12.5" HP Elitebook 2570P Owner's Lounge

[BxL]

There are some results:

My 2570p config:

i5-3340m@2,7GHz (HD4000 IGP), 2x4GB Micron KIT DDR3 1600Mhz, 2xKingston 60GB SSD sata3 @ RAID0 (one in a HDD caddy at the upgradebay), Win7 prof x64 sp1.

Boot speed is 6 to 8 seconds after the BIOS HP logo,

only Intel RST app (2 process, IaStorIcon and IaDataMgrSvc) set to delayed start.

[bjorm]

I tried to set turbo mode by throttlestop but nothing changed.

Here You comparision for 3 mission (FHD)

red: without throttle, high settings

green: with throttlestop, high settings

blue: with throttlestop, low settings

As You can see, even with low setting there are specific places where games is barely playable, it is caused by BusLoad choking. The reason is opt1.2 only, not higher.

Here is FPS with Mutli on High setting with res. 1680x1050.

Hwinfo after 3,5h of BF4 playtime. As we can see, there is 95% working on turbo mode but CPU was using only on 40-60%

[Tech Inferno Fan]

I tried to set turbo mode by throttlestop but nothing changed.

Here You comparision for 3 mission (FHD)

red: without throttle, high settings

green: with throttlestop, high settings

blue: with throttlestop, low settings

As You can see, even with low setting there are specific places where games is barely playable, it is caused by BusLoad choking. The reason is opt1.2 only, not higher.

..

Hwinfo after 3,5h of BF4 playtime. As we can see, there is 95% working on turbo mode but CPU was using only on 40-60%

@bjorm, can you get your hands on a AMD card like a HD7870 or better? During my eGPU testing I found some games function better with AMD cards than NVidia. Furthermore, AMD has more economical bandwidth usage in DX10/DX11 where x1.2Opt pci-e compression has little effect.

sskillz reported with his HP 2560P i5-2540M 2.6 +HD7970@x1 2.0 eGPU implementation :

It also plays mp bf3 ultra at 100fps and bf4 utlra with 2x msaa at 70 with hdmi sound. I'm worndering if to post a video?

[bjorm]

@bjorm, can you get your hands on a AMD card like a HD7870 or better? During my eGPU testing I found some games function better with AMD cards than NVidia. Furthermore, AMD has more economical bandwidth usage in DX10/DX11 where x1.2Opt pci-e compression has little effect.

sskillz reported with his HP 2560P i5-2540M 2.6 +HD7970@x1 2.0 eGPU implementation :

@Tech Inferno Fan hehe Few minutes ago I was browsing internet looking for good competitor for GTX760 Im not sure if I am able to let myself of buying 7970 but what AMD is arround GTX760 efficiency? Another issue is DSDT and setup stuff necessary to run AMD, so I'm not keen on to play with them

[unclewebb]

Can you guys post some screen shots of how you have ThrottleStop set up? The screen shot on the previous page showed that it was in monitoring mode and wasn't even turned on. ThrottleStop isn't going to be able to do anything to help the cause in monitoring mode.

For the Dual and Quad Core mobile CPUs with a 35W TDP, Intel lists the maximum turbo time period as 64 seconds and ThrottleStop confirms that. When adjusting the ThrottleStop turbo time limit, I would not go beyond that value. The time adjuster was left unlocked so users can experiment but it is best to start with the datasheet maximums and go from there. Going beyond the maximum might not work correctly at all.

The default long and short turbo power limits are listed as 35W and 43.75W (35W + 25%). The maximum for these are shown in the Intel datasheet as 48W and 56W. Those are the maximum values so don't go beyond those values when setting the turbo power limits.

If you are having some CPU performance issues then turn off your other monitoring software and run ThrottleStop with the Log File option checked so I can have a look at some numbers. ThrottleStop might not be able to help you out but you have to know how to properly set it up. I am interested to see some ThrottleStop CPU data when the bus load choking is going on.

ThrottleStop 6.10 beta 1

http://www.mediafire.com/download/i4jp6b8uk1cza5w/ThrottleStop_610b1.zip

The latest beta fixes a few minor things and maintains the Power Balance values which can be handy when using the Intel GPU.

