Core i7-3770K is slower than i7-3770

[DBP]

Here is the CPU window of Activity Monitor. This part does not make sense to me as I expected Prime95 to have 8 threads and nearly 100% of the processor...not 798.1% of the processor and 11 threads.

View attachment 46152

We assume that 8 of those 11 threads are using ~100% each.

Because your machine has more than 4 virtual cores (4+HT: 8) Activity Monitor is not showing you the per-core utilisation in that window, but there are two different displays you can get of it:

• double-click on the CPU activity histogram (or select Window->CPU History), and
• select Window->Floating CPU Window->(Vertical or Horizontal).

I suggested earlier that you look at the Floating CPU Window.

It does look like your machine is running with a 4-core turbo of 39, which is not what Intel says should be the default.

If all cores were busy then I would be suspecting that your BIOS might be automatically overclocking to 39,39,39,39 instead of the default 39,38,37,37 (just as this D3H motherboard does with the 3770). If you changed the settings from Auto to these specific numbers do you get the same behaviour on your 3770K?

That way we'll know if your BIOS is doing the slight overclock by default.

 

[skaker]

We assume that 8 of those 11 threads are using ~100% each.

Because your machine has more than 4 virtual cores (4+HT: 8) Activity Monitor is not showing you the per-core utilisation in that window, but there are two different displays you can get of it:

• double-click on the CPU activity histogram (or select Window->CPU History), and
• select Window->Floating CPU Window->(Vertical or Horizontal).

I suggested earlier that you look at the Floating CPU Window.

It does look like your machine is running with a 4-core turbo of 39, which is not what Intel says should be the default.

That way we'll know if your BIOS is doing the slight overclock by default.

Any multi-threaded app will have more (sometimes many more) threads than cores it's mapped to. In the very least there will be a parent thread that spawns the individual thread, but usually there are many other operations that are not parallelized (various i/o threads for example). So the fact that it shows 11 doesn't really matter. It's mapping to 8 'cores' (which is the only way for utilization to be above 700...) which is the important issue. This is not a measure of what the thread is actually doing. You could have really poorly coded threads where they spend 99% of their time waiting for synchronization, in which case the code would be incredibly slow, but still show ~800% utilization. You can get more info from using top at the command line (will give process resource use, but also give user vs. system cpu utilization).

As for the second point, not necessarily. The auto-overclock may be lower level, i.e. it may be setting a minimum active core multiplier based on the cpu's reported highest default bin. In which case manually setting the cores to 39/38/37/37 (or for that matter anything below 39, i.e. 33/32/31/30) won't change how the processor reacts.

 

[DBP]

The auto-overclock may be lower level, i.e. it may be setting a minimum active core multiplier based on the cpu's reported highest default bin. In which case manually setting the cores to 39/38/37/37 (or for that matter anything below 39, i.e. 33/32/31/30) won't change how the processor reacts.

There's one way to tell: set it to 39/38/37/37 and see if the multiplier goes to 39x. Or are you saying that if you set the max multiplier to 33 then it will actually achieve higher? Doesn't sound right.

And what do you mean by "bins"? Chips are rated into bins at the factory based on their success/failure at various tests (chips of various "speeds" can come out of the same wafer). What other meaning of the word "bin" are you using?

 

[skaker]

No I was just saying that whatever the BIOS is doing, it may be doing it below the level of the settings. I don't know, but one approach would be to query the processors maximum turbo multiplier, and set this as the lower bound on any adjustments. In this case, manually setting the multipliers to 39/38/37/37 would still yield an all core multiplier of 39. Likewise, manually setting the cores to 33/32/31/31 would also yield an all core multiplier of 39. While it's possible that what your observing on your 3770, and what others have seen with all core turbo of 39, is a result of faulty code in the CPU, I think it's infinitely more likely it's a BIOS enhancement.

Intel used to use the 'bin' terminology in their early incarnations of turbo, and I still use it sometimes by mistake =). Initially there was a single turbo 'bin', when engaged the CPU would run at that specific multiplier. Current processors are much more flexible. While they will list different steps (i.e. 39/38/37/37), the CPU can actually make much finer adjustments. Turbo now uses TDP/thermal monitoring and can make very fine adjustments to the multiplier. Take a chip that's running at 3900 up to it's throttling temp, and the multiplier will increment lower (by very small amounts) until under the thermal limit.

