SB i5-2500T vs. Underclock 2500K

[jpmucha]

Hello all,

I'm wondering if anyone has seen an analysis comparing the power usage of a 2500T and an 2500K underclocked to 2.3 GHz.
Prices are predicted to be comparable between the two processors, so it becomes a choice between the freedom and
maximum performance of the 2500K and the dynamic performance and thermal optimizations in the 2500T.

From a pure performance standpoint, the 2500K has a +1GHz stock speed and a turbo speed 400 MHz faster. The integrated
graphics has twice as many execution units running 200 MHz faster (although the 2500T strangely has a 1250 MHz turbo speed).

Before the base and turbo speeds are set, I don't believe there's be anything fundamentally different in the silicon between a
2500K and a 2500T other than any potential binning to make sure the wafers with the silky 'magic' are destined for 2500T labels.

With that in mind, would the 'effective TDP' of a 2500K underclocked to 2.3 GHz (and GPU comparably underclocked if possible) begin
to approach the 45W TDB of a 2500T? The underclocked 2500K would still have the 12 EUs chewing power (unless gated). Has
anyone described how turbo would work with an underclocked CPU? Would the max turbo speed still be 3.7 GHz, or would it only be
a +4x to the base multiplier (ex. 3.3 -> 3.7 GHz for a 2500K)?

It doesn't seem inconceivable that an underclocked 2500K could have the same idle and 'normal' power draw as a 2500T
(assuming low GPU situation where unused GPU units gate), but be able to turbo beyond the speeds achievable by a 2500T. Its this
the secret behind why the processors are the same price? If that's the case, the only drawback to the underclocked 2500K approach
is that it could spike up beyond the 45W TDP window (potentially requiring a stronger cooling solution for OEMs).

I would like to design a power efficient system, but could live with the temporary spikes in power draw (which only occur when demanded).

Thanks for the input!

Jason

 

[Zenthar]

My guess is the difference in Turbo mode is also part of the processor's difference. If you were to downclock the 2500K to 2.3GHz, you might be stuck with a maximum 2.7GHz turbo.

If I look at all the 2500 series (2500T (45W), 2500S (65W), 2500 (95W) and 2500K (95W)), the clocking of the CPU itself might be responsible for 30W. The only difference between a 2500S and a 2500 being that the 2500S clock can go as low as 2.7GHz, but the turbo can still go up to 3.3GHz.

Following the same logic, the only difference between the 2500 and 2500K (beside the unlock) is the 6 vs 12 EUs and both are rated 95W, that could mean the EUs number in insignificant in the equation.

Therefore, if you were to downclock a 2500K to 2.3GHz and it was to turbo itself only up to 2.7GHz, you could potentially get a CPU that has a an even lower TDP than the 2500T, but also be slower.

I'd also like to point-out that TDP isn't power consumption, but heat output. There is probably a correlation between the 2, but I don't know if it's linear, exponential or logarithmic.

Finally, if you are looking at a compromise between a 2500's performance and a 2500T's power consumption, the 2500S would be your best candidate instead of trying to tweak a 2500K. To that add the fact that any MB that will allow you to use the on-die GPU would also prevent you from changing the multiplier, underclocking the 2500K might not even be a possibility.

 

[jpmucha]

Thanks for the response Zenthar. I have the same concern you state in your first paragraph. Is the turbo=base_speed+X (where X is 4 for 2500/2500K, but X=10 for 2500T), or is it just a maximum speed limit (multiplier) programmed into the chip.

I agree, from my understanding, TDP is more of a guideline for the maximum strength of the cooling solution needed for a system. I guess another 'programmed' difference would be that the 'T' model's power management circuitry will be tweaked to only allow the CPU to sustain 'turbo' only to the degree that a 45w classed cooling solution can handle... where a 'S', vanilla, or 'K' variant would stay in turbo long enough to tax a 65 and 95 watt cooling solution (as specified).

I think the number of EUs figure in the actual energy usage, but as we discussed above, the TDP is something different. Intel's power gating is very good, so I expect idle conditions won't add much energy load... and I guess your loaded situations will use whatever it can within the thermal/power window allowed.

A competing site described explicitly that the unlocked processors can be set to anything from 16x-57x (meaning 5.7 GHz max overclock!), so an underclock is a sure thing for the K chips. Another poster on Tom's forum site said most modern chips can be underclocked... its the high end that gets locked down.

Thanks again!

 

[Zenthar]

I think I've read somewhere that the H66/67 chipset doesn't allow you to change the multiplier at all, but I've read conflicting posts. Some say it's only the memory interface you cannot OC on H66/67 boards.

 

[jpmucha]

I think I've read somewhere that the H66/67 chipset doesn't allow you to change the multiplier at all, but I've read conflicting posts. Some say it's only the memory interface you cannot OC on H66/67 boards.

Interesting. All the reviews I've seen so far have been on P67 boards, so that may be a possibility. I'm planning on going with an H67 board since I'm building a server... so GPU performance isn't much of an issue... and the lower power consumption is of interest. This will be something I'll have to research.

 

[Zenthar]

The P67 chipset lets you overclock CPU and memory but it lacks the flexible display interface necessary to support SNB’s HD Graphics. The H67 chipset has an FDI so you can use the on-die GPU, however it doesn’t support CPU overclocking—only memory. [...] There is a third member of the 6-series family that will begin shipping in Q2: Z68. Take P67, add processor graphics support and you’ve got Z68. It’s as simple as that. Z68 is also slated to support something called SSD Caching, which Intel hasn’t said anything to us about yet.

Source: Anandtech

 

[jpmucha]

The P67 chipset lets you overclock CPU and memory but it lacks the flexible display interface necessary to support SNB’s HD Graphics. The H67 chipset has an FDI so you can use the on-die GPU, however it doesn’t support CPU overclocking—only memory. [...] There is a third member of the 6-series family that will begin shipping in Q2: Z68. Take P67, add processor graphics support and you’ve got Z68. It’s as simple as that. Z68 is also slated to support something called SSD Caching, which Intel hasn’t said anything to us about yet.

Source: Anandtech

That might shoot down my little 2500K plan for tomorrow... The 2500T will be less flexible (at least with an H67), but closer to my 24/7 power concerns. The 2500K still seems to idle well... so I still may be ok...

Thanks for the warning...

 

[jpmucha]

Picked up a 2500K from Microcenter today. Also went for the Asus H67 EVO board. I'll report how the underclocking attempts work for H67. Thanks for all the great feedback!

 

[jpmucha]

Best answer selected by jpmucha.

 

[Maziar]

This topic has been closed by Maziar