Dufus on NBR recently showed me another way for CPUs to reduce the maximum multiplier. He wrote a quick little tool called MaxMulti which is available here.

http://forum.notebookreview.com/hardware-components-aftermarket-upgrades/531329-throttlestop-guide-176.html#post9410230

This can allow overclocking of the partially unlocked CPUs like the 3720QM, 3820QM, 4700MQ, etc and up but the chipset also has to support this overclocking feature for it to work. There wasn't much feedback about this on NBR but MaxMulti definitely worked correctly on my unlocked 3570K. I am hoping to build this feature into a future version of ThrottleStop. Thanks Dufus.

[Tech Inferno Fan]

INFO 2570P DSDT override so can host AMD eGPUs

@Tech Inferno Fan hehe Few minutes ago I was browsing internet looking for good competitor for GTX760 Im not sure if I am able to let myself of buying 7970 but what AMD is arround GTX760 efficiency? Another issue is DSDT and setup stuff necessary to run AMD, so I'm not keen on to play with them

As you know, the 2560P/2570P BIOS pci-e allocation is such that there is no 256MB contiguous space to host a AMD eGPU or GTX650/GTX750/GTX9xx cards. The solution is to either use Setup 1.x's PCI compaction OR to perform a http://forum.techinferno.com/diy-e-gpu-projects/7476-%5Bguide%5D-dsdt-override-fix-error-12-a.html#post102517.

I've made the latter easy for you. Here's a Win7 DSDT override I created from my F.42 BIOS: 2570P-dsdt-override . Just extract and run 'load', then reboot. If you get a BSOD on startup then reboot and select "restore last good menu" to effectively erase that registry override. If it loads OK then can confirm you now see a Large Memory section in Device Manager -> Video Resources by type as shown below. This may be sufficient to then have Win7 auto-allocate an AMD card into the Large Memory area so an AMD card can work in the same plug-and-play manner as a NVidia GTX560Ti/GTX660+ one.

If you use a different bios version then the addresses in the DSDT override will not match. In that case create a new DSDT override from scratch:

1. Download latest IASL compiler from here. At the time this was written that was 20131115

2. Extract the DSDT table that your Windows is using with 'iasl -g'

3. Edit 'dsdt_17df.dsl', deleting:

- This unnecessary block to make compiled DSDT smaller than original DSDT

If (_OSI ("Windows 2001"))

{

Store (0x07D1, OSYS)

}

If (_OSI ("Windows 2001 SP1"))

{

Store (0x07D1, OSYS)

}

- This block to allow compile to happen without errors:

If (CondRefOf (FPED))

{

FPED ()

}

4. Edit 'dsdt_17df.dsl', adding QWord Memory entry per http://forum.techinferno.com/diy-e-gpu-projects/7476-%5Bguide%5D-dsdt-override-fix-error-12-a.html#post102517 .

5. Compile 'dsdt_17df.dsl' with 'asl dsdt_17df.dsl'

6. Either load the DSDT table as a registry override (Win7) or as a in-memory substitution using Setup 1.x (Win8) per the http://forum.techinferno.com/diy-e-gpu-projects/7476-%5Bguide%5D-dsdt-override-fix-error-12-a.html#post102517 .

7. Reboot system to use the new DSDT table.

8. Confirm that the new DSDT table has a 'large memory' entry OR do a 'iasl -g' to decompile the DSDT table windows uses, then use an editor on dsdt_17df.dsl and confirm the QWordmemory entry is shown.

Can you guys post some screen shots of how you have ThrottleStop set up? The screen shot on the previous page showed that it was in monitoring mode and wasn't even turned on. ThrottleStop isn't going to be able to do anything to help the cause in monitoring mode.

For the Dual and Quad Core mobile CPUs with a 35W TDP, Intel lists the maximum turbo time period as 64 seconds and ThrottleStop confirms that. When adjusting the ThrottleStop turbo time limit, I would not go beyond that value. The time adjuster was left unlocked so users can experiment but it is best to start with the datasheet maximums and go from there. Going beyond the maximum might not work correctly at all.

The default long and short turbo power limits are listed as 35W and 43.75W (35W + 25%). The maximum for these are shown in the Intel datasheet as 48W and 56W. Those are the maximum values so don't go beyond those values when setting the turbo power limits.

