really. Most good X58 boards can do close to 219 BClk or so daily, and with x22 being the default multiplier for a Xeon X5650 on fully-WestmereEP-compatible desktop boards with full load on all cores (and x23 being the default multiplier all of the rest of the time), that comes out to over 5.0GHz for tasks that require 2 cores/4 threads or less and 4.8GHz daily for full load on all cores, if set up correctly. Anyone expecting to get a higher daily speed out of 1366, especially from a six-core, twelve-thread CPU may not have realistic expectations. A Xeon X5650 is unlikely to hold anyone back in any way.

Long story short here, people? Don't work harder, work smarter.

Going to be taking my M500 and a GTX 660 out of my 2500K rig, and use in Z's above this Friday, to compare some games and SATA benches. If someone has a request let me know.

Yep, hit him up.

 

[Zoson]

Just going to say don't listen to this Zorch dude, there's so much wrong with what he posted that I'm not even going to bother quoting his text vomit. He's supplied exactly 0 evidence to show his OC is anywhere near stable(it's probably insanely unstable and he just doesn't use his system much).

Bottom line, not all CPUs are created equal and Intel's binning is REAL. A higher stock clocked cpu will almost always reach higher overclocked speeds at lower voltages than a lower binned chip. Why? Because the TDP doesn't change. Higher stock cpus have to maintain the same TDP at a higher clock. That means they _MUST_ have less gate leakage, which, in turn, means the transistors are getting more bias placed on the gate, instead of losing the electricity to heat. That, in turn, directly translates to a faster switching speed.

990x reaches 4.8GHz HT on at 1.5v:


980x reaches 4.66GHz HT on at 1.58v:


970 reaches 4.4GHz HT on at 1.495v:


In the above shots you can see a very clear scaling difference between the different binned chips, and these Xeons are EXACTLY the same silicon. It's not just voltage, too. It's the amount of heat generated by each chip at the different speeds/voltages. the 990x clearly produces significantly less heat even when at higher voltage than the 970.

Long story short, this guys two cents are classic, depreciated pesos.

 

[Cecil]

How are you going to show one example of each after saying how each one is different?

I've had over 15 different 1366 CPUs across a dozen boards. 920 D0s clock as high or higher then most 930/940/950etc. For 1366 batch actually mattered. Also, when binning same CPU, lower stock VID means higher leakage, which is good for going cold (LN2) but bad for daily as it will draw more power and cause more heat when OCed.

A low VID sample is going to be a high leakage chip (good for benching with extreme cooling, high VID is better for daily). VID is determined by TDP. If a CPU has a TDP of 125W, when they are setting the VID, they are setting it so it does not go over the TDP. If the range for a certain CPU is 1.2-1.5, and a CPU is going past the 125W TDP, they will give it a low VID. So a 1.25v CPU would be a low VID CPU, but a high leakage CPU. Now you take that 1.25VID CPU, and set it to 1.4V, it will now be drawing more power then a high VID CPU that started at 1.4V.

The multi isn't the limitation with these CPUs. So there is no need to spend $150 for an extra multi bump for a platform that is supposed to offer bang for buck when you get the same thing for $70. It was the same when people would have 920s running faster then 940s and 965s.

Running a 5650 at 200x22 is going to be 4400. And 200BCLK is easy on pretty much any board. Once you get around that point cooling is what will be the limiting factor. If you have a super high end water set up, then pushing another 10-15BCLK should be fine putting it at 4.6-4.7+. Better boards will do 220BLCK resulting in 4840Mhz. I think anyone expecting a stable speed on this kind of platform of 5Ghz using safe voltage, just needs to realize that is extremely unlikely to happen.

But then this goes back to the whole point of this platform. Price. It doesnt make sense to spend a ton on a high end water setup to run a 6 year old budget platform. Those people will be on 2011/1150. People going with this kind of setup are on a budget, and will likely have air or AIO water cooling. So again, a multi bump isnt going to help anything here. Spending $150 on the CPU and $200 on the board is almost what used 2011 3930K setups are going for.

 

[ZorchThatCPU]

Just going to say don't listen to this Zorch dude.

I'm just going to say don't listen to this yayhoo Zoson, since he doesn't know me and clearly doesn't know what he's talking about.

He's supplied exactly 0 evidence to show his OC is anywhere near stable(it's probably insanely unstable and he just doesn't use his system much).

I wasn't aware that explaining to people like you how to do this required any evidence. However, I was going to include some screenies, but I do not have the option to attach photos, so I didn't. But you can certainly check it for yourself and discover that I'm exactly right.

