[AT] AMD's Seattle server SoC

[SlimFan]

So Intel wasn't lying?!?

 

[Vesku]

Kabini @ 1.9GHz x 4 = 28.1 = 3.7 SpecInt/GHz/Core
Avoton @ 2.6GHz x 8 = 106 = 5.1 SpecInt/GHz/Core
AMD A57 @ 2GHz x 8 = 80 = 5 SpecInt/GHz/Core

Going to guess AMD's 80 number for the ARM SoC comes from using very fast memory, DDR4 perhaps.

 

[Abwx]

Kabini @ 1.9GHz x 4 = 28.1 = 3.7 SpecInt/GHz/Core
Avoton @ 2.6GHz x 8 = 106 = 5.1 SpecInt/GHz/Core
AMD A57 @ 2GHz x 8 = 80 = 5 SpecInt/GHz/Core

Going to guess AMD's 80 number for the ARM SoC comes from using very fast memory, DDR4 perhaps.

Theses are usefull numbers up to a a certain point only, i extracted the perfs of the rigs submitted by intel for consumers CPUs and supermicro for the Avoton out from SPEC s list :



C2750 i5 4670K A4 4000 A10 7850

SPECint2006

17.5 56.2 23.7 31.8

SPECint_rate

103 163 37.3 85.8

Cfp2006

23.3 66.8 23.6 34.3

Cfp2006_rate

77.3 126 30.5 64.1

The rates are integer and FP related bandwiths, the int and cfp scores are actual computations.

 

[Vesku]

Yes, it's definitely not a direct comparison. Manufacturers have little reason to ensure complete hardware parity to make consumers less confused. Confusion leads consumers to spend more according to some studies.

 

[Khato]

So Intel wasn't lying?!?

Such a surprise, isn't it?

 

[AtenRa]

http://www.spec.org/cpu2006/Docs/readme1st.html#Q15

Q15. What is the difference between a "rate" and a "speed" metric?
There are several different ways to measure computer performance. One way is to measure how fast the computer completes a single task; this is a speed measure. Another way is to measure how many tasks a computer can accomplish in a certain amount of time; this is called a throughput, capacity or rate measure.

- The SPECspeed metrics (e.g., SPECint2006) are used for comparing the ability of a computer to complete single tasks.
- The SPECrate metrics (e.g., SPECint_rate2006) measure the throughput or rate of a machine carrying out a number of tasks.

For the SPECrate metrics, multiple copies of the benchmarks are run simultaneously. Typically, the number of copies is the same as the number of CPUs on the machine, but this is not a requirement. For example, it would be perfectly acceptable to run 63 copies of the benchmarks on a 64-CPU machine (thereby leaving one CPU free to handle system overhead).

Note: a speed run which uses a parallelizing compiler to distribute one copy of a benchmark over multiple CPUs is still a speed run, and uses the SPECspeed metrics. You can identify such runs by the field "Auto Parallel".

 

[AtenRa]

Estimated ??? they havent run the benchmark ???

http://www.intel.com/content/www/us/en/benchmarks/atom/atom-microserver-specint-rate-base-2006.html

Intel Atom processor C2750: FOR.INTEL. cpu2006.1.2.ic14.0.15aug2013 Intel Atom processor server-based platform with one processor (8-core 2.4GHz, 20W, B0-stepping), Intel® Turbo Boost Technology enabled, 16GB memory (4x 4GB DDR3-1600 UDIMM ECC), 250GB SATA 7200RPM HDD, Red Hat Enterprise Linux 6.4. Estimated Score: SPECint_rate_base2006=106

 

[Abwx]

Yes, it's definitely not a direct comparison. Manufacturers have little reason to ensure complete hardware parity to make consumers less confused. Confusion leads consumers to spend more according to some studies.

You re right, indeed, using the rate as basis i could easily conclude that the C2750 is about 65% of a 4670K at 20% of the TDP while it s actualy only 30% computing wise, wich is still a better perf/watt ratio than the i5, though.

Estimated ??? they havent run the benchmark ???

http://www.intel.com/content/www/us/en/benchmarks/atom/atom-microserver-specint-rate-base-2006.html

Supermicro did test a real configuration with an Avoton , SPECint_rate :

http://www.spec.org/cpu2006/results/res2014q3/cpu2006-20140617-29941.html

The other SPEC tests can be found here, Supermicro is the last on the 4 lists of benchmarks :

http://www.spec.org/cpu2006/results/res2014q3/

 

[AtenRa]

A10-7850K at 93,7 peak

http://www.spec.org/cpu2006/results/res2014q3/cpu2006-20140715-30428.html

 

[Abwx]

A10-7850K at 93,7 peak

http://www.spec.org/cpu2006/results/res2014q3/cpu2006-20140715-30428.html

I found 85.8 and it was also submited by Intel, look like there s some variability, they used ICC as compiler in all cases, the 7850 was flagged up to AVX, the compiler has a CPU dispatcher anyway but it s still very usefull for intra AMD CPUs comparisons.

Such a surprise, isn't it?

Indeed but from the tests they published scores that are the less relevant as far as performance is a concern, looking through the details we can see that Avoton is rather weak to say the least, hence the cherry picked bench...

 

[SlimFan]

Indeed but from the tests they published scores that are the less relevant as far as performance is a concern, looking through the details we can see that Avoton is rather weak to say the least, hence the cherry picked bench...

Why do you say that? It's pretty clear that Avoton has been designed for multi-threaded workloads. That's been a pretty consistent message about the "microserver" market. I'd say quite the opposite from your conclusion. Avoton actually performs very well and shows excellent multi-core scaling on a fairly heavy benchmark.

If you want the highest single threaded performance, this probably isn't the CPU for you. But then again, neither is Seattle, or most of the AMD parts. The per-thread performance in this benchmark puts Avoton ahead of an A57 at the same core count and with a lower TDP.

 

[AtenRa]

You should also check the operating system and the compiler, the ATOM C2750 was tested in RedHat Server, while A10-7850K in Win 8.1

 

[Exophase]

Intel SPECint2006 results get an advantage from ICC being far better at (if not flat out breaking) some of the benchmarks. There's some comparisons here:

http://www.behardware.com/articles/847-9/the-impact-of-compilers-on-x86-x64-cpu-architectures.html