If you are having some CPU performance issues then turn off your other monitoring software and run ThrottleStop with the Log File option checked so I can have a look at some numbers. ThrottleStop might not be able to help you out but you have to know how to properly set it up. I am interested to see some ThrottleStop CPU data when the bus load choking is going on.

ThrottleStop 6.10 beta 1

ThrottleStop_610b1

The latest beta fixes a few minor things and maintains the Power Balance values which can be handy when using the Intel GPU.

Dufus on NBR recently showed me another way for CPUs to reduce the maximum multiplier. He wrote a quick little tool called MaxMulti which is available here.

The ThrottleStop Guide - Page 176

This can allow overclocking of the partially unlocked CPUs like the 3720QM, 3820QM, 4700MQ, etc and up but the chipset also has to support this overclocking feature for it to work. There wasn't much feedback about this on NBR but MaxMulti definitely worked correctly on my unlocked 3570K. I am hoping to build this feature into a future version of ThrottleStop. Thanks Dufus.

Hey Unclewebb, thank you for the comprehensive answer. I've managed to secure a i7-3740QM which I'd like to attempt to optimize performance for. It's got a x35 4-C Turbo mode which I see engage for some short time (eg: 20s or so with a ~38W TP) after which it drops down to x34 along with a ~36W full load TDP. I haven't been able to get a constant x35 multi running. Though I haven't improved cooling (2560P heatsink would be a good start) and do see the cores hit 100c under full load after some time AND am using a 65W AC adapter as don't have a 90W one handy. Could either of those be the reason I'm getting a x34 rather than x35 4C-Turbo mode?

With your notes above, do you advise setting a 45W CPU with TPL 48W (short) and 56W (long)? I've noticed that the Intel Power Balance setting will change from 0/8 to 0/24 when I change the Intel iGPU from Maximum Battery Life to Maximum Performance. The latter seeing 3dmark11 score almost double due to improved graphics score but along with it came a 5% (or so) CPU performance drop. Clearly the whole CPU+iGPU package is being maintained under TDP.

@All, can a NBR member rehost MaxMulti for some of us that aren't (or are no longer) NBR members? They want a login to do a file download. Would be interesting to see if Dufus uncovered a way to get partially unlocked multis working. MaxMulti is at http://forum.notebookreview.com/hardware-components-aftermarket-upgrades/531329-throttlestop-guide-176.html#post9410230

Throttlestop screenshots of 2570P (F.42 BIOS) + i7-3740QM quad-core CPU

[unclewebb]

Thanks Tech Inferno Fan for the screen shots.

The 3rd Gen Core i7 mobile Quads with a 45W TDP use default long and short turbo power limits of 45W and 56.25W (45W + 25%). ThrottleStop only reports whole numbers. The maximum time limit is 64 seconds, same as the 35W TDP CPUs. The maximum Long term power limit is 60W and the maximum short term turbo power limit is 72W for the 45W TDP CPUs. These are the maximums listed in the Intel datasheet so seem like a good place to start.

ThrottleStop 6.10 correctly reports that your CPU supports +4 bins of overclocking. The problem is that it also reports that the Overclocking BIns feature is locked by the bios. The lock bit is located in MSR 0x194 - Bit[20]. You would need a modified bios to unlock this and you also need the appropriate chipset that supports overclocking. I think Dufus has a trick where he can get this overclocking working on the Intel chipsets that don't support overclocking. Unfortunately he is not available at the moment to tell me more about this.

Because this feature is locked on your motherboard, his MaxMulti program is not going to help you.

After you adjust the TDP values, if you are not getting the full 35 multi when full load testing, try setting the Power Balance to CPU 31 - GPU 0. Just do some quick testing with TS Bench and Prime 95 to see what sort of multis you get.

The Intel turbo feature is based on a time weighted average. When set to 28 seconds, this does not guarantee that you will get the full 28 seconds at the Short term turbo power limit. If you are significantly over the Long term limit then the amount of time you actually get at the higher limit is reduced.

Dufus also showed me another bit of magic that controls the TDP Level Control. Your CPU does not use this feature but there might be a hidden register that overrides the power limits that you can access in ThrottleStop. I am going to see what I can do with this new info but I might need the expert to come and help me out.

Here is some of his multiplier magic. Thanks Dufus.

