i7-870 slower than an i3-540???

[ik9000]

ok so I swapped my i3-540 for an i7-870 I picked up on ebay for a good price. I haven't had to reinstall windows, everything in the system is identical bar the CPU.

Set bclk to same frequency (160MHz), same RAM timing, voltage etc. the i7-870 is slower in single threaded applications. Consistently slower. eg passmark 1515 i3 on 22x160 (3.52GHz), but on the i7-870. I have to set the multi to 24x160 to hit 1542. But that's 3.7GHz, for which with the i3 I was hitting 1650 in the single threaded stakes.

Is there a clever reason why this is slower than a supposedly inferior CPU? Do I need to reconfigure something? It (obviously) wins the multithreaded tests hands-down, the only place I've noticed it lacking is in the single thread tests.

Tried turbo on and off, C states etc. They don't seem to make a difference in this regard.

[Spreadie]

The i3 540 was based on 32 nm Clarkdale and the older i7 870 was based on 45 nm Lynnfield.

It's common to see IPC increases with die shrinks and architecture changes.

[scaryjim]

The i3 is Clarkdale, a refinement of the Lynnfield core found in the i7. According to Hexus Clarkdale has improved IPC over Lynnfield, so clock for clock in single-threaded tests the i3 should be faster.

At stock the i3 is limited to 3.06GHz while the i7 can boost to 3.6GHz, and the i7 has twice the cores/threads, so the i7 is the better chip. But in the very specific scenario that you are testing, Clarkdale will always have the advantage.

EDIT: aaaaaaand ninja'd!

[CAT-THE-FIFTH]

The Hexus article is talking about the previous Core2 chips - the first Core i3 chips were based on the same core design as Lynnfield:

http://techreport.com/review/18216/i...ore-processors

However the core was updated with newer instruction support it seems,and since it is running on a 32NM process I suspect it might have less of a tendency to downclock.

[ik9000]

but what confuses me is the various bench sites which all suggest the 870 should beat the 540 even in single threaded usage (at stock)

eg Anandtech: http://www.anandtech.com/bench/product/143?vs=107 The only criteria the 540 is better is on power.

Similarly:
http://cpuboss.com/cpus/Intel-Core-i...el-Core-i3-540
http://cpu.userbenchmark.com/Compare...40/m841vsm2936
http://www.cpu-world.com/Compare/325...i7_i7-870.html

even passmark itself says the i7 should beat the i3 including in single threaded tests: http://www.cpubenchmark.net/compare....&cmp[]=738

[Spreadie]

but what confuses me is the various bench sites which all suggest the 870 should beat the 540 even in single threaded usage (at stock)

It should. Improved IPC means the 540 is better on a clock for clock basis. The benchmark sites show the base speeds of the CPUs but the 870 will boost to 3.6GHz whereas the 540 doesn't have a boost function and is stuck at 3.06GHz.

Your initial tests show you clocked both chips to similar levels, negating the boost advantage of the 870, so the 540's improved IPC came into play.

Basically, what Jim said earlier.

[CAT-THE-FIFTH]

Thats the thing there is no indication to show Lynnfield and Clarksdale have different IPC:

The Westmere family isn't just a straight shrink of prior parts, either. Intel has added six new SSE instructions aimed at accelerating encryption and decryption via the AES algorithm. Together, these instructions provide what the company calls "full hardware support" for AES. A seventh new instruction, PCLMULDQ, enables carry-less multiplication, which is also important for cryptographic work. Westmere processors include a few new tweaks for power savings, too, but as we understand it, that's about it. This isn't quite the overhaul that, say, Penryn Core 2 chips were compared to their Conroe predecessors. Then again, Westmere follows its mainstream 45-nm cousins by just a few months. In fact, Intel's plans originally called for a dual-core, 45-nm processor with integrated graphics code-named Havendale, but the firm canceled that product and pulled forward the introduction of Clarkdale, instead.

I suspect it might be those extra instructions coming into play - if anything maybe try Cinebench and see if the same thing happens.

[ik9000]

So this is the i3-540 at 21x180 (3.78GHz) RAM at 8x multi 1440MHz


This is the i7-870 at the same settings


comparing thread for thread (4vs8 respectively) the i3 is indeed better. Those extra instructions clearly make a difference.

But what puzzles me is this is the CPU usage during the single threaded test for the 870


notice how it is loading 2 cores. Watching the figures it rarely gets above 40% . It's almost like something is balancing the load across the cores. I've checked this is other software (CPUZ - clocks, HWmonitor, even windows task manager) all show this behaviour too. Is there a better single thread test I can run? (only problem is I've now taken out the i3 so a bit limited in what I can do. I did run some handbrake tests a while back, but don't think that will reflect single thread. These values fluctuate, with 2 and 4 hitting 30% and 2 dropping with 4 going up to 44%. Similarly if I watch the clocks in CPUz (using the CPUz bench) at stock with turbo enabled it rarely goes above 3.4GHz, with several cores running at 3.4GHz and without the drop in cores 3+4 you would expect to allow it to boost to 3.6GHz. It does hit 3.6GHz - for a split second, but that's it. And it can't be a thermal issue - it's quite happy running all 4 cores at 3.78GHz at the moment, and max temp is 65C so far, even under load.

Here is the HWmon which shows all 8 individual threads, run only during the ST test. Max on core 2 is 63%. Surely it ought to be hitting 100%?

[ik9000]

It should. Improved IPC means the 540 is better on a clock for clock basis. The benchmark sites show the base speeds of the CPUs but the 870 will boost to 3.6GHz whereas the 540 doesn't have a boost function and is stuck at 3.06GHz.

Your initial tests show you clocked both chips to similar levels, negating the boost advantage of the 870, so the 540's improved IPC came into play.

Basically, what Jim said earlier.

like an idiot I forgot to record the i3 at stock. The lowest I have is the i3 at 3.2 GHz ( a 1x bump on the multi).

The i7-870 at stock, with turbo enabled, is getting worse single thread performance than this. I'm not sure it should be.

[CAT-THE-FIFTH]

Try Cinebench in single threaded mode??

[ik9000]

ok these are the measures during cinebench runs:

single thread during and after:




multithread during and after:




It doesn't seem to want to do 100% on one thread, and still seems to split it over two.

Also interesting is that the idle state seems to have a higher Vcore than when it's running. Is this a vdroop effect?

[Spreadie]

Enable load line calibration in the BIOS and run it again - if it is vdroop, that'll help.

[ik9000]

why? It's the multithread that has the big voltage drop, but that is running more as expected. The single thread seems to be fine voltage wise but it is that one that doesn't seem to be going as expected. I'll try it, but I'm not expecting much back.

[scaryjim]

... what confuses me is the various bench sites which all suggest the 870 should beat the 540 even in single threaded usage (at stock) ...

The i7 870 has turbo, so at stock it can boost up to 3.6GHz. The i3 540 is fixed at 3.06GHz. So at stock the i7 has ~ a 20% clock speed advantage, assuming turbo is working correctly. I assume that you're turning turbo off to run your overclocked settings, so with an i7 OCed to 3.6GHz you should actually get very similar results in single threaded benchmarks to an i7 at stock...

[Dashers]

I'd test with something very simple, like one of the Pi stress testing tools. Makes no use of any fancy instructions. Should be a really good tool for comparing like-for-like.

[Troopa]

could also just be that the cpu has previously been clocked?