@pavlor

Actually, you are claiming the T4240 is 1/6th the performance of an i5 in single core performance, but cannot produce any number (only two or three levels of assumptions) to back it up.

I am claiming that in coremark the i5 is up to 20% faster per MHz, which with a double clock in single thread turbo mode could make it 2.4 times faster for single thread.

But then you screwed that up by quoting the 1.7 number for the dual-thread performance vs single-thread on T4240, which (if true) brings us back to: 1.2 * 1.7 / 2 = 1.02, so 2% faster per MHz, with a double clock frequency.

So the numbers we HAVE say the best i5 is almost exactly twice the performance of the T4240 for a single thread. I suggest you come up with some better numbers if you want to prove anything else.

_________________
This weeks pet peeve:
Using "voltage" instead of "potential", which leads to inventing new words like "amperage" instead of "current" (I, measured in A) or possible "charge" (amperehours, Ah or Coulomb, C). Sometimes I don't even know what people mean.

@Zylesea

Maybe new naming convention (I suspect prior versions didn´t mention e300, but simply 603e - like all documents from 2005, 2006).

@pavlor

Indeed it was. But one core of same type isn't necessarily identical to another core in same family. Like "G4" aka e600. Different versions have wildly different DMIPS/MHz numbers due to internal changes. They share the instruction sets, though. So software compiled for one ewhatever should run unmodified on all devices of the same family. More or less

My point: The whole "G this or G that" and "e this or e that" is a completely counter productive way of comparing performance. Features, yes. Performance, no.

_________________
This weeks pet peeve:
Using "voltage" instead of "potential", which leads to inventing new words like "amperage" instead of "current" (I, measured in A) or possible "charge" (amperehours, Ah or Coulomb, C). Sometimes I don't even know what people mean.

@olegil

P2041 1500 MHz (4 cores): 4655.00 Coremark per core, 3.1 Coremark/MHz
T4240 1800 MHz (12 cores/24 threads): 14980.25 Coremark per core, 7490.125 Coremark per thread, 4.16 (or 4.89) Coremark/MHz.

According to these Coremark results, e6500 is 1.34x (or 1.58x) faster per MHz than e500mc core.

Now, may I ask how fast is e500 core in comparison to G4 core? If you say they have similar performance/MHz (except AltiVec), I can prove thanks to series of SPEC CPU benchmark results that G4 1.5 GHz is roughly 1/10 of integer single core performance of not that new Core i5 CPU.

@pavlor

I would like to see those numbers proving exactly that. Here's what I get:

1: Best i5 SpecInt2006: 50.4

2: Now that was a quad core (no hyperthread) 3300MHz, with turbo mode 3700MHz. So best estimate single-core performance would be rougly 50.4/4*3700/3300 = 14.13

3: This is about the same SpecInt2006 result as a Core 2 Duo T7600 2.33GHz, which scores about 2460 in SpecInt2000. So let's go ahead and assume that single core performance of 3700MHz i5 in SpecInt2000 is 2460.

4: The G4 gave 307 at 1GHz. If this would scale to 1.8GHz, it gives 552.6, which is 1/4.45 (Edit: of 2460). So a 1.8GHz G4 would already be far better than your 1/6 assumption.

5: Freescale says 6 dhrystone instructions per clock on e6500, and a 70% increase from running dual threads. 6/1.7 = 3.53 dhrystone instructions per clock in single thread mode then.

6: The G4 would manage about 2.3 dhrystone instructions per clock, so the e6500 is 1.53 times faster per clock. Now we're mixing dhrystone with SpecInt2000, but we have to get the numbers from somewhere. So 1.53/4.45 = 1/2.9. Roughly.

That's more than twice your "one sixth" from your first post here.

Again, cough up some numbers to prove me wrong, I have little to no use for vague hints about your assumptions.

Interestingly enough, I now gave two completely unrelated examples showing your assumptions to not seem to hold water. Might be time to admit you got owned

Last edited by olegil on 09-Mar-2013 at 12:34 AM.
Last edited by olegil on 09-Mar-2013 at 12:31 AM.

_________________
This weeks pet peeve:
Using "voltage" instead of "potential", which leads to inventing new words like "amperage" instead of "current" (I, measured in A) or possible "charge" (amperehours, Ah or Coulomb, C). Sometimes I don't even know what people mean.

@olegil

Quote:

1: Best i5 SpecInt2006: 50.4

Right.

