@KimmoK

power org q2 2014 newsletter
https://www.power.org/newsletter/power-org-q2-2014-newsletter/

Last edited by AmigaBlitter on 09-Aug-2014 at 09:09 AM.

_________________
retired

Servergy's Power server:

They hint about 1.5Ghz version:" Cleantech server consumes around 100 watts of power at maximum load".

But almost all it's details are unclear at this point.

UPDATE:
Some Power vs Xeon readings.
Geeky read optimize for POWER.

Last edited by KimmoK on 22-Aug-2014 at 09:03 AM.
Last edited by KimmoK on 22-Aug-2014 at 08:57 AM.

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

@KimmoK

A nice video on Power 8 systems.

https://www.youtube.com/watch?v=4ZyXc12J6FA

_________________
retired

IBM Power related stuff

Chinese "Power" chip scheduled to appear this year etc.:
http://openpowerfoundation.org/press-releases/openpower-foundation-technology-leaders-unveil-hardware-solutions-to-deliver-new-server-alternatives/

Power9 coming (around 2017?) ...
http://www.anandtech.com/show/8727/nvidia-ibm-supercomputers

More at:
https://morph.zone/modules/newbb_plus/viewtopic.php?topic_id=9463&forum=3

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

Power8, Power9 in a couple of years, Chinese server chips...

Very relevant stuff for Amiga indeed!

This chip should be now sampling:
"Suzhou PowerCore Technology Co. Ltd plans to deliver working versions of its CP1 (below) this summer. It will come in 8-, 10 and 12-core versions, each with a 64K data cache and 32K instruction cache. Like the IBM chip, the CP1 has 96MB of L3 cache, 115GB/s memory bandwidth and can support 32 lanes of PCIe 3.0."

More:
https://morph.zone/modules/newbb_plus/viewtopic.php?topic_id=9463&forum=3

IBM Powe8 prices start from USD2500 I think, interesting to see how will CP1 + TYAN motherboard be priced...

And IBM insist Power8 is superb in servers....
http://www.hpcwire.com/2015/06/09/ibm-power8-outperforms-x86-on-stac-benchmarks/

Last edited by KimmoK on 11-Jun-2015 at 10:47 AM.

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

@KimmoK

Although i'm a PowerPC fan, i have to admit that Intel is kicking hard:

http://www.hpcwire.com/2015/06/02/cpu-benchmarking-haswell-versus-power8/

_________________
retired

@AmigaBlitter

As has been argumented about a million times by the x86 crowd on this very forum:
benchmarks running in cache is cheating.

When you see GFLOP/s being 20 times the number of GB/s (20 instructions run for each byte fetched or stored in memory) you gotta ask yourself, where did the data come from?

IBM has a much more realistic performance per bandwidth number there.

_________________
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.

So, in short ... in some synthetic test code intel is faster, but in some more real life tests Power is 2x faster...?

They must be faster, otherwise no-one would buy & develop Power...

Quote:

In a blog post, Gupta writes that “STAC-A2 gives a much more accurate view of the expected performance as compared to micro benchmarks or simple code loops.”

++
http://cabotpartners.com/Downloads/crossing-the-performance-chasm-with%20openpower.pdf

+++
No news in coremark results...
http://www.eembc.org/coremark/

(e6500 systems have still highest coremark per core/mhz and coremark/watt unless I'm mistaken.
And the t4240 being the most powerfull single SoC device. Too bad we have to wait for blender benchmarks, heh...)
UPDATE. The most powerfull (in coremark) single CPU might be Intel Xeon E5 2687W at 3,4Ghz
http://www.cpu-world.com/CPUs/Xeon/Intel-Xeon%20E5-2687W.html ( 150W power )
UPDATE: And Core i7 860 result is about 1/3 of T4240...

Last edited by KimmoK on 11-Jun-2015 at 02:00 PM.
Last edited by KimmoK on 11-Jun-2015 at 01:58 PM.
Last edited by KimmoK on 11-Jun-2015 at 01:57 PM.
Last edited by KimmoK on 11-Jun-2015 at 01:54 PM.
Last edited by KimmoK on 11-Jun-2015 at 01:48 PM.
Last edited by KimmoK on 11-Jun-2015 at 01:45 PM.
Last edited by KimmoK on 11-Jun-2015 at 01:39 PM.
Last edited by KimmoK on 11-Jun-2015 at 01:36 PM.