MaxMulti

https://www.mediafire.com/?ruis7l5f4bs1am5

It's possible that your 65W adapter simply doesn't have enough power to run everything at full speed.

[Tech Inferno Fan]

Thanks Tech Inferno Fan for the screen shots.

The 3rd Gen Core i7 mobile Quads with a 45W TDP use default long and short turbo power limits of 45W and 56.25W (45W + 25%). ThrottleStop only reports whole numbers. The maximum time limit is 64 seconds, same as the 35W TDP CPUs. The maximum Long term power limit is 60W and the maximum short term turbo power limit is 72W for the 45W TDP CPUs. These are the maximums listed in the Intel datasheet so seem like a good place to start.

ThrottleStop 6.10 correctly reports that your CPU supports +4 bins of overclocking. The problem is that it also reports that the Overclocking Bins feature is locked by the bios. The lock bit is located in MSR 0x194 - Bit[20]. You would need a modified bios to unlock this and you also need the appropriate chipset that supports overclocking. I think Dufus has a trick where he can get this overclocking working on the Intel chipsets that don't support overclocking. Unfortunately he is not available at the moment to tell me more about this.

Because this feature is locked on your motherboard, his MaxMulti program is not going to help you.

After you adjust the TDP values, if you are not getting the full 35 multi when full load testing, try setting the Power Balance to CPU 31 - GPU 0. Just do some quick testing with TS Bench and Prime 95 to see what sort of multis you get.

The Intel turbo feature is based on a time weighted average. When set to 28 seconds, this does not guarantee that you will get the full 28 seconds at the Short term turbo power limit. If you are significantly over the Long term limit then the amount of time you actually get at the higher limit is reduced.

Dufus also showed me another bit of magic that controls the TDP Level Control. Your CPU does not use this feature but there might be a hidden register that overrides the power limits that you can access in ThrottleStop. I am going to see what I can do with this new info but I might need the expert to come and help me out.

Here is some of his multiplier magic. Thanks Dufus.

MaxMulti

https://www.mediafire.com/?ruis7l5f4bs1am5

It's possible that your 65W adapter simply doesn't have enough power to run everything at full speed.

Thank you for the followup.

Did change those settings (TPL short=72W long=60W, duration=64s, power balance CPU=31, iGPU=0) as well as the ones at http://forum.notebookreview.com/hardware-components-aftermarket-upgrades/531329-throttlestop-guide-182.html#post9439091 and still get a max sustained x34 4-C multiplier.

Don't have a 90W AC adapter to test but I have do have "Boost Converter" enabled in the BIOS that advises "gain momentary performance gains by drawing extra power from the battery". I have a 62Whr battery and a 65W AC adapter.

@jacobsson , with your i7-3820QM and (I believe) 90W AC adapter, modified cooling system, does Throttlestop's '1024M 8-thread TS Bench' see you maintain a x35 multiplier? Or does it running x35 for what is like 20s (or less) then drop down and maintain x34?

MaxMulti didn't do anything. It reported "Allowable multi already set to max". Seems I'm treading down the same path that NBR user Tanware went with his i7-3820QM. He had the same BIOS-locked extra turbo bins. Only difference is he has a Samsung NP700G7C-S01US with HM76 chipset, a i7-3820QM and a somewhat stock Insyde BIOS. I've got a 2570P with HM77 chipset, i7-3740QM and what may be some HP modified Insyde BIOS.

Tanware was able to unlock those extra turbo bins via what appears to be alteration to Insyde BIOS UEFI variable settings : The ThrottleStop Guide - Page 180

Utility to enable MSI hidden BIOS settings without flashing a modded BIOS - Page 4

Something I'll need to investigate if it's possible on the 2570P.

[ha1o2surfer]

I read this whole thread on the 38xx series CPU and have a few things to chime in and say. I have a fully unlocked BIOS on my W110ER and have the power limits set pretty high. I will say that there is a time where the performance can be improved maybe 5-8% but the power draw will increase 10-15 watts for some reason. I would stay under x34 because anything higher can cause 55+ watts SUSTAINED power draw and this may not even be stable depending on what power delivery system the 2570p has. I also don't know the cooling system but this can almost def. cause a throttling scenario.

Also try using Intel Burn in test or Linx to test the max multi your CPU will run at. I have noticed very few benchmarks will up the multi to 32+/ power draw past 40-45 watts.

