My i5 build

Page 2 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
jy_nz: i7 Turbo seems to be working correctly and will give you the most accurate look at your multiplier. This tool also works on Core 2 CPUs.

Your screen shot shows that you are running a single threaded task which is being mostly shared between the two center cores. The two outside cores are in charge of running any background tasks while this is going on.

On a Core i5 750, you will get a 24 multiplier but you only get this when 1 core is active. The two center cores are taking turns as the active core while the two outside cores are spending most of their time asleep.

I don't have the official specs yet but I'll post a utility that you can use to confirm your CPU's capabilities in a moment.

As soon as 2 cores are active, your CPU will no longer be able to use the 24X multiplier. When all 4 cores are active, your maximum multiplier is 21. The Load Filter setting in i7 Turbo limits the data going into the Min and Max boxes. In your screen shot you have this set to 95% (the default) so it is recording the min and max multiplier only when your load is over 95%. If you change this filter to 0%, then all data will be considered for the Min and Max statistics. That's why it is only showing 21.000 because at full load, greater than 95%, that's all this CPU can do.

The 23.733 and 23.829 values for the two center cores are a very accurate average of the multipliers for those two cores during the previous 1 second interval. What is constantly happening is a core is trying to run at the maximum 24 multiplier but any background activity while you are benching kicks the multiplier down to a lower value. It might drop all the way down to 21 when two cores are active or it might drop only to 23 or 22. The multiplier on these CPUs will constantly be cycling like this, hundreds of times a second, based on load. CPU-Z rounds the multiplier off which may or may not be 100% accurate.

i7 Turbo is based on the Intel Turbo White Paper. It compares two high performance timers within the CPU to accurately calculate the average multiplier during an interval. This is the method that Intel recommends for CPUs that support Intel Dynamic Acceleration like some of the Core 2 mobile chips or Intel Turbo Boost which Core i9/i7/i5/i3 CPUs use.

Enabling C-States or C3/C6 allows the cores that are not doing the work to enter the sleep state and become inactive. The maximum multiplier available is based on how many active cores you have. The fewer cores active, the more bins of turbo boost will be available to you.

Model Specific Register (MSR) 0x1AD contains information about how many bins of turbo boost are available. My tool will show the specs for when 4, 3, 2 and 1 core is active. The 4 numbers will be side by side and are in hexadecimal format.

http://www.fileden.com/files/2008/3/3/1794507/MSR1AD.zip

When 4 cores are active, we know the maximum is 21 and when 1 core is active, we know the maximum is 24. I'm just not sure about what the two center values are for this CPU.

24 decimal translates to 18 in hex and 21 decimal translates to 15 so you should see something like:

15 __ __ 18

What do the two center numbers show?

This tool also works on the previous Core i7 CPUs.

Edit: Next time you are running wPrime, include Core Temp and i7 Turbo. The reported temperature looks like the two center cores are doing most of the work. When playing around you can also go into the Task Manager and right click on a task and use Set Affinity... to lock it to a particular core or cores. Hyper PI is another good testing utility that lets you run 1 to 4 instances of SuperPI. When running a single thread of SuperPI, it automatically locks it to one core.
 
unclewebb, thanks for the info. I know previous processors use core 0 and core 1 as the main loading(quad cores), but in this case the i5 750 loads more with the center cores, if i read correctly and understood. Correct me if I'm wrong. So in this theory wouldn't the center cores be the hottest for the i5 750? It seems so in the OP's pics.

So it seems that Intel changed their technology around..seeing how quad cores are hottest on first two cores, but now its center for i5..not sure about i7 can't remember

This thread has become very interesting umm..
 
A long time ago, some software used to always ask for core 0 but more software is multi threaded aware these days so it just asks to run on a generic core and ends up on whatever core Microsoft or Intel gives it.

In this one particular example, it appears that the work load got handled by the two center cores and they shared it fairly equally but I don't think this will always be the case. The load can be scheduled on different cores hundreds of times a second so most software is only giving you a snapshot at that one instance in time.

