Haswell overclocking discussion.

[Rvenger]

Maybe I'm too tired, but I'm not even really sure what we're discussing!

Anyways, I'm not even sure what to think. It's pretty clear the IBT linpack isn't hitting Haswell the way it could be, neither is anything else really. So what I'm guessing is that you are saying these stress tests we're using which were designed for older systems aren't optimized for Haswell, or perhaps even older uarchs and that we're kidding ourselves?

At least that's what I'm thinking, I ran the latest Intel linpack.. Jesus H Christ.

Stock: (Kept opening the file to make sure it was doing something, ops)

=================== Timing linear equation system solver ===================
 
Size   LDA    Align. Time(s)    GFlops   Residual     Residual(norm) Check
1000   1000   4      0.026      26.0931  1.002198e-012 3.417754e-002   pass
1000   1000   4      0.008      83.1892  1.002198e-012 3.417754e-002   pass
1000   1000   4      0.008      85.3023  1.002198e-012 3.417754e-002   pass
1000   1000   4      0.008      86.6677  1.002198e-012 3.417754e-002   pass
2000   2000   4      0.048      112.1212 4.327261e-012 3.764187e-002   pass
2000   2000   4      0.050      106.2897 4.327261e-012 3.764187e-002   pass
2000   2000   4      0.061      87.4023  4.327261e-012 3.764187e-002   pass
2000   2000   4      0.049      109.7069 4.327261e-012 3.764187e-002   pass
3000   3000   4      0.140      128.7982 9.653528e-012 3.717342e-002   pass
3000   3000   4      0.176      102.2257 9.653528e-012 3.717342e-002   pass
3000   3000   4      0.140      128.6445 9.653528e-012 3.717342e-002   pass
3000   3000   4      0.140      129.1520 9.653528e-012 3.717342e-002   pass
4000   4000   4      0.314      135.8601 1.565487e-011 3.412126e-002   pass
4000   4000   4      0.313      136.2905 1.565487e-011 3.412126e-002   pass
4000   4000   4      0.312      136.9157 1.565487e-011 3.412126e-002   pass
4000   4000   4      0.311      137.1751 1.565487e-011 3.412126e-002   pass
5000   5000   4      0.590      141.2435 2.512907e-011 3.504046e-002   pass
5000   5000   4      0.592      140.8687 2.512907e-011 3.504046e-002   pass
5000   5000   4      0.590      141.3670 2.512907e-011 3.504046e-002   pass
5000   5000   4      0.592      140.9005 2.512907e-011 3.504046e-002   pass
10000  10000  4      4.280      155.7971 1.012627e-010 3.570624e-002   pass
10000  10000  4      4.275      155.9875 1.012627e-010 3.570624e-002   pass
15000  15000  4      15.932     141.2575 1.979849e-010 3.118294e-002   pass
15000  15000  4      16.082     139.9325 1.979849e-010 3.118294e-002   pass
20000  20000  4      32.094     166.2039 3.564028e-010 3.154945e-002   pass
20000  20000  4      32.116     166.0878 3.564028e-010 3.154945e-002   pass
25000  25000  4      63.792     163.3114 6.220813e-010 3.537553e-002   pass
25000  25000  4      64.971     160.3481 6.220813e-010 3.537553e-002   pass
30000  30000  4      117.158    153.6544 9.270371e-010 3.654390e-002   pass

4.8GHz 3.8GHz Uncore - shut my motherboard down, the heat was insane. There is no way my cpu cooler can cope with this...