[unclewebb]

When long term full load testing, does ThrottleStop show an exact multiplier like 34.00? What does ThrottleStop report for power consumption? Is there any difference between your tests when using TS Bench or Prime95? Just curious. Dufus has a few more tricks up his sleeve. Hope he returns to NBR some day.

Your CPU and chipset support overclocking. A bios that disabled the lock bit, MSR 0x194 bit[20], might be enough for you to do some overclocking.

Can you run my MSR Finder tool and take a snapshot of your CPU registers when your CPU is being throttled?

MSR Finder

http://www.mediafire.com/download/b2tg2yfjt27csy3/MSRFinder.zip

You can Copy and Paste it to www.pastebin.com or anywhere convenient.

[Tech Inferno Fan]

When long term full load testing, does ThrottleStop show an exact multiplier like 34.00? What does ThrottleStop report for power consumption? Is there any difference between your tests when using TS Bench or Prime95? Just curious. Dufus has a few more tricks up his sleeve. Hope he returns to NBR some day.

Your CPU and chipset support overclocking. A bios that disabled the lock bit, MSR 0x194 bit[20], might be enough for you to do some overclocking.

Can you run my MSR Finder tool and take a snapshot of your CPU registers when your CPU is being throttled?

MSR Finder

MSRFinder

You can Copy and Paste it to Pastebin.com - #1 paste tool since 2002! or anywhere convenient.

CPU shows x35 on all cores for a short period (looks like 5s if that), then goes to constant x34 on all cores for the duration of the 1024M TS Bench test as shown by the screencap below.

Here's the pastebin of a MSRFinder snapshot while running x34 on all cores: 2570P i7-3740QM @ x34 4-C - Pastebin.com

There has not been a successful BIOS mod on any i-core Elitebook. HP RSA protect their BIOS. When the system boots and detects a modified BIOS it just halts and does nothing. See only Tanware/Dufus' UEFI variable method as accessing those extra turbo bins. Unless of course Dufus has some other tricks up his sleeve.

[ha1o2surfer]

If I have to make a guess here I would say that HP has configurable TDP locked on the HP BIOS. Your CPU is only running x1 under full turbo boost (for the 3740QM) due to this. In order to go up to the next level of turbo boost, and in your case x35 multi, the nominal TDP limit would have to be increased to 45 watts. I won't go into detail since Unclewebb knows all about this. I can replicate the issue on my laptop but manually setting the multis and watching the power consumption. The nominal TDP in your case is 36 watts which is default on 45 watt quad core IvyBridge CPUs. If you were to fire up the iGPU the multi would drop to 27 to stay under that limit. This would be my guess and I think only a BIOS mod would fix this.

[unclewebb]

When the system boots and detects a modified BIOS it just halts and does nothing.

Manufacturers are taking all the fun out of being an enthusiast these days.

I had a look at the MSR Finder data. There are 2 MSR registers that can limit the PP0 and PP1 power consumption but they are not being used. I looked at a few other things but I can not see anything obvious. Whatever is limiting you to the 34 multiplier is outside my level of understanding.

[Tech Inferno Fan]

Manufacturers are taking all the fun out of being an enthusiast these days.

I had a look at the MSR Finder data. There are 2 MSR registers that can limit the PP0 and PP1 power consumption but they are not being used. I looked at a few other things but I can not see anything obvious. Whatever is limiting you to the 34 multiplier is outside my level of understanding.

i7-3740QM running at x34 instead of x35

I posted about this yesterday and today I went back to the store to do further testing.

- I used a 120W AC adapter -> no power starvation issues

- I opened the bottom of the case and physically pushed down two large pieces of copper over the heatsink, risking burning my fingertip ends in the process (ouch!). Highest temps now at full CPU load were 95 degrees -> no CPU throttling due to temps.

Yet still I found my CPU was hitting x35 for a split second, then reverted to x34 never to again run at x35. I tried correcting this with thse further steps:

- I turned screen brightness down, remove battery and remove 1 RAM stick to lower system power consumption in case the systemboard was limiting current -> no difference.

- I ramped up Throttlestop bench from x30, x31, x32, x33, x34 (then got stuck there.. increase to x35 kept it at x34). This was in case the CPU was flagging some internal temp spikes and downgrading performance. With a slow ramp that there could be no high temp spike. Still I was only getting x34.

