Y500 GT650M Overclock

Just ran a benchmark with an overvolted (1.1v) 650m and got 3360 in 3dmark11, with 1228 core, and 2450 memory. 1215 is furmark stable.



That means that over a stock 650m, you can get about a 50% increase in performance. That by itself is huge, and on the order of two 650m card in SLI. Now add in the additional 50%+ bonus you get from combining two cards in SLI (both overclocked), and I think you'd likely see a 125%+ increase in performance over a single card at stock settings. WOW!

 

Just done what ever you had done but i clocked core to 1200 and memory to 2550
My score is p3330
Yes really a huge difference before modding its almost 1000 difference in score
Awesome no regreting of getting 750m anymore


Sent from my Nexus 4 using Tapatalk 2

 

Looks good. You're just throwing away free performance if you don't overclock. And the risks are minimal if you don't change voltage and keep an eye on temperatures. In 3DMark 11 I got over 1100 points higher than stock when I overclocked my 650M SLI. And I got an average of 30% increase in games, which allowed me to turn up a lot of extra eye candy while still getting great FPS.

 

Actually 650M SLI gets around 3700 at stock, so it's not quite there yet, but I agree, this card can be pushed quite a bit.

I prefer to bench and gauge performance in real games instead of synthetics, so I use an 83-second FRAPS test of the beginning of Operation Swordbreaker in Battlefield 3 starting from when the doors of the LAV open. I use this section because it is repeatable, very demanding, and an accurate indicator of what a big 64-player multiplayer battle is like. I use 1080p, 70 FoV, all High settings with FXAA, MSAA, and motion blur turned off, which are the settings I play at. Anyway, long story short, I average 26 FPS with a single 650M at stock and 63 FPS with 650M SLI overclocked to 1120/2250. So you're looking at a massive 142% increase going from a single 650M at stock to 650M SLI overclocked, which is even higher than you predicted Kukri. In Metro 2033, I can't find the exact number in my benchmarks database but I recall the delta to be even bigger, something on the order of 150%. Keep in mind though, that I consider these best-case scenarios because Battlefield 3 and especially Metro 2033 are very GPU-bound games, so another title with more of a CPU bottleneck or not-as-good SLI scaling won't fare quite as well.

 

you mean dont to change the voltage manually?
already i flashed 1.1mv overvolted bios mod. will this makes any harm to my computer??
yes temperatures are in control... during 3dmark11 benching its striked 66-67c cpu , 58c gpu and 45c hdd.
after an hour of assassins creed 3 gameplay(maxed out settings(40-43 fps)) its just striked 78c.
i think its acceptable temperatures.

 

Any kind of overclocking/overvolting is going to increase the rate of electromigration (degradation) and decrease the useable lifespan of your component. By how much is anyone's guess, although I would think the higher you push the more significant it is. A mild increase may not decrease lifespan by an appreciable amount as much as a large increase would. But even then you might be looking at the component lasting 5 years instead of 10, by which time you'll probably have outgrown the laptop. There are a bunch of other variables that come into play as well such as the quality of your hardware and whether it's free from any manufacturing defects so it will be different for everyone.

There is no such thing as "safe" overclocking or overvolting but the risk is small if you're smart and cautious about it, just keep in mind that the risk is always there. There are always the rare instances of components suddenly dying. Heat is really your number one enemy as sustained exposure to too-high temperatures will damage your hardware which is why adequate cooling is a must. Luckily, the chip companies set the threshold to throttle or shut down well below the temperatures where damage would take place

I didn't overvolt my Y500 because of the temperature issues and the desire for 24/7 stability, not to mention I was happy with the overclock I achieved on stock voltage. Since this is a laptop I don't trust the integrity of hardware components such as the power supply, VRM's, etc. to provide 100% stability when overvolted like I would for a desktop PC.

Anyway, depending on how much higher you can overclock with a voltage increase, you may find the tradeoffs I have described to be worth it. It's your call.

 

I have overclocked my 750m using nvidia inspector. I was wondering if there was any way t increase the core clock speed by more than 135 mhz? Would I have to flash the BIOS to unlock the max clock speed?

 

Got my 650M SLI (306.00 drivers) up to 1050/2250 using nvidiainspector (no voltage bios mod or anything) and have stable temps of 62c under heavy gaming.
Temps at normal clocks are 54c.
My ASIC qualities are 73.4 and 78.4 for main and SLI card, not bad I think.
Seems stable so far...