=================== Timing linear equation system solver ===================
 
Size   LDA    Align. Time(s)    GFlops   Residual     Residual(norm) Check
1000   1000   4      0.027      24.4817  1.002198e-012 3.417754e-002   pass
1000   1000   4      0.006      116.9661 1.002198e-012 3.417754e-002   pass
1000   1000   4      0.006      113.6987 1.002198e-012 3.417754e-002   pass
1000   1000   4      0.006      115.8336 1.002198e-012 3.417754e-002   pass
2000   2000   4      0.035      150.8695 4.327261e-012 3.764187e-002   pass
2000   2000   4      0.035      151.3860 4.327261e-012 3.764187e-002   pass
2000   2000   4      0.034      155.2040 4.327261e-012 3.764187e-002   pass
2000   2000   4      0.036      149.4019 4.327261e-012 3.764187e-002   pass
3000   3000   4      0.104      173.0388 9.653528e-012 3.717342e-002   pass
3000   3000   4      0.107      167.7035 9.653528e-012 3.717342e-002   pass
3000   3000   4      0.104      173.6963 9.653528e-012 3.717342e-002   pass
3000   3000   4      0.103      174.1454 9.653528e-012 3.717342e-002   pass
4000   4000   4      0.229      186.4405 1.565487e-011 3.412126e-002   pass
4000   4000   4      0.231      184.9304 1.565487e-011 3.412126e-002   pass
4000   4000   4      0.229      186.1771 1.565487e-011 3.412126e-002   pass
4000   4000   4      0.229      186.1232 1.565487e-011 3.412126e-002   pass
5000   5000   4      0.433      192.4411 2.512907e-011 3.504046e-002   pass
5000   5000   4      0.434      192.2064 2.512907e-011 3.504046e-002   pass
5000   5000   4      0.433      192.6871 2.512907e-011 3.504046e-002   pass
5000   5000   4      0.434      192.0620 2.512907e-011 3.504046e-002   pass
10000  10000  4      3.169      210.4104 2.977802e-008 1.050003e+001   FAIL
10000  10000  4      3.158      211.1710 9.805032e-011 3.457353e-002   pass

211 flipping GFLOPs? Are you kidding me?!?! This is with it throttling and failing!

4.6GHz core 4.6GHz Uncore

=================== Timing linear equation system solver ===================
 
Size   LDA    Align. Time(s)    GFlops   Residual     Residual(norm) Check
1000   1000   4      0.032      21.0860  1.002198e-012 3.417754e-002   pass
1000   1000   4      0.006      112.8623 1.002198e-012 3.417754e-002   pass
1000   1000   4      0.006      114.3003 1.002198e-012 3.417754e-002   pass
1000   1000   4      0.006      114.0276 1.002198e-012 3.417754e-002   pass
2000   2000   4      0.042      128.4695 4.327261e-012 3.764187e-002   pass
2000   2000   4      0.036      149.8790 4.327261e-012 3.764187e-002   pass
2000   2000   4      0.036      148.4749 4.327261e-012 3.764187e-002   pass
2000   2000   4      0.050      106.6145 4.327261e-012 3.764187e-002   pass
3000   3000   4      0.107      167.7921 9.653528e-012 3.717342e-002   pass
3000   3000   4      0.108      167.2586 9.653528e-012 3.717342e-002   pass
3000   3000   4      0.107      167.7909 9.653528e-012 3.717342e-002   pass
3000   3000   4      0.109      165.8319 9.653528e-012 3.717342e-002   pass
4000   4000   4      0.236      181.2239 1.565487e-011 3.412126e-002   pass
4000   4000   4      0.235      181.8316 1.565487e-011 3.412126e-002   pass
4000   4000   4      0.238      179.4995 1.565487e-011 3.412126e-002   pass
4000   4000   4      0.237      179.8140 1.565487e-011 3.412126e-002   pass
5000   5000   4      0.446      186.8200 2.512907e-011 3.504046e-002   pass
5000   5000   4      0.445      187.3511 2.512907e-011 3.504046e-002   pass
5000   5000   4      0.445      187.2766 2.512907e-011 3.504046e-002   pass
5000   5000   4      0.444      187.6106 2.512907e-011 3.504046e-002   pass
10000  10000  4      3.313      201.2619 1.012627e-010 3.570624e-002   pass
10000  10000  4      3.233      206.2836 1.012627e-010 3.570624e-002   pass
15000  15000  4      11.923     188.7423 1.979849e-010 3.118294e-002   pass
15000  15000  4      11.838     190.1115 1.979849e-010 3.118294e-002   pass
20000  20000  4      24.219     220.2491 3.564028e-010 3.154945e-002   pass

220 GFLOPS D: D: D: D: D:

Chip hit 98C and I closed it, I'm not that crazy!


I don't even know what to think. On the one hand, clearly old tools are ineffectual (all of them), on the other hand what in the heck do I need to test this for? I had a problem with the old tests, they weren't representative of actual loads I used so they made little sense outside of some people wanting to tout their cooling solution... But this?

lolz...

