20x Core i7 3770K IMC overclocking test on air

[Massman]

Hey all,

 

After the SF3D OC Gathering Vol.X binning event, I quickly tested the 20 Core i 7 3770K chips I brought over for IMC overclocking capability on air cooling.

 

Test system:

 

- Core i7 3770K (obviously)

- Simple air cooler with 1 fan

- GIGABYTE Z77X-UP5 TH

- GSKILL TridentX DDR3-2666C11

 

There are four states:

 

- "ok" means the CPU can boot up with that memory frequency

- "66" means the CPU cannot boot up with that memory frequency

- "51" means the memory was not capable of that frequency (like "55" on M5G)

- "/" means untested setting

 

Of the 20 tested CPUs ...

 

... 20 passed DDR3-2400 test

... 19 passed DDR3-2600 test

... 18 passed DDR3-2666 test

... 13 passed DDR3-2800 test

... 2 passed DDR3-2933 test

... 0 passed DDR3-3000 test

 

I've verified the testing on the M5G, the results were of course the same. I also tested the 2933 CPUs with memory under cold and then passing DDR3-2933 was no problem anymore, which indicates that "51" was related to the memory rather than the CPU capability.

 

The chart:

 

[ryba AICooling]

So boths cpus what I tested in Finland could make 2933 What a skill in binning....

[Massman]

So boths cpus what I tested in Finland could make 2933 What a skill in binning....

 

Just proves you don't have to be pretty to be lucky!

[TaPaKaH]

that's extremely lucky, only one chip out of 20 I tried could work at 1333

[Hiwa]

tests are 32m?

 

a lot of cpus can do 2933 or so but can't pass 32m even at 2800

 

Good job Pieter.

[Massman]

that's extremely lucky, only one chip out of 20 I tried could work at 1333

 

New or old batch?

 

tests are 32m?

 

a lot of cpus can do 2933 or so but can't pass 32m even at 2800

 

Good job Pieter.

 

Boot up only, didn't have time to go in OS and check stability.

[Hiwa]

New or old batch?

 

 

 

Boot up only, didn't have time to go in OS and check stability.

 

What setting u used on UD5H ? can u share bios settings?

[Splave]

sucks that there is soooo much freakin variability in chips these days,,,,need to bin Core mhz, then pray it has a decent IMC....so lame

[Massman]

I used the new UP5 TH board, not the UD5H .

 

What BIOS setting are you looking for specifically? Already flashed the BIOS of the board, so the profile I used has been overwritten.

[Hiwa]

I used the new UP5 TH board, not the UD5H .

 

What BIOS setting are you looking for specifically? Already flashed the BIOS of the board, so the profile I used has been overwritten.

 

any difference btw those 2 boards?

 

i need for UD5H i can't do +2950 with the cpu can do 32m on Asus board.

[Hyperhorn]

Maybe Sam's "could work with" is a bit different from Massman's "can boot up". So it would be interesting to know how big the difference from boot-stable to bench-stable (to rock-stable) is.

[ryba AICooling]

Guys, Massman was testing cpus at AC. That kind of memory freq. need cpu wil sub-zero as well.

[TaPaKaH]

New or old batch?

3217-3219 Costa Ricas

Maybe Sam's "could work with" is a bit different from Massman's "can boot up". So it would be interesting to know how big the difference from boot-stable to bench-stable (to rock-stable) is.

my best 3570K did 1395 suicide (boots 2666 but doesn't boot 2800/2933), 1380 32M and 1333+ memtest

The chip Christian Ney sent me does 1475 suicide (boots 2666/2800/2933), does 1400 32M may be one time out of five and HCI memtest at 1333 gives me a BSOD.

Not board's or memory's fault, tested with multiple.

 

Either way, unless you're a mem reviewer or like making those Hynix/Samsung 1400MHz+ e-peen threads, you don't need a CPU that does more than 1350 (for BBSE and PSC).

[M.Beier]

Pretty decent chips, but unfortunately there is a huge difference from bootable, to fully stable..

 

DDR3-2666 is quite good for fully stable, I've only had one Ivy that can do that...

 

- Now the funny thing, the ES Intel sends out for reviewers, pretty much every CPU can do DDR3-3000 boot

 

Guys, Massman was testing cpus at AC. That kind of memory freq. need cpu wil sub-zero as well.

 

Ivy IMC does not really impact that much from cooling.

[IanCutress]

My CPU (Malay ES) with that kit can boot 2800 in a P8Z77-V Pro, but can't in a UD3H.

[Massman]

I did do random tests with some CPUs booting at 2800 to see if they run 32M. Never had a situation where a booting 2800 wouldn't run 32M stable.

