Sizing Up Servers: Intel's Skylake-SP Xeon versus AMD's EPYC 7000 - The Server CPU Battle of the Decade?
by Johan De Gelas & Ian Cutress on July 11, 2017 12:15 PM EST- Posted in
- CPUs
- AMD
- Intel
- Xeon
- Enterprise
- Skylake
- Zen
- Naples
- Skylake-SP
- EPYC
SMT Integer Performance With SPEC CPU2006
Next, to test the performance impact of simultaneous multithreading (SMT) on a single core, we test with two threads on the same core. This way we can evaluate how well the core handles SMT.
| Subtest | Application type | Xeon E5-2690 @ 3.8 | Xeon E5-2690 v3 @ 3.5 | Xeon E5-2699 v4 @ 3.6 | EPYC 7601 @3.2 | Xeon 8176 @ 3.8 |
| 400.perlbench | Spam filter | 39.8 | 43.9 | 47.2 | 40.6 | 55.2 |
| 401.bzip2 | Compression | 32.6 | 32.3 | 32.8 | 33.9 | 34.8 |
| 403.gcc | Compiling | 40.7 | 43.8 | 32.5 | 41.6 | 32.1 |
| 429.mcf | Vehicle scheduling | 44.7 | 51.3 | 55.8 | 44.2 | 56.6 |
| 445.gobmk | Game AI | 36.6 | 35.9 | 38.1 | 36.4 | 39.4 |
| 456.hmmer | Protein seq. analyses | 32.5 | 34.1 | 40.9 | 34.9 | 44.3 |
| 458.sjeng | Chess | 36.4 | 36.9 | 39.5 | 36 | 41.9 |
| 462.libquantum | Quantum sim | 75 | 73.4 | 89 | 89.2 | 91.7 |
| 464.h264ref | Video encoding | 52.4 | 58.2 | 58.5 | 56.1 | 75.3 |
| 471.omnetpp | Network sim | 25.4 | 30.4 | 48.5 | 26.6 | 42.1 |
| 473.astar | Pathfinding | 31.4 | 33.6 | 36.6 | 29 | 37.5 |
| 483.xalancbmk | XML processing | 43.7 | 53.7 | 78.2 | 37.8 | 78 |
Now on a percentage basis versus the single-threaded results, so that we can see how much performance we gained from enabling SMT:
| Subtest | Application type | Xeon E5-2699 v4 @ 3.6 | EPYC 7601 @3.2 | Xeon 8176 @ 3.8 |
| 400.perlbench | Spam filter | 109% | 131% | 110% |
| 401.bzip2 | Compression | 137% | 141% | 128% |
| 403.gcc | Compiling | 137% | 119% | 131% |
| 429.mcf | Vehicle scheduling | 125% | 110% | 131% |
| 445.gobmk | Game AI | 125% | 150% | 127% |
| 456.hmmer | Protein seq. analyses | 127% | 125% | 125% |
| 458.sjeng | Chess | 120% | 151% | 125% |
| 462.libquantum | Quantum sim | 91% | 129% | 90% |
| 464.h264ref | Video encoding | 101% | 112% | 112% |
| 471.omnetpp | Network sim | 109% | 116% | 103% |
| 473.astar | Pathfinding | 140% | 149% | 137% |
| 483.xalancbmk | XML processing | 120% | 107% | 116% |
On average, both Xeons pick up about 20% due to SMT (Hyperthreading). The EPYC 7601 improved by even more: it gets a 28% boost on average. There are many possible explanations for this, but two are the most likely. In the situation where AMD's single threaded IPC is very low because it is waiting on the high latency of a further away L3-cache (>8 MB), a second thread makes sure that the CPU resources can be put to better use (like compression, the network sim). Secondly, we saw that AMD core is capable of extracting more memory bandwidth in lightly threaded scenarios. This might help in the benchmarks that stress the DRAM (like video encoding, quantum sim).
Nevertheless, kudos to the AMD engineers. Their first SMT implementation is very well done and offers a tangible throughput increase.
Facebook
Twitter
RSS
219 Comments
View All Comments
oldlaptop - Thursday, July 13, 2017 - link
Why on earth is gcc -Ofast being used to mimic "real-world", non-"aggressively optimized"(!) conditions? This is in fact the *most* aggressive optimization setting available; it is very sensitive to the exact program being compiled at best, and generates bloated (low priority on code size) and/or buggy code at worst (possibly even harming performance if the generated code is so big as to harm cache coherency). Most real-world software will be built with -O2 or possibly -Os. I can't help but wonder why questions weren't asked when SPEC complained about this unwisely aggressive optimization setting...peevee - Thursday, July 13, 2017 - link
"added a second full-blown 512 bit AVX-512 unit. "Do you mean "added second 256 ALU, which in combination with the first one implements full 512-bit AVX-512 unit"?
peevee - Thursday, July 13, 2017 - link
"getting data from the right top node to the bottom left node – should demand around 13 cycles. And before you get too concerned with that number, keep in mind that it compares very favorably with any off die communication that has to happen between different dies in (AMD's) Multi Chip Module (MCM), with the Skylake-SP's latency being around one-tenth of EPYC's."1/10th? Asking data from L3 on the chip next to it will take 130 (or even 65 if they are talking about averages) cycles? Does not sound realistic, you can request data from RAM at similar latencies already.
AmericasCup - Friday, July 14, 2017 - link
'For enterprises with a small infrastructure crew and server hardware on premise, spending time on hardware tuning is not an option most of the time.'Conversely, our small crew shop has been tuning AMD (selected for scalar floating point operations performance) for years. The experience and familiarity makes switching less attractive.
Also, you did all this in one week for AMD and two weeks for Intel? Did you ever sleep? KUDOS!
JohanAnandtech - Friday, July 21, 2017 - link
Thanks for appreciating the effort. Luckily, I got some help from Ian on Tuesday. :-)AntonErtl - Friday, July 14, 2017 - link
According to http://www.anandtech.com/show/10158/the-intel-xeon... if you execute just one AVX256 instruction on one core, this slows down the clocks of all E5v4 cores on the same socket for at least 1ms. Somewhere I read that newer Xeons only slow down the core that executes the AVX256 instruction. I expect that it works the same way for AVX512, and yes, this means that if you don't have a load with a heavy proportion of SIMD instructions, you are better off with AVX128 or SSE. The AMD variant of having only 128-bit FPUs and no clock slowdown looks better balanced to me. It might not win Linpack benchmark competitions, but for that one uses GPUs anyway these days.wagoo - Sunday, July 16, 2017 - link
Typo on the CLOSING THOUGHTS page: "dual Silver Xeon solutions" (dual socket)Great read though, thanks! Can finally replace my dual socket shanghai opteron home server soon :)
Chaser - Sunday, July 16, 2017 - link
AMD's CPU future is looking very promising!bongey - Tuesday, July 18, 2017 - link
EPYC power consumption is just wrong. Somehow you are 50W over what everyone else is getting at idle. https://www.servethehome.com/amd-epyc-7601-dual-so...Nenad - Thursday, July 20, 2017 - link
Interesting SPECint2006 results:- Intel in their slide #9 claims that Intel 8160 is 2% faster than EPYC 7601
- Anandtech in article tests that EPYC 7601 is 42% faster than Intel 8176
Those two are quite different, even if we ignore that 8176 should be faster than 8160. In other words, those Intel test results look very suspicious.