Quote:

2: Now that was a quad core (no hyperthread) 3300MHz, with turbo mode 3700MHz. So best estimate single-core performance would be rougly 50.4/4*3700/3300 = 14.13

What?! You can´t divide by 4 in SpecInt2006 (it is not SpecInt_rate2006). Yes, auto-parallel adds some speed boost, but not in scale of four (see dual-core or six-core SpecInt2006 results for comparison).

Quote:

3: This is about the same SpecInt2006 result as a Core 2 Duo T7600 2.33GHz, which scores about 2460 in SpecInt2000. So let's go ahead and assume that single core performance of 3700MHz i5 in SpecInt2000 is 2460.

2460 SpecInt2000? That is nearly same value as in case of Core 2 E6600 2400 MHz. I have here both Core 2 Q6600 (same single core performance) and Core i5-2500K (which scores about 40 in SpecInt2006). I can assure you, my brother´s i5 is more than two times faster (in sense Y=2*X) than Q6600 in single core demanding applications (eg. DosBox - multiple cores can´t help here).

Quote:

4: The G4 gave 307 at 1GHz. If this would scale to 1.8GHz, it gives 552.6, which is 1/4.45 (Edit: of 2460). So a 1.8GHz G4 would already be far better than your 1/6 assumption.

As I pointed in previous article, your numbers are simply wrong - 1/4.45 of 2460 SpecInt2000 means 1/4.45 of single core performance of Core 2 Q6600, that is nearly half as fast as i5-2500K in single core operations. So, with your own numbers, 1.8 GHz G4 would be (at best) 1/8.9 as fast as i5-2500K.

If e6500 core is 1.34x (or 1.58x) faster per MHz than G4, it will be 1/6.64 (or 1/5.63) as fast as i5-2500K. As I wrote, this i5 is more than 2 times faster than Q6600, so even these numers are probably better than reality.

Quote:

Again, cough up some numbers to prove me wrong, I have little to no use for vague hints about your assumptions.

See above.

Quote:

Interestingly enough, I now gave two completely unrelated examples showing your assumptions to not seem to hold water. Might be time to admit you got owned

I don´t think so...

@pavlor

Quote:

pavlor wrote: What?! You can´t divide by 4 in SpecInt2006 (it is not SpecInt_rate2006). Yes, auto-parallel adds some speed boost, but not in scale of four (see dual-core or six-core SpecInt2006 results for comparison).

Interesting. I didn't know that. Thank you. Now you're starting to provide some numbers, not just hints (I don't care about your brothers machine seeming to be twice as fast as whatever, unless you can provide a number for it). And the 2460 came from SpecInt2000.

_________________
This weeks pet peeve:
Using "voltage" instead of "potential", which leads to inventing new words like "amperage" instead of "current" (I, measured in A) or possible "charge" (amperehours, Ah or Coulomb, C). Sometimes I don't even know what people mean.

@olegil

Quote:

(I don't care about your brothers machine seeming to be twice as fast as whatever, unless you can provide a number for it)

As you wish.

Quote:

And the 2460 came from SpecInt2000

Core 2 Duo E6600 SpecInt2000: 2595 (Source)
Core 2 Quad Q6600 SpecInt2006: 18.2 (Source) (Note: There is no SpecInt2006 result for E6600 on the spec.org page, Q6600 should be close enough in single core performance)
Core 2 Duo E6700 SpecInt2006: 20.0 (Source) (higher clocked than E6600, clearly shows more cores don´t give massive speed increase in this benchmark)
Core i5-2500K SpecInt2006: 43.3 (Source)

Enough?

@olegil

going a BIT off-topic now: Actually, the more I read about it, the more I think I want an i5 in my next laptop. What's the advantage of an i7 anyway? Must admit I've sort of "tuned out" from it all lately.

_________________
This weeks pet peeve:
Using "voltage" instead of "potential", which leads to inventing new words like "amperage" instead of "current" (I, measured in A) or possible "charge" (amperehours, Ah or Coulomb, C). Sometimes I don't even know what people mean.

@olegil

Quote:

think I want an i5 in my next laptop

CPU is not that important, it is GFX what makes difference. Laptop I use right now has Core i5-2430M 2400 MHz (2core/4threads) with both IntelHD Graphics and Radeon HD7470M - enough for most of my favourite games. If it was only a little bit less loud...