What CPU performance is possible with a non "K" sandy bridge?

[geofelt]

There has been some discussion and division of opinion on what one can do to improve the performance of a non "K" sandy bridge cpu.
In particular, I am talking about the cpu performance, not the integrated graphics.

1) As I understand it, the BCLK can be overclocked to a limited amount from 100 to 103-106.

2) Is this overclock available to H61, H67, P67, Z68 chipsets, or some or all?
Does it depend on the motherboard and bios?

3) The normal turbo capability of sandy bridge can increase the performance of one core to some extent. Can this upper limit be changed by changing the max multiplier? If so, what will be the frequency of the remaining cores? Does one get a higher maximum by specifying it or are you still getting the default boost?

4) The core speed is the product of the multiplier and the bclk. CPU-Z is commonly used to detect the core speed being used. Is this the speed of the fastest core, the minimal speed, the average speed, or what?
Is there an app that will display the core speed of all the active threads?

5) For an app that can use all threads, can you improve throughput by specifying a higher turbo multiplier, or, will the throughput remain the same because the average core speed is unchanged?

No doubt, someone has done some controlled testing of some or all of these variables, and I would like to know what the results were.

 

[cbrunnem]

overclocking by the multi 4.2 is the highest you can go. its a turbo work around by the mobo manufacturers from what i understand to the best of my knowledge. now if you overclock with the base clock you can go as high as 110 i believe or whatever is stable. my mobo allows upto 110.

 

[quickmana]

1) Changing the BCLK multiplier effects everything on your mobo 🙁, so moving it only slightly can cause serious issues like network problems. That's why you don't see this going beyond 110, ever.

2) All are "overclockable" but again, not having a k series means you can't change what you really want to change, the multiplier.

3) The max multiplier effects all cores. The "Turbo Boost" effects a single core. Because most apps run in a single core, it's a good idea to keep heat low by leaving your multiple at the default settings, and getting the turbo boost as high as possible. For example, 3 of my cores run at 1.6 to 3.33 ghz as needed, and one core runs at 4.2 ghz as needed. The processor determines which core needs the speed boost, and to the user (or windows) it will look like your processor is running at the full 4.2 ghz (in my case).

4) CPU-Z displays the *current* speed. If you have speedstep or similar enabled (like me), then your processor will constatly change speeds depending on current demand. Some people will use CPU-Z with another program like prime95 to display the speed "under load." Your processor will run at maximum speed under load. Some more ambitious overclockers will force their processor to always run at max speed by disabling speedstep technology.

5) The turbo multiplier doesn't help much with true optimized multithreaded apps. In this scenario, it could be argued that AMD is a better choice right now. To say it more clearly, you'll want to higher clock speed, and a lower "turbo boost."

I have personally found the best performance in almost everything by setting the turbo boost as high as possible, and ignoring everything else. This is because it's not worth the heat you'll produce to speed up all the cores for most apps today.

 

[k1114]

1) The "Turbo Boost" effects a single core.

We are talking specifically about SB's turbo which can effect all cores. Non-k can change the multiplier on a capable motherboard. And I don't mean to rain on your party but pretty much everything you said is also false.


1. You can look at the tom's mobo reviews of z68 and p67, most of the time they get to 107, one has gotten to 108, you can google to see what other have gotten to though tom's is my reliable source of choice.

2. There are a few h67 that can overclock but are limited to the non-k way of overclocking only 4 bins. And there are also some p67 that are partially locked (missing settings).

3. At stock SB has 4 bins to distribute to "overclock" aka turbo boost. That's why a 2500k can do 3.7 on 1 core, 3.6, on 2 cores, 3.5 on 3 cores, 3.4 on 4 cores. The other cores will be in idle state as they are idle. Technically overclocking on a k and non-k is the same, you are adjusting the full load state but the only difference is the k takes advantage of a separate code to have the multipler be the max on all cores instead of 1 as with a non-k. Asus has actually taken advantage of this code to "auto OC" the cpu to have the stock 3.7 on all cores. Modern cpus have multiple load states which is what makes turbo and speedstep possible.

In essence, any time you manually set a memory or BCLK setting in the P8Z68 Deluxe's UEFI (and this applies as far back as P67-based boards, according to Asus, though this is the first we've seen of it), the firmware engages an automatic rule that sets Turbo Boost to its maximum level across all cores, even when all four are active and you'd normally expect to see fewer bins of added frequency. This rule can be manually overridden in the UEFI if you want to cut power use. Or, if you set the BCLK and memory to be configured automatically, the rule is not applied. Asus says this is actually a feature of the processor architecture's microcode, if a vendor digs deep enough. It was first revealed by an Intel engineer and purportedly does not go against any of Intel's guidelines. Based on our results, it'd appear that Asus is currently the only company utilizing it.

http://www.tomshardware.com/reviews/z68a-gd80-p8z68-deluxe-z68xp-ud5,3025-11.html

4. CPUZ states core 0 by default, you can right click in the cpu area and select the other cores/see speeds of all cores.

5. Even if comparing multithreaded apps, all are different in how they utilize each core, which is why benchmarks test multiple real world apps/synthetics. In theory more cores would be better in a fully multithreaded optimized app but of course you still have to keep in mind, ghz, architecture, etc. and depending on the workload, the rest of the pc specs.