 

Haven't been in here in a while. I'm currently running SLI, 1.1mv bios, and Dell 240w PS. At 1200/2600mhz in 3DMark11, it is completely stable, no artifacts. Probably will push the core up higher but I think the memory is at it's limit. I did briefly run it at 2700mhz and saw artifacts so I just backed out of the benchmark. Temps maxed out at 77C for the internal and 84C for the Ultrabay. CPU maxed out at 88C with ThrottleStop. Pretty cool day here in Korea, showing 19C outdoor temp, indoor is much warmer but I don't have a thermometer. No cooling pad, just sitting on my table. I did order a CM U2 cooling pad so it should help keep the temperatures down once summer hits. The 240w powersupply made all the difference in the world as I was running with the standard 120w PS all this time and I was unable to take advantage of the overvolt bios. I'll go single gpu later and see what kind of scores/temps the laptop can max out at.

NVIDIA GeForce GT 650M video card benchmark result - Intel Core i7-3632QM,LENOVO INVALID score: P5088 3DMarks

 

That's crazy dog. Is that even stable for gaming?

I'm gonna see how good the GPU quality in my new replacement Y500 is. I wonder if it would be good enough for me to get to 1200 MHz core on stock voltage. My previous one was decent and got 1120/2250 on stock volts and about P4900 in 3DMark 11. That was my 100% stable OC mind you and the one I ran 24/7 and in games for hours on end without crashing, at least before the laptop bricked. I was able to run 3DMark 11 at 1150/2250 and landed just barely south of P5000 but that was not stable enough for anything outside of benchmarks.

 

It seems stable in 3DMark11, nice and smooth and been running benchmarks back to back for a few hours now. I know that 1200mhz is a good number for my core because I could almost get it to work at stock voltage when the laptop was brand new (see first page, 2nd post). It would run through the benchmark but would crash back to desktop if I run it on a continuous loop in demo mode. I used to run the laptop continuously at 1155/2555 using Nvidia Inspector back when I was single GPU playing Far Cry 3 for 4-5 hours straight on the weekends. Don't really see the need to right now since all I'm playing is Simcity 5 but with the overvolt bios, 1200/2600 should be stable 24/7 if I wanted to run it that way. The key is this new PS, that 120w PS just wasn't cutting it. Also, seems like the 3632QM runs quite a bit cooler so I don't get into any real temp issues. I've also replaced the thermal grease (AS5) back in January but seemed like that actually ruined my temps as I was running cooler out of the box. I'll play around with it some more and report back, only got the PS through Amazon yesterday.

I have no doubt that you'll be able to hit 1200mhz out of the box because the 750m runs a higher stock voltage. I've had this thing since the beginning of December and it's been a love/hate relationship. Overall, still love the piece of but again, wish the execution was even half as good as the concept. Can't wait to get my new msata ssd in the mail, this harddrive is painfully slow.

 

Actually the specs of my replacement are in my sig. It's still the same 650M model but with the 16GB SSD cache. This and the cheapest model with the 750M and no other upgrades were the only configurations on the Lenovo website that ship out without a 3-week delay. And I can just use my current 650M Ultrabay instead of waiting an indefinite period of time for the 750M Ultrabay to become available, since people have confirmed that the one advertised on the Lenovo website as 750M is actually still 650M.

I was asking about the long-term stability of your overclock because from what I have seen 3DMark 11 Basic Edition doesn't push the GPU as much as recent games or Unigine Heaven, so I could get away with a higher overclock that would crash elsewhere. Unigine Heaven maxed out at 1080p is just about as good a GPU test as you'll find anywhere. If you can loop it for 24 hours without locking up, then your overclock is 100% stable. I was at 1125/2250 before and it was completely stable even in the most demanding games, but it failed Heaven in the 20th hour. So I reduced the core by 5 MHz and that made it perfect.

I'm completely in agreement with you on that last statement. It's a laptop that looks great on paper but not in actual use.

 

Hmm...my GPU tends to be around 48-50 on stock...it seems a little high and maxed out at 72C when I ran furmark test at 925/2250...seems a little high for temps, I even have a cooler, or am I okay?

I haven't installed any modded BIOS either just stock everything and ran a script to get that OC. But I see people with much higher OC and lower temps.

