Phoronix were recently given access to three servers running three different POWER9 Talos II configurations and compared them to EPYC and Xeon. On paper these systems look amazing, thanks to the architecture supporting four threads per core; they tested a dual 4-core Talos II system, a Talos II Lite with a single 22-core CPU and a Talos II with dual 18-core processors with thread counts of 32, 88, and 144 respectively.
There were certainly usage scenarios where the dual 18 core system could outpace even the EPYC 7601 but could not surpass the dual Xeon Gold 6138 system. The review covers a fair amount of benchmarks and configurations but doesn't begin to scratch the surface of wide variety of server configurations you need to consider before abandoning POWER9 altogether but the key metric, performance per dollar, shows these architecture solidly in the middle of the pack.
"Back in April we were able to run some IBM POWER9 benchmarks with remote access to the open-source friendly Talos II systems by Raptor Computer Systems. We were recently allowed remote access again to a few different configurations of this libre hardware with three different POWER9 processor combinations. Here are those latest benchmarks compared to Intel Xeon and AMD EPYC server processors."
Here are some more Processor articles from around the web:
- AMD Ryzen 7 2700 and Ryzen 5 2600 @ Modders-Inc
- Intel Pentium Gold G5600 3.9 GHz @ TechPowerUp
- Intel Core i7-8086K @ TechARP
- Intel Core i7 8086K Linux Performance @ Phoronix
- Intel Kaby Lake G Core i7-8705G @ TechSpot
I’ll not fully trust Phoronix
I’ll not fully trust Phoronix as much as the Server oriented websites and those Talos systems were done with Remote access by Phoronix. But still for some uses maybe Power9’s PCIe 4.0 support will come in handy but who knows without having the hardware in house to really see how things can be tested.
The Phoronix Closing Remarks stated:
“Overall, the POWER9 performance is looking increasingly competitive with Intel Xeon and AMD EPYC server hardware but there still are some areas that could use improvement around multimedia/encoding and interpreted languages. The added benefit to POWER9 with the Talos II is that these systems manufactured in the USA are among the most fully open-source systems available today” (see article link provided above)
Those are OpenPower Power9s in this system so there will be plenty of OpenPower licensees(In Addition to Talos) to help make things a 3 way competition. And maybe begin to Include Cavium’s ThunderX2 ARM servers in there also with their 4 threads per core that are using that Broadcomm Vulcan Processor IP that Broadcom Sold off to Cavium after Broadcom was acquired by Avago(stupid for selling that Vulcan Processor IP in the first place). Now Marvell has acquired Cavium and some damn good custom ARM core server IP.
I was surprised at how
I was surprised at how competitive the Power9 was.
Member a few years ago when
Member a few years ago when every single post was over-ridden in comments by that anon power8 fan who kept calling us all idiots and screaming about the future of power8 32 core tablets and stuff and making it annoying as hell just to come to the site? Good times!
I wonder what happened to that dushbag?
It may lag behind on
It may lag behind on performance, but it leads on liberty.
Those who forsake the latter to gain a little of the former…
Also, 4 threaded Power 9 CPUs
Also, 4 threaded Power 9 CPUs are just one type, 8 threaded ones should be way more powerful (with less cores though). And they are already shipping them.
The 8 way smt models
The 8 way smt models basically fuse two cores and only really exist for per core licensing server apps. ie. some apps charge per “core” this way it’s cheaper because there are less cores.
The L4 eDRAM model that’s
The L4 eDRAM model that’s upcoming is supposed to double memory bandwidth:
“Now, it is getting close for ‘Zeppelin’ and ‘Fleetwood/Mack’ to take the Power9 stage with the ‘Cumulus’ Power9 variant, which sports up to 12 cores per chip and eight threads per core and which also makes use of the ‘Centaur’ memory buffer chip and L4 cache memory to provide roughly twice the memory capacity and bandwidth per socket.
This is the big iron that many IBM i and AIX customers have been waiting for, and maybe even a few customers doing in-memory and other kinds of analytics processors on Linux who are sick and tired of managing clusters to get big jobs done. For NUMA ware applications, a NUMA system is just so much easier to program, even if it does cost a bit more. (Not if you count the cost of that programming on cluster systems, of course.)”