Sandy Bridge processors - how do they compare?

Let me preface the following by saying that I am in no way familiar with processor technology and that I ask you to forgive my ignorance.

That said, I am wondering how Intel's Sandy Bridge processors compare. For example, what is the difference between an i5-2410M, i5-2540M, and i7-2620M? I understand that the i7 has a higher clockspeed and more L3 cache, meaning it can process more calculations per second than the i5s when fully taxed. But I don't understand how all three can be rated at the same 35W tdp if they are based on the same microarchitechture. Does the i7 use more power than the i5s when fully taxed? In other words, how does its efficiency compare to the i5s'? Is there a difference in power consumption and/or performance if all three are not fully taxed?

Any expert guidance would be much appreciated. Thanks in advance!

 

Core i7s are made with better silicon, so they can operate at a slightly lower voltage while maintaining a higher level of performance.

 

Food for thoughts.







Intel Sandy Bridge Processors Gaming Performance: Part II - Notebookcheck.net Reviews

 

ah, wut?

 

I'm not quite following your question...?

Are you saying, is there a difference when say the i5-2410m operating at 1.5 ghz vs the i7-2620m operating at 1.5ghz?

 

TDP is not reflective of power consumption. The faster the CPU and the same cache of the same generation, the more power it will consume in idle (though the difference is very little) and load. So yes the faster i7 dual core will consume more than the i5. They are equally efficient as they are from the same silicon. Only the lower TDP CPUs are binned, the same TDP CPUs are just differentiated by marketing.

 

This is interesting. The GPU seems to be doing all the work when playing games in high resolutions. In such scenario the i3 and quad i7 are almost equal

 

Yeah, I think most CPUs have come to a point where they are powerful enough and the bottleneck is the GPUs. Looks like there is some difference when using anistropic filtering though, but i5s are very close to the i7s anyways.

 

i3-2310m is around the speed of a mid level first gen i series proc.

 

Thanks for all the answers thus far!

Yes. From what I'm reading it seems like both would process calculations equally quickly and consume identical amounts of power -- is that correct?

 

No, the i5 will be about 18% slower than the i7 and the i7 will consume much more power when pushed to the edge of it's performance envelope.

See:
PassMark - Intel Core i5-2410M @ 2.30GHz - Price performance comparison

See:
PassMark - Intel Core i7-2620M @ 2.70GHz - Price performance comparison

 

Those are at different clocks? And from what I've heard, passmark isn't exactly the most reliable benchmark.

At the same frequency the i7 and i5 should perform similarly, considering that they're the same architecture, though differences in cache sizes might make a little difference. The power consumption should be pretty much the same.

 

Better manufacturing techniques and better design allow for better performance while still maintaining a fair power profile. Silicon is silicon.

 

Very unlikely. A wafer is manufactured then cut. Some of those dies from the same silicon will be rejected just by visual defects. A small function test will be made and some more chips may be rejected. Then during burn-in further chips from the same silicon can be rejected. Then Class testing will determine the dies max speed and binning.

A word about TDP, this is a figure provided to manufacturers to provide a cooling solution such that when operated in worse case scenario but still in specification, the chip will not run more power than the rated TDP.

An example I can give is with C2D. I run an overclocked P8400 (2.26GHz). For me to run at 2.8GHz/3.0GHz requires elevated Vcore to 1.27V. Running Linpack the CPU draws up to 45W-50W measured from the battery taking into account idle power. For the higher rated P9700 (2.8GHz) with more cache and lower required VID, it should run within it's 28W TDP at 2.8GHz.

P8400 (TDP 25W) @ 2.8GHz >45W
P9700 (TDP 28W) @ 2.8GHz <28W

So much for equality. Sorry I don't have any SNB figures for you but hope this helps.

 

It may not be the most reliable, but when over 700 different people (between the two cpu's...) are reporting, then it's better than your guess or mine.

As for the different clocks - of course - the i7 can do much more work, that's why I stated 'when pushed to the edge of it's performance envelope'.


Dufus, yeah, no equality (silicon is not just silicon...).

 

This might not be the best example because the actual architecture of the P8400 and P9700 are of different steppings (R0/M0 vs E0. The P9700 is basically an undervolted and undermultiplied T9900/T9800). Also, their multipliers are different, which forces parts of the chip to work at different speeds. Their cache sizes are also different, but that works to the advantage of the P9700. So in your tests above, you are outlining the power advantages of a different stepping CPU at different multipliers, instead of outlining the power advantages of similar stepping but differently binned CPUs.

tl;dr: apples to oranges.

 

Yes, there can be different TDP's for different steppings even for the same model but are you suggesting the difference in stepping can account for a >60% difference in power?

Of course the multipliers are different, the P8400 is locked to a lower multi to reduce it's maximum frequency. What would you consider a better example?

What about the same SKU where one chip can be undervolted more than another? Or with SNB one chip of the same SKU/stepping running hotter than another or being able to run 110MHz BCLK vs 104MHz. Silicon is not equal even on the same SKU/stepping let alone different binning.

 

Not quite sure where you're pulling the 60% difference in power from... but no I am not saying that. Actual power consumption and TDP are two different things. In rough order of importance...

Here's what goes into the TDP number:
Marketing, stepping, voltage, multiplier.

Here's what accounts for actual power consumption:
Voltage, multiplier, stepping, base frequency, power features (DFFS/IDA/Speedstep).

So when you say a >60% difference in power, it's quite ambiguous. You are going off of TDP, yes? 25W vs 35W. However you are assuming these CPUs are running at stock frequencies and multipliers, which already means you have introduced a significant difference in power consumption (2.80 vs 2.26 GHz).

TDP is so inaccurate because in reality, the faster a chip the hotter it will run, and therefore the more thermal dissipation you need. An E0 T9400 is going to run slightly cooler than a T9600 at load because it is clocked slower. But they have the same 35W TDP. An even better example is the voltage differences across "35W TDP" chips. A T9900 uses a lot more voltage than even a T9600. My point is, TDP determines what class of laptop a CPU should fit in. It's not a scientific number.

A better example to test binning would be to get two similar stepping CPUs. For example, a P9700 and a T9900. Both are E0 stepping, but one is certified to run an extra 266MHz over the other.
If you wanted to test different steppings, say compare the P8400 to the P9700, you could lock the P9700 down to the same multiplier and start the overclocking competition from there.

The same SKU where one can be overvolted more than another is a decent example of binning. Even better would be which one could use a given voltage to clock higher, as you said with the BCLK.

I'm not trying to be condescending or whatever, I know you know what you're talking about in terms of binning but I'm just getting a bit technical on this particular situation because I feel it's not the best way to test the physical/electrical qualities of silicon.

 

Um, no I was running the P8400 OC'd using 333MHz BCLK to make the comparison at around the same CPU clock. The TDP as I said should be worst case scenario while the CPU is still working in specification, ie worst/heavy load. TDP for P9700 is 28W last time I looked. P8400 is 25W but as you can see I'm operating well above it's thermal designed power for cooling by OC'ing.

Fair enough. Unfortunately I don't have an inexhaustible collection of CPU's but if I get the chance I'll try to look into it from a closer perspective to try and make a fairer comparison. But don't wait in anticipation lol.

 

Silicon may be silicon, but isn't silicon binned?