Anyways I'm going to bed, maybe I'll dream happy thoughts where I wake up and every game is optimized like this and I can downclock to 3GHz and get more performance than I was getting at 5.3GHz three hours ago.

I guess my 73c stock temps on IBT are about right then? lol

 

[Face2Face]

Don't forget to check for WHEA errors, they don't (always) cause BSODs so they can be occurring even if you think you are stable. (See sig for simple app to quickly check)

Nice sales pitch there
I do find your program quite useful, thanks for posting it.

 

[wand3r3r]

Nice sales pitch there
I do find your program quite useful, thanks for posting it.

Yeah I just saw a perfect opportunity for a perfect use case.

Btw, I gain nothing from the use of it, just being helpful being it exists.

 

[NFarnzy]

I did drop the uncore down a little.. gonna drop it down some more and see what happens..I heard tho you was supposedly to keep the uncore between 100-300mhz of the actual speed.. Maybe I am wrong but I thought that is what JJ from Asus was saying in his video.

 

[BallaTheFeared]

I just wanted to make sure people weren't kidding themselves when they pick and choose which "stress tests" they are using by how much >>Cooler<< they can show their processor as being by running an >>Inferior<< stress tester.



The whole point of running stress tests is so that you don't have to run every single possible combination of factors in every single one of your software in every scenario yourself.
If your stress testing application doesn't seem to actually be stressing your hardware, its probably because it's not, especially if there are other stress test applications that stress it more.

All previous gen stress tests are giving similar temps, which are 20-25C hotter than non avx (pretty much everything we currently use)tests.

At some point, doesn't it become a bit silly? 220 flops is over twice as fast as a 980x at 4.2GHz.

Also you'll notice from 3.4/3.4 to 4.8/3.4 the cache even with Intels linpack wasn't holding the flops back if at all

But I digress, to each their own. If anyone is interested in downloading Intel's latest Linpack which will utterly destroy anything you can get with Prime95, AIDA, LinX, or Intel Burn Test here is the link: http://software.intel.com/en-us/articles/intel-math-kernel-library-linpack-download

I'd actually be interested in seeing a comparison between Ivy/Sandy at stock 3.4/3.5 vs Haswell, Face2Face?

 

[Face2Face]

I'd actually be interested in seeing a comparison between Ivy/Sandy at stock 3.4/3.5 vs Haswell, Face2Face?

No problem - Just waiting for my new/used mobo - My current one is flaky with IVY -Fine with Sandy. New/used board should be here today, I can make some comparisons this weekend. If you are wanting something sooner, maybe someone else can chime in. I can also do some 4.8Ghz comparisons and maybe the magic 5 as well
What are you wanting to compare? FLOPS?

 

[BallaTheFeared]

No problem - Just waiting for my new/used mobo - My current one is flaking out on me. New/used board should be here today, I can make some comparisons this weekend. If you are wanting something sooner, maybe someone else can chime in. I also can do some 4.8Ghz comparisons and maybe the magic 5 as well
What are you wanting to compare?

Well currently the question is really if the current stress test tools we've used for several generations are actually "stressing" the processors or not.

So unless someone has something better, the Intel optimized linpack benchmark seems to be the way to go. It's putting out considerably more heat, drawing more power, and producing far higher GFLOPs than other tests are (the ones the measure at least).

Intel® Optimized LINPACK Benchmark is a generalization of the LINPACK 1000 benchmark. It solves a dense (real*8) system of linear equations (Ax=b), measures the amount of time it takes to factor and solve the system, converts that time into a performance rate, and tests the results for accuracy. The generalization is in the number of equations (N) it can solve, which is not limited to 1000. It uses partial pivoting to assure the accuracy of the results.

Do not use this benchmark to report LINPACK 100 performance because that is a compiled-code only benchmark. This is a shared-memory (SMP) implementation which runs on a single platform. Do not confuse this benchmark with:

MP LINPACK, which is a distributed memory version of the same benchmark.
LINPACK, the library, which has been expanded upon by the LAPACK library.

Intel provides optimized versions of the LINPACK benchmarks to help you obtain high LINPACK benchmark results on your genuine Intel processor systems more easily than with the High Performance Linpack (HPL) benchmark. Use this package to benchmark your SMP machine.

Also I added some utilities to the OP, if anyone has other suggestions just link to the download area of the website and I'll be happy to add it to the OP. :thumbsup:

 

[BallaTheFeared]

I guess my 73c stock temps on IBT are about right then? lol

IBT at stock is probably 100 GFLOPs, whereas this is 150 GFLOPs.

