A couple followups to the Blackbird semi-review
For proof, Phoronix now has Blackbird benchmarks of that same basic single-4 (16 thread) system. They also tried to measure the performance impact of the Spectre mitigations, which Raptor enables on every machine they ship (both kernel and user-level), and discovered that while there is an impact on POWER9 of about 8% it is still rather less than the effect of the mitigations on Intel chips (about 18%, not counting microarchitectural mitigations required to deal with ZombieLoad et al). There's a little weirdness in the numbers in that sometimes the mitigated throughput was slightly better performing, possibly some unaccounted-for statistical artifact, but it's good to see that the price paid for better security on our systems is rather less than on x86_64. That said, if you really want to wring out that extra 8%, you can configure the protection level in the BMC based on your anticipated risk.
Finally, a big thumbs up to Red Hat, who fixed this ppc64le-specific bug in LibreOffice I reported via Dan Horák in record time.
This seems as good a place as any to ask those in the know: should I buy a POWER9 system now, or continue to soldier along with my 2014-era x86 system and hold out for a POWER10 Talos Real Soon Now? I've been sitting on the fence for so long that the fence is starting to show permanent conformity to my backside, and CPUs don't really get better with age.
ReplyDeleteOnly Raptor can say how far away they would be from a P10 system, but it took at least a year for the T2 to emerge after the P9 was introduced, so I'd say you'd be waiting until late 2020 or early 2021 at the very earliest -- and that assumes the 2020 release date for P10 is in any way accurate. Meanwhile, there is at least one more generation of P9 to come.
DeleteIt'll still have to be your call, but in my relatively uninformed estimation I don't see P10 desktop systems being close, while good P9 systems are available now and work well.
The bug in LibreOffice was another example that ppc64le is a good platform for finding bugs in code that other platforms magically survive (or ignore) :-) And really kudos to Caolán for the prompt fix.
Delete"price paid for better security on our systems is rather less than on x86_64"
DeleteThere is AMD too in the x86-64 world, you know. Not sure how it looks in Phoronix's tests but the impact of Spectre mitigations and SpeculativeStoreBypass is much smaller (ARM's/Power's/Intel's Meltdown and most of the Intel-only holes are not applicable).
https://www.phoronix.com/scan.php?page=article&item=mds-zombieload-mit&num=10
DeleteAll current mitigations (including MDS vuln.) have -16 % performance hit on Intel, but just -3% performance hit on AMD CPUs (Zen) according to Phoronix.
AMD does have mandatory PSP though, which was already found to contain some nasty security holes including their vaunted "encrypted memory" leaking keys. Highly privileged black boxes that implement DRM are just not something you want anywhere near a "secure" system unless that system is running your own DRM code and obeying you alone -- AMD signs the PSP firmware with their vendor key, and the hardware enforces it, so that's right out.
DeleteIntel and AMD both weakened their systems for speed and DRM. That's an acceptable tradeoff for what 90% of consumers use an x86 box for (Netflix and games), but you can't use the word "secure" to describe the result unless you're referring to keeping content locked away with the DMCA as the weapon of choice.
Choices, as always.
Intel, AMD and even ARM are more foolish now than ever, on multiple accounts.
ReplyDeleteGlad to learn the Blackbird sold better than Raptor anticipated!