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

@KimmoK

Take the coremark per core, divide by frequency of e5 (3400). Multiply by frequency of T4 (1800). Result: 13239.1535294. That's pretty badly beaten by the T4 with its 15656.13. The Intel way of doing multithreading really does not impress me. On P4 they were getting 20% extra, now they're getting 20% extra.

While freescale manages 70-80% extra when running two threads.

But I guess it matters nothing to the consumer when they manage to clock it so much higher.

_________________
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.

There is a guy on Real world tech forums with a POWER8 machine - he says he ran multiple benchmarks and in single thread POWER8 is about as fast as Sandy Bridge. Only in multithreaded enviroment it can compare to Haswell(IBM designed POWER8 primarily for multiple threads). Haswell core is much more sophisticated with greater IPC and better vector unit. Skylake Xeons will probably have SMT4(4 threads per core), Knight's Hill is confirmed to have SMT4(Knights's hill uses a highly modified Silvermont-Atom core).
We'll see what POWER9 brings but so far Intel is kicking butt(and at much lower TDP)

_________________

@olegil

Quote:

olegil wrote: @AmigaBlitter As has been argumented about a million times by the x86 crowd on this very forum: benchmarks running in cache is cheating.

IBM's POWER8 has 96MB of cache, whereas Intel's Xeon E5v3 has only 40MB. So IBM is cheating...
Quote:

When you see GFLOP/s being 20 times the number of GB/s (20 instructions run for each byte fetched or stored in memory) you gotta ask yourself, where did the data come from?

By the more cache usage? Floating point code tends to be more linear and localized, so it makes better use of caches. That's also the reason why the SIMD paradigm became so much important and widespread: because it's common to manipulate local data doing the same operations.
Quote:

IBM has a much more realistic performance per bandwidth number there.

With so much big cache it should run in cache more often than Intel's chip. Cheating?

@olegil

Quote:

olegil wrote: @KimmoK Take the coremark per core, divide by frequency of e5 (3400). Multiply by frequency of T4 (1800). Result: 13239.1535294. That's pretty badly beaten by the T4 with its 15656.13. The Intel way of doing multithreading really does not impress me. On P4 they were getting 20% extra, now they're getting 20% extra. While freescale manages 70-80% extra when running two threads. But I guess it matters nothing to the consumer when they manage to clock it so much higher.

Because single thread/core performance still counts.

@WolfToTheMoon: can you post some link? Thanks.

@cdimauro

Don't bother. All benchmarks are invalid here unless they show that Power is superior solution.
Benchmark of choice here seems to be coremark which is made for embedded solutions which have quite different needs that desktop programs.

EDIT2: haven't caught the irony. Sorry. Better to go sleep.

Last edited by cdimauro on 11-Jun-2015 at 08:05 PM.
Last edited by cdimauro on 11-Jun-2015 at 08:03 PM.
Last edited by cdimauro on 11-Jun-2015 at 08:01 PM.

I report some numbers from another comment:

Intel® Xeon® Processor D (8 cores, 1.9 GHz, 45 W, ES2): SPECint*_rate_base2006 (16 copy) = 268

But I report also some data about the Avoton family, which is comparable to the T4240 one, as I state before (sorry for the typo).
Intel® Atom™ processor C2750 (8 cores, 2.4 GHz, 20 W): SPECint_rate_base2006 (8 copy) = 102

@cdimauro

As someone who actually works in field of HPC KimmoK link tells me what I know already.

"Only 10 percent less performance than the best-performing solution, a Supermicro server with two 10-core Intel Xeon E5-2690 v2 @ 3.0GHz (Ivy Bridge) and one NVIDIA K80 GPU accelerator."

You can get that setup for ~$10k. Guess what you pay for the system which is only 10% slower than that? 65k+tax.

Yes. There are things where Power8 is superior compared to X86 but on those it will get utterly destroyed by CUDA.

@Rose: "very competitive".

BTW, have you some experience with some Xeon Phi?

http://www.freescale.com/webapp/sps/site/prod_summary.jsp?code=DUAL-QORIQ-T4240&fsrch=1&sr=2&pageNum=2#pspFeatures

_________________
retired

@AmigaBlitter

It looks like a real powerhouse.

Powerfull indeed.

(they have had 40 virtual cores LSxxxxP processor in the roadmap in 2013, that should deliver almost the same as that dual processor system linked above)

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