Do they ever do compiler speed tests or boot speed tests? Which of the existing ones would be most similar?
GCC is part of spec2006, which they optimize for. But the actual compilers people use (IE not that version of GCC) are rarely well-optimized for CPU. In fact, most compiler profiles are not flat yet, and where your compilation takes a long time, it's because some optimization or algorithm has gone nuts and needs to be fixed. Until that kind of state of the world changes, measuring compile time is IMHO, a bit pointles…
>In fact, most compiler profiles are not flat yet, and where your compilation takes a long time, it's because some optimization or algorithm has gone nuts and needs to be fixed.
For some reason one of my sideprojects needs more than two minutes to link but the task manager tells me it's neither CPU nor RAM nor disk bound. I haven't found the cause yet.
I'm someone who's still running an ancient (2007, IIRC) Nehalem i7-920 based workstation for daily use and was eagerly waiting for this launch to finally decide on a new build. After reviewing all the day one benchmark data, I've come to the odd decision that I'm probably going for an i7-7700k, especially if Intel drops the prices by even a tiny amount. I say "odd" because, apart from basic daily tasks and occasional…
You're right about the AVX2 part. If that's important to you. Otherwise I'd have to disagree, waiting all those years and going to a quadcore is sad. This is a monumental shift to see $330 8C/16T chips start to be mainstream. I also hate that Intel went with that cheap TIM that requires delidding. You're in a tough spot given the focus on AVX2 IMO. My thoughts on Ryzen in general though is that the base 1700 is the chip for everyone building completely from scratch in 2017 and beyond. I personally won't be building another quadcore from this point onward. Just feel too much like a sucker to do so and I do think it's foolish (even for gaming! ... where almost everyone's system is going to be GPU bound at 1080P or higher anyway) unless you have some specific need like what you're describing with a massive 50% penalty affecting you.
I'm optimistic, but I'd like to see benchmarks for compiling code. Right now I buy really expensive 8 core cpus, because multiple threads do speed up encoding quite a bit. If Ryzen can match that at half the price, it will be much welcome.
ECC is a feel good feature, but is mostly useless. Memory either is corrupted (which can be detected by running MemTest) or it works perfectly fine without ECC. I have over 50 10+ year old Itanium servers with 192GB ECC RAM each, there are 2 ECC errors logged total, while running 24/7.
ECC is great for detecting errors more than correcting them. Then one can do thorough memtests and decide if it's time to replace memory modules. It doesn't really prevent problems (there are unrepairable ECC errors too) as much as it leads to early detection and repair. Well worth it when you're optimizing for MTTR. It's not a silver bullet if the user is just paranoid about data integrity (nothing beats end-to-end…
Personal anecdote. I have a Phenom + Asus mobo system with ECC memory in it that has been working fine for years without reporting a single ECC error. And then they started. At first there was a correctable error once every few days, then few errors per day, and then almost a constant stream of errors. Fortunately, they were all correctable. From the reported ECC syndomes I deduced that it was always the same single bit lane. That's why only a single ECC symbol was affected and the errors were correctable. I reseated the CPU and DIMMs and the errors went away, and the system keeps chugging on. Not sure what it was, maybe some dust particle getting in a wrong place, maybe some oxidation.
Anyway, I can only imagine how much trouble those errors could do if they were not detected and corrected. Especially, at the stage when there was a moderate number of them. I guess when they would become too many, then the system would just crash all the time.
Wow. AMD has really pulled through on this one. They are really giving Intel a run for their money. This is going to be great for consumers. What is interesting is the HUGE jump in multi-threaded performance on AMD. Very interesting.
> What is interesting is the HUGE jump in multi-threaded performance on AMD. Very interesting. Indeed. Single threaded performance, however, seems bad. Virtual cores most likely share less of the actual core than on Intel parts, allowing one to compete less with the other and get more done. On my i7 laptop, on anything more demanding, I see half the "cores" idling, probably because one is using the actual underlying…
Single threaded perf isn't bad, it's just not outstanding. Many of the cases where I care about perf are cases in which the app is multithread savvy.
But it's still losing in every web and office single threaded benchmark. EDIT: compared to 7700k that costs about the same
Or, seen another way (better representation of the overall reality): The 7700k is losing in all aspects below w.r.t. the Ryzen 7 1800X, except absolute single-threaded perf. Intel is losing in: - absolute multi-threaded performance - relative multi-threaded performance per $ - memory bandwidth - L3-cache bound workloads (AMD has twice more L3) - price of motherboards
> What is interesting is the HUGE jump in multi-threaded performance on AMD. Very interesting. Indeed. Single threaded performance, however, seems bad. Virtual cores most likely share less of the actual core than on Intel parts, allowing one to compete less with the other and get more done. On my i7 laptop, on anything more demanding, I see half the "cores" idling, probably because one is using the actual underlying…
Hyperthreading is not a virtual core and as it doesn't exist as a separate core its hard for it to be idling. Bad analogies make for less than adequate understanding.
They are also called "logical cores" (I think Intel documents prefers "logical" and Sun - they invented the concept - prefers "virtual"). One logical core can appear to be idling to the OS if the parts of the physical core are in use by another (SPARC and POWER go up to 8) logical core (as it's resource starved). It seems, from the benchmarks, the AMD parts have less shared resources between logical cores (or a finer granularity for allocating) than the i7 in my laptop.
> What is interesting is the HUGE jump in multi-threaded performance on AMD. Very interesting. Indeed. Single threaded performance, however, seems bad. Virtual cores most likely share less of the actual core than on Intel parts, allowing one to compete less with the other and get more done. On my i7 laptop, on anything more demanding, I see half the "cores" idling, probably because one is using the actual underlying…
>"half the "cores" idling" Right, thats exactly what happens with Hyper-Threading. If you can fully utilize the physical core, Hyper-Threading can't do much.
It seems the AMD parts can squeeze a little more work from the two threads, judging by the multi-threeaded benchmarks, than competing Intel parts.
You do have to consider that this is an 8 core, and more and more applications are being developed with this in mind. Whoever still thinks that Single threaded performance is the most important, lives in the past.
Consumer applications? Doubt it. Not a whole lot of consumer-oriented applications parallelize that well to be honest.
don't you run more than one of them at a time though?