Something like 4.6GHz in IBT is 126~ GFLOPs whereas in this it's 220 GFLOPs...

 

[Face2Face]

I will try this Intel optimized linpack benchmark. I have used all of them and I always found IBT to generate the most heat, but I never kept track of the GFLOPS. This is a good find, I will try this out when I get some time this weekend. My board just got here and my H100i is coming tomorrow.

 

[JoeRambo]

The new Linpack is the same as old Linpacks, except it is compiled and optimized for Haswell. Key optimization in it is FMA support, that can double FLOPS if not bottlenecked by bandwith or some other resource.

So Linpack "evolution" as exposed for example by Linx was - without AVX ~65GFLOPS, wAVX - ~125GFLOPS and now with new version of Linpack and FMA support - ~250GFLOPS.

Needless to say, to move things @250GFLOPS requires a truckload of bandwith and execution resources, end result is HEAT

 

[AkumaX]

If it's bouncing like that it's because you're using offset voltage and running an avx instruction.

If you use fixed while doing your testing it won't bounce, even 1.12v could be excessive for 4GHz. You should try toning that back, while using fixed voltage. Also make sure you're setting your Uncore to x38 as to not infer any sort of added chance of failure from different aspects of the overclock you aren't aware of.

Eureka has a ASRock board and his is automatically keeping his uncore (cache ratio) in a 1:1 ratio with his core.

Also something else to consider, if you aren't going to delid is using a more practical method of testing for your needs and cooling solution. Haswell will throttle so there is no risk of instability in non avx programs if you're getting too hot, assuming of course you don't currently plan to run avx based programs in your daily tasks.

I learned a lot of new terms today

However, a lot of these terms have different names in the motherboard I'm using (MSI Z87), and I'm trying to make sense of it...

It was much easier o/cing SB/IB since I literally only changed 1 setting - the CPU mult. Never even touched any of that other stuff.

But anyways, messing with the BIOS I have CPU Multiplier Ratio and CPU Ring Ratio. I don't remember what I messed with, but now CPU-Z 1.64.3 says 800MHz @ 0.7V idle. Yay! Still can't go past 4.0GHz...

My BIOS has no mention of uncore or vcore offset. So again, stock voltages for everything.

 

[Elfear]

Balla, kudos on putting together a Haswell overclocking guide. Even though I probably won't be getting one I appreciate the time you spent putting everything into one thread.

 

[BullHorn]

I'm trying to understand how to make use of Asus' Adaptive CPU voltage feature but it just keeps BSODing. Using the Manual voltage feature is unacceptable, it's like revving your car engine to 100% while standing at a red light.

 

[BallaTheFeared]

I learned a lot of new terms today

However, a lot of these terms have different names in the motherboard I'm using (MSI Z87), and I'm trying to make sense of it...

It was much easier o/cing SB/IB since I literally only changed 1 setting - the CPU mult. Never even touched any of that other stuff.

But anyways, messing with the BIOS I have CPU Multiplier Ratio and CPU Ring Ratio. I don't remember what I messed with, but now CPU-Z 1.64.3 says 800MHz @ 0.7V idle. Yay! Still can't go past 4.0GHz...

My BIOS has no mention of uncore or vcore offset. So again, stock voltages for everything.

Which board do you have?

Here are my settings for 4.8/4.7 testing, this is the overclock I showed in the first post.

In screen 2 it shows your voltages.

You can set these manually.

CPU Core Voltage Mode - Override = Fixed, it stays at what you set regardless of if you're at idle or load. This is good for testing AVX code because it won't bump the voltage past what you set, it's not great for 24/7 because it won't drop at idle.

I leave offset alone, I just set what vcore I want - that's it. The CPU handles the rest.

CPU Ring Voltage - Same thing as core. After you get stable you can switch these to Dynamic, so it will drop voltage at idle. However remember anything running avx will boost your voltage past what you've designated!

That's all I change.

CPU Ratio; Ring Ratio; Dram and their voltages.

Everything else is handled by the board.

Edit:

CPU Ratio mode controls if the cpu downclocks or not, CPU Ring/Core Voltage mode controls if the voltage is fixed, or if it is dynamic (locks in, or adjusts based on load/cpu speed).