[M.Beier]

My CPU (Malay ES) with that kit can boot 2800 in a P8Z77-V Pro, but can't in a UD3H.

 

By any chance can you try it in a MV G/E/F ?

[Leeghoofd]

- Now the funny thing, the ES Intel sends out for reviewers, pretty much every CPU can do DDR3-3000 boot

 

Then we got the 3 duds then Marc. Our best one was the 3570K stable at 6400 2800Mhz 32M, 2933 not in a million lightyears

 

Both the ES 3770K didn't even boot at 2800 on the Gene V... 23 FTW

 

I know some Giga related people have good IMC capable CPUs, wonder how they did the binning.

 

My retail was busy with 32M 7th loop on air, with the TridentX kit at -20°C, rebooted for more MHz ( from 4 - to 5ghz ) 55555555555555555555555,23333333333 post code errors looooooooool

[Leeghoofd]

My CPU (Malay ES) with that kit can boot 2800 in a P8Z77-V Pro, but can't in a UD3H.

 

loosen tRRD, tRRSR and tWWSR, it's too tight on Gigabyte. Boards should be able to do the same as the ASUS ones...

[TaPaKaH]

- Now the funny thing, the ES Intel sends out for reviewers, pretty much every CPU can do DDR3-3000 boot

I had enough of Intel's ES Ivies to claim the opposite - their are not superior to retails.

[riska]

I have had 2 malay that can both do more than 2900mhz mem speed on water stable and more on cold

 

The one i have now can do 2850mhz+ on the tridentx 2666 kit at stock voltage

[Massman]

I had enough of Intel's ES Ivies to claim the opposite - their are not superior to retails.

 

Yup, I can agree with that as well. No way Intel is shipping "pre-binned" CPUs.

 

In fact, from the 50 chips we tested at the SF3D OC Gathering 10 (retail) came straight from Intel. They were the most shit ones we had! So either Intel's not doing any screening whatsoever or they are doing it completely wrong

[IanCutress]

By any chance can you try it in a MV G/E/F ?

 

Not got those in, sorry

 

loosen tRRD, tRRSR and tWWSR, it's too tight on Gigabyte. Boards should be able to do the same as the ASUS ones...

 

That's for Gigabyte to decide on how they recognise their kits. Shouldn't have to go around fiddling with settings 99.99% of people don't understand. Set the strap and go - if the board works then great, if not then too bad, really. I do sense the folly of saying that on an OC forum, but as I review from the perspective of the 'normal enthusiast' user, we're lucky XMP is even set sometimes, let alone XMP then strap as required.

 

Let's be honest though, GSkill isn't a preferred vendor for Gigabyte. Corsair and Kingston get top priorities, then it filters down. Don't get me wrong, they still work with each other, but in general the later the BIOS the better it seems to work with my kits. In my experience, anyhow - YMMV

[sin0822]

nice testing massman.

[I.M.O.G.]

That's for Gigabyte to decide on how they recognise their kits. Shouldn't have to go around fiddling with settings 99.99% of people don't understand. Set the strap and go - if the board works then great, if not then too bad, really. I do sense the folly of saying that on an OC forum, but as I review from the perspective of the 'normal enthusiast' user, we're lucky XMP is even set sometimes, let alone XMP then strap as required.

 

Let's be honest though, GSkill isn't a preferred vendor for Gigabyte. Corsair and Kingston get top priorities, then it filters down. Don't get me wrong, they still work with each other, but in general the later the BIOS the better it seems to work with my kits. In my experience, anyhow - YMMV

 

The rule of benchmarks prevails... In motherboard reviews where benchmarks are heavily utilized by readers, it gives Gigabyte a perceived "efficiency" advantage because they tighten some things up more than other boards. In most reviews, the reviewer and reader pays no attention to the resulting sub-timings - they assume RAM timings are detected by SPD and set identically. So in a motherboard review where a board is put head to head against a different board, the variance between products is typically 1-3% with all other factors identical (within margin of error)... But a little tweaking like this can give a perceived edge to one board in the benchmark results.

 

Gigabyte isn't the first to do it. With enough memory module testing while in engineering/bios development, most popular memory will just work and no one pays any mind... The only reason anyone is talking about manually changing these settings on Gigabyte is because some DIMMs weren't tested enough so the auto-settings don't work on some DIMMs at top-end frequencies, which lead to some people figuring out what was up.

 

Finally, your point doesn't entirely make sense to me. The sticks do just work without fiddling with settings that 99.99% of people don't understand. It's only when pushing aggressive memory frequencies that are pointless for anything other than benchmarking itself that fiddling is required - and fiddling is exactly what we're benchmarking for.

Edited July 25, 2012 by I.M.O.G.