Bottom line, not all CPUs are created equal and Intel's binning is REAL.

Of course binning is real. But these are all Xeons, genius, meaning that they are already binned. They are all the cream of the crop.

A higher stock clocked cpu will almost always reach higher overclocked speeds at lower voltages than a lower binned chip.

Only if you're talking about a very low-binned chip versus a higher one. My Xeon X5650 is capable of maxing out the potential of this architecture, so having a higher stock clock would be irrelevant. The best these chips can do is the best these chips can do. There is no such thing as infinite scaling. If they released a super-duper-binned Gulftown chip that could run 4.8GHz stock, it would still hit the same ceiling as any other Gulftown chip. Use your head, it's not complicated.

Long story short, you're just embarrassed that all of you have been blundering around between the "blunt force trauma" approach and the "just throw more money at it" approach, neither of which is an intelligent way to approach this. Don't worry, we all have to learn, and now you know.

 

[GonzoP]

Yeah, after you mentioned on the phone that people in this thread were throwing away money on the most expensive Westmere-EP chips when their results for half the CPU cost or less would most likely be the same, I figured I owed it to the community to stop by and offer a personal testimonial about the X5650 (for $75 or less) being plenty to max out almost everyone's cooling. I don't even know if the chips that are faster at stock could do better under better cooling, really. Most good X58 boards can do close to 219 BClk or so daily, and with x22 being the default multiplier for a Xeon X5650 on fully-WestmereEP-compatible desktop boards with full load on all cores (and x23 being the default multiplier all of the rest of the time), that comes out to over 5.0GHz for tasks that require 2 cores/4 threads or less and 4.8GHz daily for full load on all cores, if set up correctly. Anyone expecting to get a higher daily speed out of 1366, especially from a six-core, twelve-thread CPU may not have realistic expectations. A Xeon X5650 is unlikely to hold anyone back in any way.

Long story short here, people? Don't work harder, work smarter.


Yep, hit him up.

Yeah I suppose you're right. The reason I got my 5650 was because it was $70 on ebay and that's all I spent. No new ram, no new mobo. $70 and I have a 4.4Ghz Hexacore on par with a 3960 or a 4770 at a fraction the cost.
My ram I've had for years is ddr3 1333 so when I run the BCLK at 200 I'm at 1600. I tested my 5650 up to 214x22 [4.7Ghz] and that's about where her ceiling was. Whether I'm hitting the BCLK ceiling or my 1333 ram's limit at 1700+ or my fear of feeding more than 1.45v vcore it almost doesn't matter.
You're right I'm not going to spend more cash on new ram which might net me a few more Mhz. And yes I suppose what are the odds I'd find a 5670 which could do 214x24 [5.1Ghz] on 1.45v vcore? Not bloody likely.

I guess I'll call it good enough at 200x22 across six cores with HT $70 and I'm out. If anything I probably can move up to 210x22 and be under any BLCK, ram or voltage limits I've self imposed. 214x22 wasn't purely stable across the boards.

 

[GonzoP]

I'm just going to say don't listen to this yayhoo Zoson, since he doesn't know me and clearly doesn't know what he's talking about.


I wasn't aware that explaining to people like you how to do this required any evidence. However, I was going to include some screenies, but I do not have the option to attach photos, so I didn't. But you can certainly check it for yourself and discover that I'm exactly right.


Of course binning is real. But these are all Xeons, genius, meaning that they are already binned. They are all the cream of the crop.


Only if you're talking about a very low-binned chip versus a higher one. My Xeon X5650 is capable of maxing out the potential of this architecture, so having a higher stock clock would be irrelevant. The best these chips can do is the best these chips can do. There is no such thing as infinite scaling. If they released a super-duper-binned Gulftown chip that could run 4.8GHz stock, it would still hit the same ceiling as any other Gulftown chip. Use your head, it's not complicated.

Long story short, you're just embarrassed that all of you have been blundering around between the "blunt force trauma" approach and the "just throw more money at it" approach, neither of which is an intelligent way to approach this. Don't worry, we all have to learn, and now you know.

Yeah like I said above. I'm pretty much in your camp.
If you don't mind me asking how are you cooling your 5650? And what voltage are you pumping thru that baby for your 4.6 daily?
Myself I run my 5650 at 1.375v BIOS 1.36v CPUZ for 4.4 and 1.425v BIOS 1.416v CPUZ for 4.6.
I can run some things at 1.45v BIOS 1.44v CPUZ at 4.7 but it was getting wonky there. Didn't want to take my vcore higher than that either. I'm back down at 200x22 on the 1.375BIOS 1.36CPUZ as it seemed like the best compromise. Since like you said I'm in this for value and didn't wanna fry my cpu and have to get another one.
And yet Zoson has been running his 990x [essentially a Xeon] for years at 1.45v and I have to assume your own vcore is in the 1.4v neighborhood for that 4.6?