Dynamic is best used after you've set your overclock, it will overshoot your voltage in AVX there is no fix for this that I'm aware of. Override or "fixed" is what should be used when you're testing for stability at a given overclock, this gives you fine grain control over the voltage explicitly.


You might just have a bad chip, there isn't much you can do about that.

If I were you I'd start off setting core ratio to 38 and ring to 38, then setting the voltage on ring to auto and on core to 1.0v (leaving everything else on auto, offset, LLC, everything) then go test, if you pass, bump core to 39 and keep going - only change the core clock ratio and voltage, but don't get ahead on core voltage try to keep it as low as you can and don't increase it much if you fail before you retest. Once you fail, bump core up 10 to 15mv and retest.

 

[cytg111]

I dont own a haswell, but i'd like data on how well you all do on specific motherboards and bios revisions.. imo that is a big leftout on the haswell oc-potential.

 

[pm]

I have an Asus Z87 Plus (stock BIOS) and a 4770K and I just changed all of the multipliers to 45, changed my memory to XMC left everything else stock, and was able to run IBT and Aida (although temps were just shy of 100C). About the only problem is that it didn't like running 2 sticks of my GSkill Ripjaws at their XMC profile so I ended up getting memory errors on those but I removed them and the errors went away. 4.5GHz seems fine but I've been mentally preparing myself to put my precious CPU into a vice and whack it with a hammer by watching YouTube videos because the next step is to de-lid it.

 

[cytg111]

I have an Asus Z87 Plus (stock BIOS) and a 4770K and I just changed all of the multipliers to 45, changed my memory to XMC left everything else stock, and was able to run IBT and Aida (although temps were just shy of 100C). About the only problem is that it didn't like running 2 sticks of my GSkill Ripjaws at their XMC profile so I ended up getting memory errors on those but I removed them and the errors went away. 4.5GHz seems fine but I've been mentally preparing myself to put my precious CPU into a vice and whack it with a hammer by watching YouTube videos because the next step is to de-lid it.

I wouldnt ...

http://forums.anandtech.com/showpost.php?p=35151079&postcount=976

 

[ilogik]

I pushed the OC Genie button on my Z87 Mpower max which OC'd the chip to 4.2 and I've been runnin on that since day 1

 

[BullHorn]

I'm on Adaptive voltage mode yet as you can see the yellow graph, the voltage is constantly at the maximum. Are you guys having a similar issue?

My guess is that as long as the CPU is maintaining a frequency higher than default, it uses the boosted voltage. The problem is that it always maintains the high frequency so eventually, Adaptive mode and manual/locked voltage mode is the same.

Right?..

-EDIT: DISREGARD THIS. My Windows 7 Power Plan was on Performance. Once I changed it to Balanced everything sorted itself.

 

[BallaTheFeared]

Depending on what your board calls it, there should be a "CPU Ratio Mode" which can either be fixed, which is what you're probably using or "adaptive/dynamic" which would allow it to downclock.


If you already have those set, make sure your windows power options allow the cpu to operate at a lower frequency.

 

[Kougar]

http://www.overclock.net/t/1401976/the-gigabyte-z87-haswell-overclocking-oc-guide

This might be worth a read for those interested in this thread.

Sin's guide is a big must read for all GB users.

Bumping my ringbus voltage from 1.12 to 1.2v not only got 4.2ghz finally WHEA error free, but I was able to drop the vcore to 1.3v and still get the chip stable at 4.3ghz too. 4.4 was not stable even with 1.32vcore, but temps are already nearing limits for useful loads anyway. Runs the uncore at 3.9Ghz all day long though now.

Don't forget to check for WHEA errors, they don't (always) cause BSODs so they can be occurring even if you think you are stable. (See sig for simple app to quickly check)

This needs to be quoted again just so people see it. Coming from a i7 920 I wasn't aware WHEA errors like this existed and it was chance that I stumbled into this one given the system kept running fine and tested fine initially. Got that WHEA error at the same time as Civ V crashed, so they have very real implications for stability....

 

[BallaTheFeared]

Added two more written guides and two more videos, one more video from JJ (Newegg) and one from Linus.