The Core i7 - 9xx series worked very similar to this when the motherboard properly supported this feature. Unfortunately, when overclocking, most motherboards limited turbo boost to +1 no matter how many cores were active. It's great to see this new P55 board doing this right even when he's overclocking and allowing access to the full +4 turbo boost that the i5 750 is capable of.

rge over on XtremeSystems did some testing of his Core i7 - 950. These have a default multiplier of 23 and are capable of either 1 or 2 bins of turbo boost, just like an i7-920 is. When 2, 3 or 4 cores are active, you get a maximum of +1 and when only one core is active you can get a +2 turbo boost which on a 950 is equal to a maximum of 25X.

Here's a screen shot of a 950 running a single thread.

v8248.jpg


You can see that at this particular instance, most of the load is locked to one thread which is allowing the multiplier for that thread to approach 25 similar to what this i5 is doing.

I don't know if the CPU chooses a particular core when it discovers a task that is running continuously like a benchmark or if a task that starts running on a particular core stays mostly on that same core until it is completed.

This new i5 is looking like a hell of a new low cost CPU. If you can run single threaded tasks with the multiplier spending most of its time at 24 while the BCLK is up at 180, you're going to have a very fast computer for the majority of tasks. It will be interesting to find out how many bins of turbo boost you get when 2 cores are active. Of course, running it locked at 4.2 GHz (21 X 200) would be a lot of fun too. With adequate cooling you should be able to get similar clocks out of one of these compared to an i7-9xx. This i5 doesn't have hyper threading and is a more efficient design so with less heat output, clocking one up into the low 4 GHz range should be pretty much standard for any serious enthusiast.

Edit: Interestingly, when Super PI was locked to the first thread using Task Manager, it was not able to use the 25 multiplier as much. This seems to show that letting the CPU manage tasks itself might be the best thing to do.

v8245allcoresc6enable.jpg
 
A bench mark to start - single threaded Super PI 32M, 9 minutes and 49.6 seconds of overclocking goodness at 4.2 GHz with resaonable temps:


pi32mk.jpg





Then a question for unclewebb:

spi42.jpg


Note still stable with a bunch of programs open at 4.2 GHz, up one MHz base clock step from the above Super PI run.

Even with the Load FIlter threshold dropped down to 50% I don't seem to be recording the Max Min numbers.

By the way I loaded up your MSR 0x1AD tool as you can see.

A not to viewers from [H]ard|Forum - no cliams being made here other than what's stated. As pointed out in past postings we're still thermal limited by the Intel cooler.
 
my i5 750. had it since last thursday (im from Australia)

is sitting at 3.6ghz stock volts and stock cooling and maxed out at 72deg on all cores in prime 95. wasnt prime stable at 3.6 on stock vcore though. but was prime 95 stable at 3.4ghz

i will shoot for 4ghz prime stable on stock cooling tonight. (will need a little more volts to be stable though)

I will try to update this.

theres a difference between "stable with programs open" and stable under stress testing.

Still this clock just owns overclocking. stock i got 24XXX 3d marks (3d mark 05) with a gtx275. and overclocking to 3.4 i got over 26XXX and about 27000 cpu marks
 
unclewebb - thank you for a great discussion and for access to your utilities.

Psychosaysdie - thanks for the encouragement.

slickncghia - if you run into problems with high base clocks you might want to look at dropping your DRAM or QPI clock multipliers. I'm running the machine fine 24x7 at the 150 MHz base and if that is OK for a week I'll crank it up another notch or two.

heapssickaye - yep same here, can't understand how people can get their knickers in a knot when the processors have been on Trade Me, eBay, Fry's Electronics (famously), a whole bunch of NZ retailers, and every computer store in Singapore!

As for non disclosure agreements (NDAs), these are individual contracts signed between the disclosing party (Intel) and the receiving party. If you signed one you'll know when it expires ... if you didn't sign one you’re not bound by contract.