- modified the TPL settings with increase short/long TDP and duration -> no difference.

So my CPU runs to spec with 1 or 2 cores active, but with 4 cores it runs like a i7-3720QM and a hot one at that. I'm concluding that I have a dud CPU. Am annoyed paying a premium for a i7-3740QM over a i7-3720QM yet getting the same 4C performance as one. Read another non-2570P user with a i7-3820QM at http://forum.notebookreview.com/sager-clevo/683364-i7-3740qm-5.html#post8785803 also with no thermal issues sees his CPU max out at x34 rather than x35 though he was able to modify the TDP settings to correct it. Hmm . . .

During this process I discovered that the stock heatsink lacks the heatsoaking capability required to run such a hot CPU. It needs more mass in the form of a few copper pieces. Aikimox's RAM heatsink mod a great way of doing this.

Above: Problematic Intel i7-3740QM SR0UV

Devastatingly fast eGPU performance

The i7-3740QM + [email protected] is giving my DX9 eGPU numbers (presumably a single/dual-core app) that nudge me to the top of the DX9 eGPU leaderboard. My > 26k 3dmark06, the first to break that barrier, edges me slightly ahead of the reigning DX9-leader, gothic860's Lenovo W530 i7-3720QM + [email protected]. Proof below:

3dmark06: 26121 - NVIDIA GeForce GTX 660 Ti video card benchmark result - Intel Core i7-3740QM Processor,Hewlett-Packard 17DF

RE5-dx9-var: 228.9

[bjorm]

Devastatingly fast eGPU performance

The i7-3740QM + [email protected] is giving my DX9 eGPU numbers (presumably a single/dual-core app) that nudge me to the top of the DX9 eGPU leaderboard. My > 26k 3dmark06, the first to break that barrier, edges me slightly ahead of the reigning DX9-leader, gothic860's Lenovo W530 i7-3720QM + [email protected]. Proof below:

3dmark06: 26121 - NVIDIA GeForce GTX 660 Ti video card benchmark result - Intel Core i7-3740QM Processor,Hewlett-Packard 17DF

RE5-dx9-var: 228.9

unfortunately It doesn't mean anything. on Polish board we found many games where PCI-E BUS limits many games! in very strong way! it is little bit curious if anybody didn't realized this (didn't make a games comparison) yet. We made it and limitation in games is really visible, games a completly different environment than 3dmarks.

btw, in elitebook 820 G1 manuals we can find something like this:

12.5” HD8 Premium AG UWVA slim (1366 x 768), so it means there will be IPS option, probably something like in X220,X230,X240. another thing is optional backlit keyboard. It's interesting if screen panel and keyboard (rather poor chances) will fit to 2570p.

[jacobsson]

..On Polish board we found many games where PCI-E BUS limits many games! ...We made it and limitation in games is really visible...

Could you please link to these game limitation comparisons?

I agree that this is an important issue that needs to be benched and discussed.

[bjorm]

Could you please link to these game limitation comparisons?

I agree that this is an important issue that needs to be benched and discussed.

There is a lot of information, lot of plots, configurations etc. We are going (with @jot23 )to collect it and post it, but there is a lot of work ahead of us.

[ha1o2surfer]

So in case anyone is wondering... Here are 2 different benchmarks causing the CPU to draw significantly more power than the other..

LinX: imgur: the simple image sharer

TSbench: imgur: the simple image sharer

Notice the multipliers.. they are EXACTLY the same. I didn't leave them running for very long because no matter how much time into the benchmark TS bench never uses more than 34-36watts. This is despite my TDP being unlocked...

[ha1o2surfer]

There is a lot of information, lot of plots, configurations etc. We are going (with @jot23 )to collect it and post it, but there is a lot of work ahead of us.

One game that hates low bandwidth links is Assassin's Creed 4..

[bjorm]

One game that hates low bandwidth links is Assassin's Creed 4..

GTA4, Crysis3, Civ5, BF3 Multi probably as well.

[Tech Inferno Fan]

An inexpensive and easy way of improving the 2560P/2570P cooling system without modding/glueing stock heatsink/heatpipe

Jacobsson and I have been discussing this privately. Might as well make it public.

