The IBM POWER8 Review: Challenging the Intel Xeon
by Johan De Gelas on November 6, 2015 8:00 AM EST- Posted in
- IT Computing
- CPUs
- Enterprise
- Enterprise CPUs
- IBM
- POWER
- POWER8
"Per Core" Integer Performance: 7-Zip
The profile of a compression algorithm is somewhat similar to many server workloads: it can be hard to extract instruction level parallelism (ILP) and it's sensitive to memory parallelism and latency. The instruction mix is a bit different, but it's still somewhat similar to many server workloads. Testing single threaded is also a great way to check how well the turbo boost feature works in a CPU.
We ran this benchmark on the POWER8 a few months ago, but there are several reasons to do this again. First of all, we can now use GCC 4.9.2, which has specific support for POWER8 (-mcpu=power8). It is good to note that POWER8 is not a radical new design compared to POWER7. So we only expect modest gains from the compiler.
Secondly, last time we ran on top of PowerKVM, inside a virtual machine. Although that should not make a big difference either - as the benchmark runs almost completely (99%) in user modus and thus runs at 100% - it's still worthwhle to rule out the influence of the virtual machine.
So we recompiled the 7-Zip source code on every machine with the -O3 optimization with GCC 4.9.2.
It is important to note that Intel is extremely aggressive with Turbo-boost on the Xeon E5-2699v3. Running code on one core causes the 2.3 GHz Xeon to boost to 3.6 GHz. As a result, the typical clockspeed advantage of the POWER8 was minimized to a measly 90 MHz, with the POWER8 CPUs boosting from 3.425 GHz to 3.690 GHz.
We found that the POWER8 needs more than one thread to deliver good performance: with one thread we only achieve 62% of the performance of a Haswell core at the same speed. Using the mcpu=power8 compiler flag did little more than boost the performance by 1-3%, which is within the margin of error of this benchmark. So your (occassional?) single threaded code will fare badly on POWER8.
Once you fire off 8 threads however, the POWER8 CPU outperforms the hyperthreaded Haswell core slightly (4%).
How about decompression which is even more (IPC) unfriendly to our brainiacs?
With a single thread, performance of a POWER8 core is about 25% slower than a Haswell core. The Haswell core is still clearly better in extracting Instruction Level Parallelism out of this ILP-unfriendly code. However, let there be no mistake about the integer crunching power of POWER8: it delivers 35% higher performance than the hyperthreaded Xeon E5, core per core, clock per clock (give or take a few MHz).
Compression depends more on the datacache and OoO engine. It is remarkable that the Haswell core with its smaller L1-datacache does a lot better than the POWER8. The many unpredictable branches of the decompression code underutilize these very wide modern cores, and as a result the SMT-8 capable POWER8 outperforms the dual-threaded (SMT-2) Haswell. Notice that running two threads instead one thread on the POWER8 offers 61% better performance. Running 8 threads delivers 2.4x higher performance, a clear indication that the POWER8 CPU has a very wide integer execution engine, but can only deliver if enough threads are active.
Facebook
Twitter
RSS
146 Comments
View All Comments
usernametaken76 - Thursday, November 12, 2015 - link
Technically this is not true. IBM had a working version of AIX running on PS/2 systems as late as the 1.3 release. Unfortunately support was withdrawn and future releases of AIX were not compiled for x86 compatible processors. One can still find a copy of this release if one knows where to look. It's completely useless to anyone but a museum or curious hobbyist, but it's out there.zenip - Friday, November 13, 2015 - link
...>--click here-Steven Perron - Monday, November 23, 2015 - link
Hello Johan,I was reading this article, and I found it interesting. Since I am a developer for the IBM XL compiler, the comparisons between GCC and XL were particularly interesting. I tried to reproduce the results you are seeing for the LZMA benchmark. My results were similar, but not exactly the same.
When I compared GCC 4.9.1 (I know a slightly different version that you) to XL 13.1.2 (I assume this is the version you used), I saw XL consistently ahead of GCC, even when I used -O3 for both compilers.
I'm still interested in trying to reproduce your results, so I can see what XL can do better, so I have a couple questions on areas that could be different.
1) What version of the XL compiler did you use? I assumed 13.1.2, but it is worth double checking.
2) Which version of the 7-zip software did you use? I picked up p7zip 15.09.
3) Also, I noticed when the Power 8 machine was running at full capacity (for me that was 192 threads on a 24 core machine), the results would fluctuate a bit. How many runs did you do for each configuration? Were the results stable?
4) Did you try XL at the less aggressive and more stable options like "-O3" or "-O3 -qhot"?
Thanks for you time.
Toyevo - Wednesday, November 25, 2015 - link
Other than the ridiculous price of CDIMMs the power efficiency just doesn't look healthy. For data centers leasing their hardware like Amazon AWS, Google AppEngine, Azure, Rackspace, etc, clients who pay for hardware yet fail to use their allocation significantly help the bottom line of those companies by reduced overheads. For others high usage is a mandatory part of the ROI equation during its period as an operating asset, thus power consumption is a real cost. Even with our small cluster of 12 nodes the power efficiency is a real consideration, let alone companies standardizing toward IBM and utilising 100s or 1000s of nodes that are arguably less efficient.Perhaps you could devise some sort of theoretical total cost of ownership breakdown for these articles. My biggest question after all of this is, which one gets the most work done with the lowest overheads. Don't get me wrong though, I commend you and AnandTech on the detail you already provide.
AstroGuardian - Tuesday, December 8, 2015 - link
It's good to have someone challenging Intel, since AMD crap their pants on regular basisdba - Monday, July 25, 2016 - link
Dear Johan:Can you extrapolate how much faster the Sparc S7 will be in your Cluster Benchmarking,
if the 2 on Die Infiniband ports are Activated, 5, 10, 20% ???
Thank You, dennis b.