as above:

Freescale published the highest CoreMark score ever for an embedded processor

https://community.freescale.com/community/the-embedded-beat/blog/2013/03/04/top-embedded-networking-system-on-chip-processor-highest-eembc-coremark-score

_________________
retired

@AmigaBlitter

Good test

@AmigaBlitter

Multicore performance of T4240 is impressive, but single core speed is still far behind current desktop CPUs (roughly 1/6 of performance of recent Core i5).

@pavlor

It's an embedded CPU afterall, so it's pretty OK. I hope they can turn their fortunes around, but atm it doesn't look too goo for Freescale.

_________________

@pavlor

How did you get to that 1/6 estimate?

In pure single thread MIPS test the difference is something like 2-4MIPS/Mhz per core, IIRC (that would indicate 40-60% faster per Mhz .... so it needs 18Ghz i5 core to be 6x faster???). Perhaps you meant single thread performance? ((There 9Ghz i5 should be 6x faster??))



other than that...
As it seems the mainstream is flooded with CPUs well below 2Ghz, I think the same should be pretty ok for faster OS + optimized apps ... (even if the MIPS/Mhz per core is lower)


Too bad there is no demand for PPC desktop caliber chips at 3Ghz. As we have seen from game consoles (3Ghz PPC) and Power Servers (5+Ghz) PowerPC cores can handle very high GHz.

Last edited by KimmoK on 08-Mar-2013 at 07:18 AM.
Last edited by KimmoK on 08-Mar-2013 at 07:14 AM.
Last edited by KimmoK on 08-Mar-2013 at 07:13 AM.

_________________
- KimmoK
// For freedom, for honor, for AMIGA
//
// Thing that I should find more time for: CC64 - 64bit Community Computer?

@KimmoK

Quote:

How did you get to that 1/6 estimate?

Quote:

Perhaps you meant single thread performance?

Assuming 1.5 GHz G4 (2.3 DMIPS/MHz) would probably score around 4 SpecInt2006 (single core integer benchmark), 1.8 GHz e6500 core (single thread, 3.3 DMIPS/MHz) would score around 7-8 SpecInt2006. Current i5 CPUs give scores well above 40 SpecInt2006.

Has anyone compiled this for their Sam, X1000 etc. The source is open but I believe you aren't allowed distribute any binaries you make.

@pavlor

Methinks thou dost assume too much.

I don't have specint2006 for any powerpc chips, but the fastest i5 reported (3100MHz) scores 51361.07 in coremark, for 4 threads (2 cores). The T4240 scores 179763.04 for 24 threads.

If we do the numbers:
51361.07/4*1800/3100*24 = 178935.340645

So the i5 is actually slower per MHz per thread (and core) in coremark than the e6500.

I hope you'll follow up with some proof, not just assumptions.

Edit: slight typo.

Edit2: Now, obviously if you force the i5 into single-thread turbo mode you'll get slightly higher numbers for that one thread vs the e6500 which doesn't have said turbo mode. But good luck reaching tens of gigahertz even in turbo mode.

Edit3: Found a faster i5, still 3.1GHz. ~62000 total score, so a bit better. Still not worlds apart, though.

Last edited by olegil on 08-Mar-2013 at 06:12 PM.
Last edited by olegil on 08-Mar-2013 at 06:06 PM.
Last edited by olegil on 08-Mar-2013 at 06:04 PM.

_________________
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:

So the i5 is actually slower per MHz per thread (and core) in coremark than the e6500.

Coremark? Benchmark where 5200B can be faster/MHz than Core 2?


I don´t think 1800 MHz e6500 core will even surpass old 970MP 2500 MHz in single core operations (except probably AltiVec).

In older SpecInt2000 benchmark:
970MP 2500 MHz: 1586 SpecInt2000 (that is around 10-12 SpecInt2006)
Core 2 Q6600 2400 MHz: 2500 SpecInt2000 (18 SpecInt2006)

Core i5 in computer of my brother is more than 2 times faster (in sense Y=2*X) than our Core 2 Q6600 (single core operations - eg. emulators like DosBox, UAE, QEMU).

@pavlor

I don't really see anything wrong with the entry for the 5200B, at 2.81 coremark/MHz.

A couple of entries for the Core2Duo seems suspect, about half of what they should be. Could it be a case of single core performance being submitted as in place of dual core numbers?

_________________
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 really see anything wrong with the entry for the 5200B, at 2.81 coremark/MHz.

Then look at PowerPC 970: 2.22 coremark/MHz.


Coremark is useful benchmark, but one must be very careful when making conclusions.

@pavlor

What's wrong with that number? I don't mean, "why don't you like the number", I mean what's WRONG with the number?

The T4240 runs 24 threads and total of 99.87 per MHz, which gives 4.16125 per thread per MHz. Remember that each core in the T4240 is much closer to a dual-core than a single-core, only a very limited amount of resources is shared between the two executing threads.

But since you wanted single-core (I still suspect you mean single-thread), 4.16125 for e6500 compared to 2.81 for e300 and 2.22 for G5. I really don't see anything suspect about any of those numbers. Where is the discrepancy?