I found a very easy way of improving cooling without glueing or modding the existing heatpipe/heatsink. Just sandwich a 25x25mm (or maybe 30x30 - have to check, 25mm works coz I have one in now .. also need to get thickness right.. 30mm might need to be drilled to pass the screwholes) copper plate between the heatsink and the CPU. No glue necessary just dab on some thermal paste on the CPU and then use the heatsink to hold it down. The extra copper drops temps nicely. Something I found simply by holding two pieces on top of the original heatsink.

Better yet would be one that is rectanguler that fits under the heatsink but extends out and over the chassis towards the fan like shown in the representation below, though haven't found one on ebay like that. That would soak cpu heat and get some cooling as well.

For your consideration:

5X 25mm 25mm 0 5mm Heatsink Copper Pad Shim for Xbox 360 PS3 GPU VGA | eBay (25x25x0.5)

5 Laptop GPU CPU Heatsink Copper Shim 25mmx25mmx0 8mm | eBay (25x25x0.8)

5pcs Laptop GPU CPU Heatsink Copper Shim 25mmx25mmx1mm | eBay (25x25x1.0)

5 Laptop GPU CPU Heatsink Copper Shim 25mmx25mmx1 2mm | eBay (25x25x1.2)

5X 30mm 30mm 0 8mm Heatsink Copper Pad Shim for Dell XPS | eBay (30x30x0.8)

Laptop Heatsink Copper Shim 3cmx3cm Thickness 1 0mm | eBay (30x30x1.0)

Laptop Heatsink Copper Shim 3cmx3cm Thickness 1 2mm | eBay (30x30x1.2)

5PCS Graphics Laptop CPU GPU Heatsink Copper Shim 42mmx42mmx1mm | eBay (42x42x1.0)

Jacobsson mentioned he'll see if a custom copper shim specifically for the 2560P/2570P, sized correctly with screwholes in place could be made. Nice if it leads to an ebay product.

The 'sandwich of copper shims' would do even better if some small holes were drilled on the bottom cover just underneath the fan to bring in some cool air rather than recirculate air within the chassis. Recommend to insulate the bottom-most layer of the copper sandwich with say black electrical tape as a preventative measure. It it ever perhaps to fold over from heat then it won't short anything on the systemboard.

[jacobsson]

I've ordered the 5 pieces of 42x42x1mm, so let's make a metal sandwich guys! =)

[bjorm]

http://forum.techinferno.com/diy-e-gpu-projects/2109-diy-egpu-experiences-%5Bversion-2-0%5D-234.html#post76185

there You have compariosn in BF4 with i5-3230M vs i7-3632QM. but really interesting thing Ypu have here, for the first time I tried Gelid GC Extreme.

As for 17minutes of almost 100% core load, temperatures are silly low even for i5

[Aikimox]

See there's been some progress here! A couple of things:

The CPU is definitely limited at about 35W, hence all high end quads show reduced clocks. Your processors are not faulty, just locked at reduced clocks by 100-200mhz. The good thing is you still benefit from the relatively high clocks, extra threads and cache.

Also, the reason there's a temp drop when putting a cooper shim between the cpu and the hs is because of the increased pressure. You could also put thin washers on the HS screws and get better results without using any shims, the fewer layers you have between the cpu and pipes - the better are your temps. I was gonna do that but my current mod is adequate to control the temps as is, even at 30c room temp. But if you are aiming at the perfection:

1)put washers on the hs screws to increase the pressure between the cpu and HS.

2)glue extra copper on top of the hs. (if you can mod an extra pipe in such a way that it would get enough surface contact with the radiator, would be legenDARY!!)

3)drill holes in the back cover to improve air circulation.

4)* If you don't mind the noise, max out the fan whenever benching or gaming.

[Tech Inferno Fan]

The CPU is definitely limited at about 35W, hence all high end quads show reduced clocks. Your processors are not faulty, just locked at reduced clocks by 100-200mhz. The good thing is you still benefit from the relatively high clocks, extra threads and cache.

Yep. Got a second i7-3740QM. I can confirm the CPU will decrease the multiplier to keep TDP < 38W. This second one runs x33-4C mode due to having higher voltages across the board. My previous one was more efficient and could hold x34-4C under full load.