Should I disable the cpu turbo boost?

 

I may try some other stability tests later on but I am quite happy with how the laptop is performing so far. I really don't have the urge to run a stress test for 24 hrs because that would dig into my gaming time and seems like excessive stress on components. For me, I gauge the max overclock based on the 15min 1080p Furmark stress test because I can see where the temperatures are at after the test and determine where the max clock is based on my comfort level. Speaking of which, I have not ran Furmark since flashing to the overvolt bios.

 

No, 72C in Furmark at your overclock is exactly where you need to be. Here's the closest from me for comparison. When you say 48-50 stock, do you mean your idle temp or are you talking about in Furmark?

 

You should use Unigine Heaven. Max out the settings at 1080p and let it loop for a while. Guaranteed to crash an overclock that is even slightly too high, something which FurMark cannot claim to do. SLI doesn't work in FurMark, even if you run it in fullscreen and change SLI rendering mode to AFR 2. I've stopped using FurMark and similar tests for reasons I explain below.

After my experience overclocking the Y500, I have zero faith left in the fur donut rendering benchmarks like FurMark, MSI Kombustor, and OCCT as far as determining overclocking stability. They're completely sythetic tests that just do a bunch of repetitive calculations designed to generate maximum heat. They tell you absolutely nothing about how stable an overclock will be in actual games because they're not based on real-time 3D game engines. I stopped using FurMark for overclocking since a few years ago because I found that I could go super high in FurMark but crash instantly when firing up a game. I tried OCCT this time around because it seemed like it would be more accurate since it is Direct3D not OpenGL and SLI works properly, but it's the same thing. I was up to 1200 MHz on the core and still going strong in OCCT, but it felt fishy so I fired up Heaven and BAM instant crash. I eventually had to reduce it all the way down to 1120 MHz before it was really stable so that should tell you about the misinformation OCCT was feeding me.

Plus, if you are overclocking VRAM you need something that will run a benchmark preset and generate a score at the end. GDDR5 is crash-resistant and the only way to know you've gone over the limit is if your score goes down since GDDR5 pretty much will not crash or artifact.

Anyway, Heaven is great because it is DirectX 11 and extremely demanding when cranked up plus has a scored test. It will heat up the GPU just as much as FurMark, if not more, and is more accurate for stability testing. It's become my one GPU test of choice.

 

Ok, installed Heaven. So, when you say run it with settings maxed out, do you mean run it in the Extreme preset? How long am I supposed to run this thing for it to be stable by your book?

 

No use the Custom preset and turn everything up as high as it goes as such:



Extreme only runs at 1600x900 in a window.

I'd say at let it loop for at least an hour. If your overclock can survive that it should be stable in games. I like to do 24-hour runs to fine tune the core down to the last 5 or 10 MHz of stability but that's just me.

 

Yup, just finished with the hour loop and ended with the benchmark to end the test.



Of note: The ultrabay gpu got hot as hell, but settled at 97C. The internal gpu stayed cool and topped off at 77C. Kind of make me want to find a better way to cool the ultrabay, hopefully the temps will drop once my CM Notepal U2 arrives. As usual, did the test with the laptop sitting flush on the desk. Outside temperature was 75F according to the Weather Metro App.

In other news, ran at 1225/2600mhz yesterday and benched 3DMark11.

 

Wow. Your VRAM is overclocked WAY past its limit. If it were stable you would not be getting such a low score, especiallly with the core so high. That's actually lower than what 650M SLI gets on stock clocks. This is what I get in Heaven on the exact same settings at 1120/2250:



This is exactly what I was referring to when I said you need to use a scored benchmark when overclocking VRAM. As soon as your score drops, that's when you know you've reached your limit. GDDR5 does not crash or artifact if pushed too far. What happens is error checking kicks in and it resends the data that failed the first time due to instability, which causes your score to drop. I see so many 650M SLI owners running their VRAM at 2500 MHz or higher and I really have to question whether they're aware that they're most likely reducing performance across the board without even knowing it.

The reason this didn't stick out like a sore thumb in 3DMark 11 is that memory clock and bandwidth increases/decreases seem to have very little effect on the score. The Performance preset at 720p with no AA is just not long or demanding enough to highlight a problem like this. Just another reason I stick with Unigine Heaven.

Now that you're aware of this, it's a good time to revisit that VRAM overclock. You've been inadvertently lowering your performance even below stock clocks but I guess you didn't notice it when gaming.