Just trying to figure out a best practice considering my expecting to use this cpu-mobo-ram configuration until something genuinely worthwhile is on my upgrade map i.e. probably several years.

 

[Deimos]

214x22 wasn't purely stable across the boards.

I couldn't even hit 209 BCLK with everything else at stock and now with the LSI Raid card installed my max BCLK is more like 170.

I'm getting rid of this thing, the difference in speed isn't worth the hassle, constantly having to deal with boot issues, low BCLK, slow boot speeds, 8 lanes on the video card etc.

Shit, just loaded up CPUz and the video card is running on a single lane

 

[ZorchThatCPU]

GonzoP, yep, it sounds like we're getting very similar results. For 4.4 flat, I use 1.375 in BIOS, which comes out to 1.36 by CPU-Z. For 4.6 on 2 cores/4 threads or less, dropping down to 4.4 when all cores are loaded, I use 1.415 in BIOS (1.392 in OS) and turn off LLC so that when it drops down to 4.4 with all cores loaded, it's back around 1.36 again.

As for cooling, I use a 2.120 rad with built-in res, so nothing special. Ambients in my computer room during the summer are terrible, starting at around 90F and ranging well above 100F depending on the weather. I won't get a chance to see this rig really shine until fall/winter.

Deimos, these chips should definitely be able to do 214 x 22 stably. The only two questions are A) can you cool your CPU sufficiently at that speed and B) are you comfortable throwing enough QPI/"VTT" at it to make it stick and use those settings daily.

On a side note, in regard to overbuying Xeons, I ran into the same thing playing with the 771-to-775 mod and Harpertown Xeons. I bought an X5450 and loved it, so I figured an X5470 would be way more awesome. It wasn't. Both could do about 4.6-ish benchable and around 4.5 daily-stable. The X5470 was not better in any way, not even better temps, but it cost more than twice as much. The X5470 got sold and I kept the X5450, a much wiser purchase.

 

[Deleted member 12106]

I can't seem to find it but had this chip run occt for 24hrs at 4.6.

 

[ZorchThatCPU]

And yet Zoson has been running his 990x [essentially a Xeon] for years at 1.45v.

Ah, I missed this earlier reading on my phone. That explains why he's so bitter about this stuff, and why he's desperately trying to convince himself that he was justified to drop a grand or whatever on his CPU while the rest of us got the exact same performance for around $70. His must be better because it was more expensive, and that never steers you wrong.

I have to assume your own vcore is in the 1.4v neighborhood for that 4.6?

Following up since I didn't respond in detail about this earlier... I love that my CPU is willing to do 4.4-4.6 without ever exceeding 1.4v. But I am not nearly as concerned about VCore (other than in regard to temps, which balloon over 1.4) as I am about QPI/"VTT". I wouldn't want to go much over your stated personal VCore limit of 1.45 unless my cooling could genuinely handle that. But within reason, my position is that if you can keep it cool, VCore limits are malleable to an extent. QPI/"VTT" voltage is what scares me. That'll kill your chip dead out of nowhere without any warning, whereas degradation from overvolting the VCore is more likely to occur gradually over an extended period of time, so that you'd probably have time to notice a loss in stability and adjust things before the chip would actually die. Unless irresponsible temps are involved.

Just trying to figure out a best practice considering my expecting to use this cpu-mobo-ram configuration until something genuinely worthwhile is on my upgrade map i.e. probably several years.

That's exactly how I feel about it. It's exciting the memory speeds that Haswell can hit, but that's just shy of being able to entice any serious interest on my part. I'm sure something mind-blowingly cool will be released by Intel soon, but for now a smoking fast 1366 rig (or two) does the trick just fine.

Anyway wrapping up, I actually forgot to thank Zoson for his thoughtful compliment.

Long story short, this guys two cents are classic, depreciated pesos.

Which might seem sort of dickish at first, but it isn't when you actually think about it. Pesos, while undeniably deflated, are worth 7.5 cents each. I was offering two cents, but he said they're really pesos, not cents. So what Zoson was actually telling me is that he thinks my opinion is seven and a half times more valuable than I was giving myself credit for. I think I've made a new friend.