You'll almost never see this on a consumer chip because the thermal throttling setpoint is so high (105 on ivybridge, 95 on SB). On server chips it's set a lot lower. For instance on a E5-26xx machine I run, the CPU throttles at 80degrees, while E5-46xx's throttle at 70 degrees. The E5 machine occasionally throttles (it's in a room that can get warm...), and I'll see the freq drop from 3000 (all core turbo speed) to 2990, 2987, 2946 (just throwing numbers out there as examples, possible numbers are basically anything...it is a stepwise effect, but the steps are very small).


EDIT: Sorry for edit, reading that not sure it was very clear. When a chip is listed as having an all core turbo multiplier of 37, that does not mean the multiplier is locked at 37, it means the maximum multiplier when all cores are active will be 37. The different 'steps' of turbo are not set multipliers, but caps on what the multiplier can be for a given profile of CPU activity. When you overcock, and change the all core turbo to 44, you are not setting the multiplier at 44, you're allowing the multiplier to go as high as 44, assuming thermals and TDP allow.

 

[Shuckey]

Sorry about that. Wrong Activity Monitor window.

So here are the default settings with the Floating Window





Here are the files with the typed in settings 39/39/38/37. HW Monitor shows turbo at 3.7...emmm

 

[DBP]

Here are the files with the typed in settings 39/39/38/37. HW Monitor shows turbo at 3.7...emmm

Bingo!

Just so everyone's clear, my analysis of this says:

1. With all cores busy the clock multiplier has gone to 37 (3.7 GHz). This matches the "37" at the end of 39/39/38/37.
2. Previously it was showing 39, when you had the multipliers set to Auto.
3. According to Intel (and hinted at by the display in the BIOS) "Auto" for each of these multipliers should be 39/39/38/37.
4. It seems that on your motherboard with that CPU when you set it to Auto it actually overclocks the CPU slightly, just as it does on my GA-Z77-D3H.

The question remains open as to whether this is special behaviour that only comes out when all four multipliers are set to Auto, whether "Auto" gets the highest previous multiplier (e.g. if you set them to 36/Auto/Auto/Auto maybe they might all end up at 36), or whatever. However I'm not sure whether anyone wants (or needs) to spend the time doing that analysis.

Can anyone fault the above reasoning that concludes that Shuckey's GA-Z77X-UD5H is slightly overclocking the i7-3770K by default?
I suppose in this case the 3770K is exactly the same speed as a 3770 would be.

 

[DBP]

Interesting. According to Intel's specs (the PDF linked earlier in this thread) the i5-3570K should have a max turbo of 38/38/37/36 (thus with 4 active cores the max clock should be 3.6 GHz, and even with only one core active it should never go above 3.8 GHz).
However I've just found a GA-Z77MX-D3H TH board with the i5-3570K when set to "Auto" is running with a max turbo of 39. This chip is also being overclocked by default settings!

So when people say that if you don't need hyperthreading you may as well get the i5 for the same power as the i7, that's true-ish (at least the 3.9 GHz part) if your board is overclocking the i5.

Sigh...

 

[jwk]

This is a great thread, I'm both glad and disappointed I ran across it! I was trying to decide what the advantages & disadvantages of the i7-3770 vs the i7-3770K were, whether VT-d in the 3770 and not in the 3770K would be an advantage or disadvantage to someone like me that hasn't used any virtualization in the past and hasn't done any over-clocking (but might possibly consider some non-agressive over-clocking) and isn't a game-player, but also thinking about future-proofing.

Some of the discussion has included the extra cost of the 3770K vs the 3770... making my decision even more difficult is that MicroCenter is selling the i7-3770 for $259.99 and the i7-3770K for $229.99- that right, the i7-3770K costs $30 less than the i7-3770! I don't really have a feel for what the long-term implications might be for VT-d on other-than virtual machines and games, what are your thoughts on whether VT-d would be implemented into apps like Adobe Creative Suite, Final Cut, Avid Media Composer, and other professional media creation applications???

 

[DBP]

So far we haven't found any OS X virtualisation software which uses VT-d. Parallels Desktop, VMware Fusion, VirtualBox: none of their OS X versions do.
Guested operating systems work fine without it (I run several Linux/BSD/Win7 guests on my CustoMacs) so I wouldn't worry about it too much.

Still waiting on someone to identify something that actually uses VT-d inside OS X...

 

[jwk]

So let me ask you this question DBP, knowing what you know now, if you had it to do over again, and if the prices were the same, would you buy the i7-3770 or the i7-3770K??