That Ultrabay GPU temperature sounds pretty alarming as well. Maybe it's time to blow out the dust or repaste. I know you are overvolting to 1.1V but the difference shouldn't be that huge. At stock voltage and 1125/2250 my Ultrabay doesn't exceed 80C on the hottest days here in California.

 

I'm talking idle. Should I disable turbo boost? Because I know that greatly increases the temps but a small performance boost

 

I'm tempted to do the bios unlock but I have to update mine to 2.02 first. There's a lot of different links floating around, which one did you guys use?

Sent from my SCH-I535 using Tapatalk 2

 

I'll play with the memory settings when I get home. I ran that benchmark after an hour of running the demo so maybe that had something to do with it also. I literally let it run for an hour and then hit the benchmark button on the upper left corner. The ultrabay is definitely running way too hot, I'll probably repaste it with AS5 this weekend but I want to wait till the cooling pad gets here to see if that helps with the temp first.

 

Idle is a little high, I usually idle in the mid 30's but I wouldn't be too concerned if it is only getting up to 72C in Furmark. I don't know what you mean by turbo boost.

 

Nevermind...idle it is at 35-37C I guess it was still hot from the last test I ran OCed.

Turbo Boost is in the CPU I believe and I think in the y500 the CPU and GPU share the same cooling system. I read somewhere in this thread that disabling turbo boost reduces temps as well. Even though it reduces the performance, it is very minimal.

Once again I am new at this but I'm 99% sure I read this somewhere in this thread.

EDIT: Okay so I got home today and tried to run the OC.bat and the GPU clock would change but wouldn't go past 924 and current clock was stuck at 135...yesterday it was OCing just fine..could it be the drivers? Because I was using the same drivers yesterday and it was working just fine and I could OC as much as I wanted (latest ones btw, 320.18)...on Mobius 1's guide it says it should work, or am I misunderstanding?)

The command I am running is...
@echo off
C:\nvidiaInspector\nvidiaInspector.exe -setBaseClockOffset:0,0,135 -setMemoryClockOffset:0,0,250 -setGpuClock:0,2,925 -setMemoryClock:0,2,2250 -forcepstate:0,0
||-forcepstate:0,5

EDIT2:After reinstalling the drivers it was fixed and after using the command :
@echo off
C:\nvidiainspector\nvidiaInspector.exe -setGpuClock:0,2,790 -setMemoryClock:0,2,2000 -setBaseClockOffset:0,0,0 -setMemoryClockOffset:0,0,0 -forcepstate:0,16

The clock is supposed to be reset (790/2000) but still runs pretty hot (49-52C again...) ...do I just need to wait a while for it to cool back down?

EDIT 3: wow..lotsa things...I ran one unigine heaven benchmark at 1100/2250 and my temps went up all the way to 78C...is that good or too high? all ultra btw

and 790 isn't the original..right? because on full ULTRA in WoW I used to run at max 15 FPS now with the reset command mentioned above it goes to 790mhz and i run 40+ fps ... lol

 

No letting it loop for an hour beforehand should have no impact. If it's completely stable you'll get the same score within +/- 5 points if you run it a thousand times. My result was from after letting it run for 24 hours. You'll find that you'll get a much higher score if you reset memory to stock and run it with just the core overclocked which validates what I said about the VRAM errors. Interestingly enough, I got the exact score you did when my VRAM became unstable, so there seems to be some consistency for this GPU when the memory is clocked too high.

When you play with it again I recommend starting low at a speed you know is completely stable, like say 2100 MHz, and slowly going up from there. Keep the core overclocked while you do it. It's better to start stable and make it unstable instead of vice-versa because if you go from high to low you might get some false-positives. Plus the score could stay low even if you reset memory to stock due to errors still resident in VRAM from when it was clocked too high, which would require you to set memory back to stock and reboot the system to flush it out. So start low and go up slowly, and if your VRAM has passed its limit it should be obvious because the score will suddenly fall off a cliff.

 

I discovered the issue and it was staring me in the face the whole time. The ultrabay gpu was running too hot due to the raised voltage. I flashed the ultrabay back to stock voltage and ran using your settings (1125/2250) and got this:



I also ran the benchmark at 1200/2600 while the laptop was cool and managed this:



Might be the reason why the new Y500s with the updated 750m is single gpu only at the moment while they sort out the cooling issue with the ultrabay. I'll probably open it up later and see if there's anything I can do to help reduce the temps. The good part is that my internal gpu/memory is very overclockable.