Oh, and one more thing ... I mentioend that a mate had the same build except his system has 8GB with all 4 DRAM slots filled. He is seeing essentailly the same numbers.
 
just realised you didnt keep it at stock volts... that makes me feel better. i was prime95 stable at 3.4ghz at 1.1vcore.

i was away all weekend and have only played with it a little bit. theese chips have HEAPS of OC headroom. i think we are going to see some huge numbers on air. maybe 5ghz+ from good coolers
 
slickncghia - the CPU volts are on auto so they're managed by the CPU. You can play with then in the BIOS but I've not done that yet.
 
cheers mate ill give that a try.

auto volts worry me a bit. i always thing they will go too high and fry it. when you are set on your clocks. start manually bumping it down till its as low as possible but still stable.

you are doing well with that ram btw. what latency are you running?

im running 4gig patriot viper 1600 7-7-7-20.

it dosent like sitting on 1600 but just under is fine. hopefully i can get it close to 1600 when im done with the clocks or maybe even tighten it up more?
 
MSR 0x1AD shows the maximum amount of turbo boost available. Your CPU shows 15 15 18 18. These numbers are in hexadecimal so converted to decimal they are:

21 21 24 24

This means that:

When 3 or 4 cores are active you get a maximum multiplier of 21.
When 1 or 2 cores are active you get a maximum multiplier of 24.

That is what your i5 750 is capable of and it's up to the motherboard to support this feature. It looks like your MSI is doing this correctly, even when overclocking the BCLK.

When running a single threaded task like a Super PI benchmark, the multiplier should be mostly at 24. If there are some background tasks and 2 cores are active, it should still try and use the 24 multiplier. Only when there are more than 2 cores active will it drop down to 21.

In your previous screen shot where it showed the hardest working core at an average of 23.829, that means that this core was using the 24 multiplier 94.3% of the time and the 21 multiplier 5.7% of the time.

If you're not sure about i7 Turbo, drop the Load Filter down to 0% and then all data should start showing up in the Min and Max boxes. If the Load Filter is at 50% and nothing is showing up in Min and Max then that is because your average load has not exceeded 50%. You need to put a load on that CPU.

Here's another simple testing tool.

http://www.fileden.com/files/2008/3/3/1794507/LoadTester.zip

It allows you to put a gradual load on your CPU from 0 to 100%. You can run 4 instances of this program so that you can load each core in a controlled manner and then you can watch what happens to your multiplier. As the load goes up, your average multiplier should go down. You can also run HyperPI which I think I mentioned before. This is a GUI that lets you run 1 to 4 instances of SuperPI at the same time. This will also let you see how your CPU responds to various loads and how stable it is. That's what all the sharks here are waiting to see. :)
 
Slickncghia

I wouldn't worry about auto volts for the moment. All it means is that the mother board is obeying the Intel VID outputs from the CPU. In other words a stock set up. That’s what I mean when I say stock anyway. Have a read back up this thread for more info.

A lot of people don’t get how VID works. You’ll see lots of question on forums like “what volts are you running?” or “what’s VID?”.

Put simply the Intel CPU usually knows what’s best for it. Clever logic in the chip evaluates loads, temperatures, and timing requirements to establish what voltage it requires. The mother board voltage regulator takes power from your power supply and steps it down to suit the CPU.

The output voltage of this mother board regulator is set by the VID bits output by the CPU. Typical mother board voltage capability range from 1.6V down to 0.5V, although the CPU will probably only operate over a subset of that range. For example the i7 datasheet (I don’t have an i5 sheet yet) spec’s the maximum permissible core voltage as 1.55V.

Sure enough when I examine the CPU voltage under a variety of loads the voltage swings between a low of 0.94V at idle up to 1.44V (for altec).

The MSI BIOs does not seem to allow the CPU voltage setting to be taken off AUTO. This might be a restriction of the REV 1.0 BIOS or there may be a cunning setting somewhere else that needs to be enabled first. In any case at the moment I’m leaving it a stock (that’s ‘stock VID controlled’ for the pedantic).