Also added a bit of a description to Wand3r3r's program so it wasn't lost in the wall of text :thumbsup:

 

[Face2Face]

The new Linpack is the same as old Linpacks, except it is compiled and optimized for Haswell. Key optimization in it is FMA support, that can double FLOPS if not bottlenecked by bandwith or some other resource.

So Linpack "evolution" as exposed for example by Linx was - without AVX ~65GFLOPS, wAVX - ~125GFLOPS and now with new version of Linpack and FMA support - ~250GFLOPS.

Needless to say, to move things @250GFLOPS requires a truckload of bandwith and execution resources, end result is HEAT

So did this debunk the need to do a GFLOPS comparison between Ivy and Haswell?

 

[24601]

Just a heads up.


Ambient 94F = 34.4444C
Thermalright True Spirit 140 in Antec 1100 case
4.6ghz 3570k 1.296v Maximum Temp; Coolest Core: 93C (Delta T:58.5556), Hottest Core: 103C (Delta T:68.5556)

Intel(R) Optimized LINPACK Benchmark data

Current date/time: Fri Jun 21 09:06:43 2013

CPU frequency:    4.596 GHz
Number of CPUs: 1
Number of cores: 4
Number of threads: 4

Parameters are set to:

Number of tests: 12
Number of equations to solve (problem size) : 1000  2000  3000  4000  5000  10000 15000 20000 25000 30000 35000 40000
Leading dimension of array                  : 1000  2000  3000  4000  5000  10000 15000 20000 25000 30000 35000 40000
Number of trials to run                     : 4     4     4     4     4     2     2     2     2     1     1     1    
Data alignment value (in Kbytes)            : 4     4     4     4     4     4     4     4     4     4     4     4    

Maximum memory requested that can be used=4210869504, at the size=40000

=================== Timing linear equation system solver ===================

Size   LDA    Align. Time(s)    GFlops   Residual     Residual(norm) Check
1000   1000   4      0.008      83.0541  1.029343e-012 3.510325e-002   pass
1000   1000   4      0.009      70.5085  1.029343e-012 3.510325e-002   pass
1000   1000   4      0.008      84.7495  1.029343e-012 3.510325e-002   pass
1000   1000   4      0.015      44.4607  1.029343e-012 3.510325e-002   pass
2000   2000   4      0.066      80.3683  4.298950e-012 3.739560e-002   pass
2000   2000   4      0.066      80.6317  4.298950e-012 3.739560e-002   pass
2000   2000   4      0.076      70.7375  4.298950e-012 3.739560e-002   pass
2000   2000   4      0.070      75.9826  4.298950e-012 3.739560e-002   pass
3000   3000   4      0.215      83.6154  8.755385e-012 3.371489e-002   pass
3000   3000   4      0.217      82.9458  8.755385e-012 3.371489e-002   pass
3000   3000   4      0.209      86.0551  8.755385e-012 3.371489e-002   pass
3000   3000   4      0.212      84.9031  8.755385e-012 3.371489e-002   pass
4000   4000   4      0.413      103.3760 1.896949e-011 4.134580e-002   pass
4000   4000   4      0.406      105.0830 1.896949e-011 4.134580e-002   pass
4000   4000   4      0.402      106.1255 1.896949e-011 4.134580e-002   pass
4000   4000   4      0.406      105.2547 1.896949e-011 4.134580e-002   pass
5000   5000   4      0.904      92.2299  2.581643e-011 3.599893e-002   pass
5000   5000   4      0.792      105.3461 2.581643e-011 3.599893e-002   pass
5000   5000   4      0.775      107.5885 2.581643e-011 3.599893e-002   pass
5000   5000   4      0.774      107.6761 2.581643e-011 3.599893e-002   pass
10000  10000  4      6.082      109.6394 9.603002e-011 3.386116e-002   pass
10000  10000  4      5.699      117.0143 9.603002e-011 3.386116e-002   pass
15000  15000  4      21.306     105.6241 2.042799e-010 3.217442e-002   pass
15000  15000  4      18.325     122.8052 2.042799e-010 3.217442e-002   pass
20000  20000  4      42.368     125.8991 4.097986e-010 3.627616e-002   pass

I think I may need to reapply thermal paste. I used the packet that came with the cooler when I had to remove it to change motherboards.
I know my AS5 is somewhere but I couldn't find it.

Just posting this up as a baseline for people to compare Haswell to Ivy Bridge in a standard test.

 

[crashtech]

94°F ambient?