 

[GonzoP]

Zorch what is your personal qpi/vtt ceiling? 1.35v?
Myself I'm at 1.3v for 4.4Ghz and was still at 1.3v for 4.7. Its likely I need more than 1.3v to stabilize the OC at 4.7 just not sure I wanna push much more

 

[ZorchThatCPU]

Zorch what is your personal qpi/vtt ceiling? 1.35v?

I guess I would say 1.38-1.4 would be the absolute max if I had unusually remarkable settings I could run daily. Like if I could run 5.0GHz daily, even with HT off, I would definitely risk frying that CPU to run 1.4v QPI/"VTT". Almost anything short of that, I'm at my max right now, with 1.36v. I am quite sure that as seasons change and ambients drastically improve here, I can cut way down on this voltage. But at the moment, ambients are so unfavorable in this room that I am deliberately substituting other voltages where I might normally use VCore, to get the same result. Just because the water in my relatively average loop starts off warm and gets hotter right now.

I'm trying to go the whole summer without putting in my window air conditioner, even on the third floor during the summer in Ohio, to offset the electrical cost of some cryptocurrency mining I recently did. In case anybody is interested in the backstory why I don't just seek more favorable ambients. It's early August and still going strong, so I hope I didn't just jinx it.

Myself I'm at 1.3v for 4.4Ghz and was still at 1.3v for 4.7. Its likely I need more than 1.3v to stabilize the OC at 4.7 just not sure I wanna push much more

Hey, if you don't need more, I'd say enjoy not needing more. I think this is helped by your RAM limitation, actually. I'm pushing the envelope somewhat by running 3 x 4GB at DDR3-2000 and 8-8-8 timings. Other voltages need goosed on occasion to maintain that, as I'm sure you know.

In the vast majority of situations, 200 BClk is a very easy goal for both of us to achieve, and it shouldn't need much extra QPI/"VTT" voltage if you don't get greedy with your memory settings, as I have sort of done here.

Edit to add: What sort of UnCore speed are you running at 4.4?

 

[GonzoP]

Yeah honestly I've never messed with uncore so its the standard 2x memory.
I'm doing a trial run where I took vcore to 1.48 and got 4.8Ghz 219x22. I definitely have to disable speedstep and c1e for stability at this speed though. QPI is still just 1.32 though. I'll say this, perhaps its the memory controller or just the superiority of the Xeon over my old i7. But I never came close to 220BLCK on my i7. It was wonky past 210. My problem now is I'm already at 1.48v for a 219 BCLK. Few x58's can do more than 220-225. I mean now I'm just doing things for the e-peen wars.
I am pretty surprised, pleasantly of course that my ddr3 1333 is maintaining its 9-9-9-24 timings here at 1750. vdimm is 1.65v
I just know I'd need to crack 1.5v for 4.9Ghz and that's if the ram and BCLK will hold up. Yeah I can drop the ram multi to 6 instead of 8 but I'd need 223BLCK and don't know if that's in the cards for me.
But no I'm just messing around right now. Probably will still drop back to 4.4Ghz maybe 4.5 as its just better for my long term peace of mind.

 

[GonzoP]

So this is worthless

But it does make me think with enough tinkering I might just be able to get 4.7Ghz stable. Anyway, back down to 4.4. I mean really what will 4.6 or 4.7 do for me I can't do at 4.4? Nothing. I am open to trying 4.5 as my 24/7 since I can keep my vcore below 1.4 or more accurately right at it and my QPI at 1.3 where I like it. Temps at 4.5 only go up about 2ºC from my 4.4 settings.

 

[ssnyder28]

Wow its pretty amazing that you guys can get those x5650s to such a high clockspeed.

 

[ZorchThatCPU]

GonzoP, you're crushing it! These chips are too fun.

Your results have me drooling waiting for autumn to get here. What kind of load temps are you getting? I see you're leaving HT on, so I can only assume you're comfortable with temps so far.

Have you messed around with disabling any cores as well as HT, to keep temps extremely friendly? When I was feeling out this chip, I had all sorts of weird combinations stored in my BIOS, but when I went a little too far, this GA-X58A-UD3R exercised its nasty habit of sometimes reflashing the BIOS from the backup BIOS chip, erasing all of my stored profiles. I'll recreate them all eventually, but one that was fun to play around with was running this as a 5.0GHz dual-core. Temps were actually fantastic, and I believe it wanted around 1.5v to do it, but it was stable enough to break bricks on. Could be a useful gaming profile, not that I'm much of a gamer. Easy as pie too, because with only two cores active, the highest Turbo multiplier becomes the full-time default multiplier (x23 on my X5650), so I only needed 218 BClk.