Edit: I think I see it. You mean e6500 and G5 should score higher due to Altivec than 5200B without it. Right? Well, I've benchmarked the 5121e a bit myself (doing practical tasks like jpeg decompression etc), but I only compared against my old desktop P4. The 400MHz 5121e scored about on par with a 700MHz P4 in that test (so about the same score per MHz as a P3). So slightly surprising numbers, but I wouldn't say wrong without more investigation. We had a similar discussion about the pA6T here last year, to sum up: Rather disappointing.

So Freescale designs this new core which doesn't have those disappointments and you refuse to believe it on the grounds of "IBM fucked up, Freescale can't possibly have done anything smarter than them"?

Last edited by olegil on 08-Mar-2013 at 08:33 PM.
Last edited by olegil on 08-Mar-2013 at 07:53 PM.

_________________
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:

You mean e6500 and G5 should score higher due to Altivec than 5200B without it. Right?

No. Old 603e simply can´t be faster/MHz than G5 or even Core i5 2500 (5.09 coremark MHz with 2 threads... 2.54 with one thread).

As you see, difference between Freescale CPUs is reasonable (all 1 thread):
5200B: 2.81 Coremark/MHz
e500mc: 3.1 Coremark/MHz
e6500mc: 4.16 Coremark/MHz (or 4.89 - two threads are as fast as 1.7x single core performance according to Freescale)

Assuming 400 MHz 5200B is half as fast as 800 MHz 440EP and assuming 800 MHz 440EP is roughly half as fast as 1133 MHz 7447, e6500 1800 MHz will be as fast as 1.9 GHz G4 (or 2.2 GHz). That is much lower value than my original estimate you disputed.

@pavlor

You jumped from from listing e300/e500mc/e6500 to assuming e300 vs 440 vs e600 vs e6500 without listing any numbers again. Do I really have to dig through all that to find the numbers? Why not just list the numbers you're using?

_________________
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.

@pavlor

I dispute the i5 coremark/MHz of 5.09, I'm fairly certain that's single-thread performance. That really sticks out like a sore thumb among the rest of the i5 coremarks.

_________________
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.

@pavlor

And an e300 (not 603e) CAN be faster per MHz than a G5. Not at Dhrystone, obviously. But coremark != Dhrystone. And Dhrystone is a pretty bad benchmark for modern computers.

_________________
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:

You jumped from from listing e300/e500mc/e6500 to assuming e300 vs 440 vs e600 vs e6500 without listing any numbers again.

e300 vs 440EP - benchmarks published on this very site confirm my opinion (lame, Blender, OGR etc.)
440EP vs G4 - same benchmarks.

Quote:

I dispute the i5 coremark/MHz of 5.09, I'm fairly certain that's single-thread performance. That really sticks out like a sore thumb among the rest of the i5 coremarks.

I chosed this result to clearly show how dangerous is to compare Coremark results. There are other even more "weird" results. 5200B is one of them.

Quote:

And an e300 (not 603e) CAN be faster per MHz than a G5. Not at Dhrystone, obviously. But coremark != Dhrystone. And Dhrystone is a pretty bad benchmark for modern computers.

5200B is 603e (at least according to Freescale webpage).

@pavlor

I still say the 5200B benchmark isn't weird. G5 has always underperformed compared to the theoretical numbers (could they have optimized the design for Dhrystone, forgetting the real world?)

And indeed I was wrong, 5200B is 603e, with 1.8DMIPS/MHz (I was thinking of the 5121e, which is e300 with 1.99DMIPS/MHz)

And it's much worse to compare Dhrystone than coremark, but I'm guessing you won't ever accept any good numbers for the T4240 no matter the benchmark.

Last edited by olegil on 08-Mar-2013 at 09:07 PM.

_________________
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:

And it's much worse to compare Dhrystone than coremark, but I'm guessing you won't ever accept any good numbers for the T4240 no matter the benchmark.

I accepted Coremark result of T4240, I only disagree with your comparison to result of Core i5/i7.

Only real "application benchmarks" can show, if e6500 is realy that fast. My bet is, it isn´t.

@olegil

5200B is e300 not 603e according to the manual from http://www.freescale.com/files/32bit/doc/ref_manual/MPC5200BUM.pdf&sa=U&ei=P1k6UfLbD4nCswbs3YBo&ved=0CCAQFjAB&usg=AFQjCNETXHXR6inKKZRWZVU_pnxu_AUrUA (page 1-5). The 5200B overview page links to the 5200 (_not_ 5200B) data sheet which incorporates a 603e (G2) core. I think that's an error on Freescales's site.
The 51xx and 52xx overview at http://www.freescale.com/webapp/sps/site/taxonomy.jsp?nodeId=01624606C10898 lists that correctly again: 5200B-e300, 5200-603e. Quite a mess that a single B is not just _a_ revision, but a revision with different core.

Last edited by Zylesea on 08-Mar-2013 at 09:45 PM.

_________________
My programs: via.bckrs.de
MorphOS user since V0.4 (2001)