 

Yeah, your idle temps are good at 35-37, that is exactly where I'm at. I use ThrottleStop to stay in turbo so I'm doing the exact opposite of what you are thinking about doing. My processor runs cool so I don't have to worry about the heat associated with it. Performance is pretty significant, especially in the physics score in 3dMark11 which helps greatly with the overall score. Your unigine heaven temps are still good, nothing to worry about at 78C.

 

guys just a noob question read the entire thread but couldnt figure it out. i have the latest 320 drivers from nvidia and nvidia 650m 1x. using nvidia inspector however using offset or setting clocks in any performance mode (P0 P5 P8) the clock remains the same at 790mhz. i cant get the clock to increase in any mode it shows est max at 925 mhz but remains at 790 mhz. any help?

 

Cool! So it looks like the low score was due to heat not unstable VRAM, although make no mistake running your VRAM clocked too high does reduce performance. Was the 1200/2600 managed on stock volts or at 1.1V? If at 1.1V I'm guessing you benched Heaven as soon as you fired it up before temps got out of control?

Consider yourself lucky. It looks like you got a good GPU and are able to push a lot out of it. Now to do something about that Ultrabay temperature.

If it's not too much trouble, would you mind telling me the ASIC quality of both your GPU's in GPU-Z? I don't know if it's accurate or even has any correlation with how high you can overclock but it would be good to know. I'm at 73.5% and 84.5% for the internal and Ultrabay GPU's respectively. Once I get my replacement Y500 and return this one I'll check the reading on the new machine. Hopefully if it's higher than 73.5% I can overclock higher than 1120 MHz core on stock volts since I'll be keeping my current Ultrabay. I don't plan to overvolt in the forseeable future because of heat and longevity concerns.

 

The forum in Techinferno is much better when it comes to knowledge about this stuff. Unfortunately, I've since stopped using nvidia inspector to overclock so I can't really help you. I'm now running a modded bios and overclocking with MSI Afterburner. The issue with the new drivers and you probably know this already is that they are locked but you should still be able to overclock it by +135mhz. Again, my knowledge about this stuff has slipped quite a bit. Sorry.

 

Yes, still using the overvolt bios, I flashed the ultrabay back to stock for a brief moment just to confirm the heat issue but now back to the overvolt again. The 1200/2600 run was done right after startup and it deteriorated soon after when the temps got up to 97C. It was probably throttling like a mother f*(ker. There's really no good solution to this cooling issue, I'm testing out some ideas but the little hole they have sucking in air for the fan is just too small. The best way is to Dremel that metal piece that lines the bay so that it's not blocking most of the air intake on the bottom cover.

Internal:

Ultrabay:

 

what's the purpose of using the over volt bios?

 

Those ASIC numbers look very good, among the best I've seen for the 650M yet. If the numbers mean anything it could explain your particular machine's overclocking prowess. I recall reading somewhere that earlier production runs of a GPU tend to have less leakage and since you bought yours way back in December that would make sense. I'm just hoping when my replacement Y500 arrives that its internal GPU is close to the 84.5% ASIC of my Ultrabay. Apparently Lenovo is not binning for the 750M because I've seen owners of the new machines with ASIC quality in the 60's and 70's. So the higher clock speed of the 750M is mainly due to higher voltage and not better quality chips.

Have you opened up the Ultrabay and repasted yet? I'm thinking it would be wise to do that first before physically modding the machine. Who knows, you might find poorly applied or low-quality TIM or even loose heatsink screws and just repasting and tightening down the heatsink could fix all your problems. I've had laptops in the past that had major heat issues right out of the box and when I opened them up there was more than a quarter-turn left in the heatsink screws and it looked like they used an entire tube of TIM on the die. You might want to consider using another TIM besides Arctic Silver 5 if the contact between heatsink and die is not great due to lack of pressure or a lot of surface imperfections. I've been using OCZ Freeze for a few years now and it's given me better results than ICD 7 and AS 5, plus I don't have to worry about curing time, electrical conductivity or capacitativeness, or the metal/diamond particles causing microscratches. OCZ Freeze is discontinued (I stocked up on it a few years ago ) but there are a lot of other good ceramic-based TIM's worth a look.