By the way Intel sepc’s the maximum DRAM voltage at 1.875V with a typical of 1.5V so the usual ‘performance’ DDR3 voltage of 1.9V is out. Your stuff looks OK at 1.65V.

My DRAM is running at the default settings of 1302 MHz 9-9-9-24.

The cheap Kingston value DDR3 sepc’s are:

JEDEC #1 6.0-6-6-17-23 @ 457 MHz
JEDEC #2 7.0-7-7-20-27 @ 533 MHz
JEDEC #3 8.0-8-8-22-31 @ 609 MHz
JEDEC #4 9.0-9-9-25-34 @ 685 MHz


Unclewebb

Your attention has been a real help. Thanks for the turbo boost info – very interesting, particularly the possible of two cores operating at x24. We’ll have to sus that out more carefully. As you know wPrime can be set from 1 to 4 threads, it was 1024M with 4 threads at x20 and 3.6 GHz that got things to the 75 deg C quoted earlier.

Thanks for the load tool it working but I don’t have time to play with it at the moment. And thanks for the tip about Hyper PI, hope for a change to test later tonight.
 


hmmm that got me a little worried. as my stuff is 1.9v
http://www.patriotmemory.com/products/detailp.jsp?prodline=5&catid=18&prodgroupid=107&id=669&type=1

when i was buying it i asked around to make sure the voltage wasnt too high and no-one said a thing.

have you got any more information on this?

cheers mate
Ben
 
your link dosent work for me but mine does.

its 7-7-7-20 at 1.9v

but yeah ill check tonight what volts its actually running at.

why is the dram voltage limited to 1.875? cpu? mobo?

this might explain why its not stable at the full 1600

edit: sorry to hijack btw 😛
 


JEDEC ddr3 standards allow a max dcv of 1.95, so your modules are within specs. Might check what your MB supports voltage-wise.
 
I was curious to see how these new CPUs compare to the previous Core 2 so I ran a couple of Super PI benches on my E8400 at 4000 MHz and 4250 MHz for comparison:

superpie84004000mhz.png


superpie84004250mhz.png


I think a Q9550/Q9650 will need to run at close to 4.5 GHz to be competitive with this new i5 750. I can't wait to see an i5 running with a decent cooler or water.
 


***, why didnt anyone warn me when i asked...

and why are they making ddr3 that wont work with the chips that most people will use ddr3 at???

i guess i will just set the mobo dram voltage to 1.65 and i might have to loosen up the timings for higher speeds at lower volts? does that sound about right
 


what are your scores? i cant see the pic/

Ill run superpi tonight as a comparisson if you would like
 


Not wanting to start an argument, I just quoted the max vdc in the JEDEC ddr3 standards. MB mfg's are allowed to use, underuse, or abuse the standards as they see fit. There are no standards police, yet, to my knowledge. The min vdc, from memory, is .09vdc.
 
E8400 4000 MHz DDR2-1000 CL 5-5-5-18
SuperPI 4M - 1m 05.770s

E8400 4250 MHz DDR2-1000 CL 5-5-5-18
SuperPI 4M - 1m 02.790s

The i5 750 already has me beat.
Tom's has a full i5/i7 review up and not surprisingly, they like these new CPUs too.
 
uncleweb

From HardOCP's 'official' first review of the new i5s:

"If you are first and foremost a gamer and an overclocker, the Core i5-750 makes every other solution look like wasted money."


slickncghia

Don't fret too much as the mother board will probably limit your DRAM voltage and your DRAM will work fine at 1.5V but maybe not with 7-7-7-20 timing. Just might need to play with the available settings.

This i5 stuff is all so new that even today MSI announced an BIOS upgrade from 1.0 to 1.1 on a mother board they still don't list on their web site. BTW you didn't mention you mobo?


crock

Noted. Didn't wish to offend, after all the full Intel Core i5-700 datasheet has only just become publicly available.