Also, it doesn't sound like you're hitting any speedbumps on clock speed, but if you do hit any, you could try dropping that UnCore a little. Bloomfield called for memory speed times two or some said times two plus one, but Gulftown/Westmere can get by just fine all the way down to memory speed times 1.5, or so they say. Just something to try if you ever need to.

 

[Zoson]

There's binning within Xeons. Just because it has the Xeon title doesn't mean it's binned like a 990x. I have yet to see any of these Xeons match the core quality of my 990x. Bottom line, a 5650 does not have the same quality transistors as a 5670.

If you're looking for the highest clocks, you will get them out of the higher binned parts.

And yeah, keep trying to pretend you know more than I do about x58 overclocking. All of the history is in the XS forum where we investigated everything going on here. I'm trying to bring that information to the table, and you guys are all finding out the hard way that I'm right.

Also, I paid $230 for this 990x 4 years ago, and I'm upgrading to x99 in october... So yeah, I've been enjoying the performance of a top end platform for a long time. There's nothing to be jealous about.

Anyway, go ahead and follow what this zorch guy is saying. As I said in the other thread to the guy who lost his cpu by basically doing what zorch is saying... If it takes you ~$100 to learn a lesson, that's on you.

I tried to help, and the proof is in the pudding. I'm the only one who has cited the actual intel documentation, as well as actual stress test results. So yeah, you can go ahead and believe a guy with basically no experience or credibility, or you can trust the guy who's been here for 13 years, has been overclocking for that long, and has had this platform for 5 total years.

Edit: also that low vid to high gate leakage is only partially true and not applicable in this situation. That statement was made in reference to chips all binned at the same level. I.E. if you have two 990x's and one has a vid of 1.1 while the other has a vid of 1.15, you'd want to use the one with 1.1 vid for extreme cooling cases, while you'd use the 1.15v one for regular air. That is NOT to compare two chips of different bin levels, as, again, TDP doesn't change, just the transistor quality. If you have one chip running at 2.93GHz at 1.1v and another chip running 3.47GHz at 1.1v, the higher binned chip will beat the snot out of the lower binned chip *every time*. Yes, batch is important, but even within batches, the chips are all binned at specific speeds, so the scaling principle still holds true.

 

[ZorchThatCPU]

LOL, yeah I bet you're a real 1366 expert, considering you bought the. 990 Noob Edition. No one who knows what they're doing would buy that chip. It's a complete waste of money, literally one of the worst purchases you could make. Noob.

 

[Cecil]

There's binning within Xeons. Just because it has the Xeon title doesn't mean it's binned like a 990x. I have yet to see any of these Xeons match the core quality of my 990x. Bottom line, a 5650 does not have the same quality transistors as a 5670.


Edit: also that low vid to high gate leakage is only partially true and not applicable in this situation. That statement was made in reference to chips all binned at the same level. I.E. if you have two 990x's and one has a vid of 1.1 while the other has a vid of 1.15, you'd want to use the one with 1.1 vid for extreme cooling cases, while you'd use the 1.15v one for regular air. That is NOT to compare two chips of different bin levels, as, again, TDP doesn't change, just the transistor quality. If you have one chip running at 2.93GHz at 1.1v and another chip running 3.47GHz at 1.1v, the higher binned chip will beat the snot out of the lower binned chip *every time*. Yes, batch is important, but even within batches, the chips are all binned at specific speeds, so the scaling principle still holds true.

Then obviously you didnt read the full post. Otherwise you wouldnt have had to repeat what was said in it.

Also, when binning same CPU, lower stock VID means higher leakage, which is good for going cold (LN2) but bad for daily as it will draw more power and cause more heat when OCed.

The Xeons are cut from the same wafer. They dont know when they pull them out if its going to be a X5650 or X5670 or what. They test to see what speed it will do at what TDP/voltage, then make that determination. That doesnt mean a higher spec CPU will clock better then a lower one though. As the examples given have said, how 920s can out clock higher spec 9xxs, and its the same with any other line of CPUs. 8120s can out do 8150s, E8500s that outrun E8600s, etc... It all depends on the luck of the draw. Going for the highest number does not equal best performance.

The biggest point you are overlooking, its the price to performance. Coupling a $150+ CPU with a $200 board is not a good deal for a 5+ year old platform with low SATA3 and USB3 performance. The X5650 at $70 is the only reason to consider this as a platform worth owning.