Also, is there a large temperature delta between the internal and Ultrabay GPU's for you? Mine are usually run about the same temperature except in a very GPU-intensive scenario like Heaven where I have observed up to a 10C increase in the Ultrabay. Neither ever exceeded 80C though so I'm not worried. Mind you, I was running at 1120/2250 on stock volts so that would definitely help, but would 1200 MHz at 1.1V really increase Ultrabay temperature by 20C? I don't know, maybe you live in a very hot area, but that's some food for thought.

 

So you can overclock the core higher than what would be stable on stock volts.

 

Lol, my Asic is 88.7%, but I can only reach a stable 1205 mhz at 1.1v :/. Most cards I saw from the early releases had Asics in the 90s.

 

But that increases the temps, correct?

 

Yes.

10char.

 

10char?...sorry if dumb question lol

 

Type a short one-word reply and see what happens.

 

Oh Lol okay

 

I re-pasted the ultrabay heatsink with AS5 and from the last few benchmarks, the ultrabay has been running just over 4C of the internal gpu under load and 3C when idle. Looks promising, maybe the temperatures will drop some more after the AS5 settles in. Disassembly of the ultrabay couldn't have been easier. I'm sure if the cooling pad ever gets here in the mail, the temps will drop even more. Still have the laptop suffocating itself on the desk.

Here's the pile of crap Lenovo used on the gpu and heatsink.

 

Lenovo Energy Management lets you choose power plans the same way you can through Windows. It was the only Lenovo software I kept when I did my clean install of Windows 8 Pro because it's got the nifty dust removal feature. The reason your scores were much lower on the Balanced plan was that PCI express link state power management was enabled which reduces the performance of the GPU. I just use the High Performance plan and customized it to my liking and don't bother switching between them. If you want ThrottleStop to manage your CPU correctly you need to set both the minimum processor state and maximum processor state to 100% for the current plan.

 

Looks like the factory paste job put way too much TIM on the die. Look how much spilled over the sides. Mine looked exactly the same when I opened it up. You can see the large amount around the edges underneath the plastic: http://forum.notebookreview.com/ide...-pictures-disassembled-y500-ultrabay-gpu.html

What's interesting in my case, though, is that I repasted and it didn't decrease the temperature at all, not even a single degree. I even did it a couple times just in case my application wasn't very good the first time, but nada. My Ultrabay was running almost identically temperature-wise to the main GPU in games anyway so I wasn't worried. I'm wondering if yours was simply a case of loose heatsink screws as your factory paste job wasn't any worse than mine.

 

Interesting to see both your gpu temperatures running in sync of each other. The only game stress intensive I've played is battlefield 3, and temperatures being 5-6 degrees C difference between the two with the ultra bay gpu running hotter. I know by default that the ultra bay gpu is supposed to run hotter.Any thoughts?

 

So how do you exactly do this, keep your y500 in turbo? Yes I know you use throttlestop, but could you explain more in detail.

 

Here are my ThrottleStop settings:

 

Very good insight on the PCI express link state and processor state, thanks! I'll mess with those settings and run the benchmarks again.

Experimenting with my clocks still, here's 1225/2575:

 

Think it was reading your post that had me hold off on re-pasting so long but 97C in Heaven made me do the re-paste. I have not seen it get into 90C yet and today was a warm day. Also, it stayed pretty close to the internal gpu temps so I'm pretty happy for now. Like I said, once I have the cooling pad under there, I think it'll be even better.

 

I haven't followed the Y500 thread for a while. But PhysX should show up as ticked. Like mine. It didn't show up at first for me, I am suspecting it's a problem with the installer. I used driver sweeper to delete the Physx driver and manually install Physx, then GPU-Z shows it. It could be either GPU-Z not detecting it or the driver's installer not installing Physx properly.

View attachment 96768

By the way, I am not sure how accurate is the ASIC quality in GPU-Z. Higher quality ASIC means lower leakage, but high leakage chip may sometimes overclock better, although it comes with much higher heat. I have seen AMD's cherry picked ultra high leakage CPU that're used for LN2 overclocking.

Oh I found it. The AMD TWKR Phenom.

http://www.tomshardware.com/news/amd-phenom-twkr-contest-overclocking,8166.html
http://www.overclockersclub.com/reviews/amd_twkr_edition_cpu_preview/