AMD FX-8150 Bulldozer On Ubuntu Linux

[nepwk]

Phoronix benchmarking had issues at least in the past by using a broken methodology (comparing perfomance without having states the expectations, the methodology or making it in a bogus way, testing OSes some with composite enable, some don't, and so on).

I agree with a lot of what you say, of course there are inconsistencies. However, compared to other benchmark sites, like Anandtech continuing to use Futuremark even after the entire industry cast shame on it as being Intel-biased, etc... OB.org is actually pretty reasonable, as long as you understand that there may be inconsistencies. Unlike most of the other benchmark sites, the inconsistencies don't always favor Intel,<sarcasm> by coincidence </sarcasm>, of course.

[Kivada]

Their Result site sounds not very possitive

Maybe not in respect to wattage, but have a look at the Mad Shrimps tests, they're very thorough in testing the new platform's limits.

[Kivada]

I agree with a lot of what you say, of course there are inconsistencies. However, compared to other benchmark sites, like Anandtech continuing to use Futuremark even after the entire industry cast shame on it as being Intel-biased, etc... OB.org is actually pretty reasonable, as long as you understand that there may be inconsistencies. Unlike most of the other benchmark sites, the inconsistencies don't always favor Intel,<sarcasm> by coincidence </sarcasm>, of course.

Yeah, its very annoying to see them still picking "tests" that are basically designed specifically for Intel CPUs. If it was compiled in ICC then it's not a reliable benchmark.

[raj7095]

Right. This article tells us that if you use modern software that scales well with semi-real cores, that Bulldozer just may be the best bang for the buck. Of course, you chose to take one tiny part of my post completely out of context, rather than to argue my actual point, which pretty much says it all.



I never said you should. Stop trying to take things I say out of context.



No, you said that this article was "unfair" because it didn't have a hyperthreaded CPU, as if that would suddenly make Intel shine, you didn't say it would be "nice". Big difference. Now you retract it, and that's fine, but I stand by my point about hyperthreading, it's not going to change the fact that AMD caught up quite a bit with this new CPU.

I'm not sure why you feel so strongly about Bulldozer, surely you've never even been in the same room as one before to have an opinion. For that matter, you quite possibly have never used a Sandy Bridge machine either, I'm not sure why your panties are in a wad just because Phoronix/OB.org, quite possibly the most scientific and unbiased benchmark site in the world, has declared that Bulldozer beat Sandy Bridge in a few benchmarks, and still loses in many. Considering how open and automated OB.org is, Michael would have a really hard time pulling off sleazy benchmarketing propaganda like Tom, Anand and others do.

well, i am just trying to decide between buying a 2600k and fx-8150 for a friend of mine. he uses linux too, and he says he needs multithreaded performance, and doesn't care much about single threaded performance. and calling it unfair was my bad, just used a wrong word. and yes, i have never used a bulldozer before, but i do own a 2600k right now. I tested out turning hyperthreading on and off a few times, and it definitely seems to make difference in multithreaded performance. now, in multithreaded benchmarks, bulldozer seems really amazing compared to 2500k. i want to see if hyperthreading is good enough to beat it, or if he should just go for fx-8150. what do you think?

[nepwk]

I tested out turning hyperthreading on and off a few times, and it definitely seems to make difference in multithreaded performance.

I would totally expect the operating system to be more responsive under load with hyperthreading turned on, but actually improving the performance of said load is very inconsistent. At any time, your computer may have thousands of threads running, but most of the time the threads are just sleeping. Hyperthreading is good at allowing some of those low-intensity operating system threads to still do work while the cores are maxed-out working on something, but from my personal experience, the occasional performance gains are never justified by the times that HT actually hurts performance.

i want to see if hyperthreading is good enough to beat it, or if he should just go for fx-8150. what do you think?

Without knowing exactly what your friend's doing, it's a little hard to make a good recommendation. If he's creating a Linux home server or something similar, I'd have to lean torwards the Bulldozer solution, as AMD's desktop CPUs are comparable to server-class Opterons in both features and number of cores, whereas Intel gimps their non-Xeon CPUs by removing various features like certain advanced features of virtualization acceleration, and offers fewer real cores for the money(which is more important for servers than single-threaded performance).

I intend to buy a Bulldozer system for that exact purpose as soon as availability improves. I've built Linux servers out of Phenom II X6 CPUs, and they do an admirable job of being a cheap server. From everything I've seen, it looks like Bulldozer would do even better.

[fmmjqtft]

I added two tests sets to give a vague idea of how the 2600k compares:

http://openbenchmarking.org/result/1...hgv=2600k-3400

In highly threaded tests (where 8150 easily beats 4100), the 8150 and 2600k seem roughly equal.

[fmmjqtft]

Two tests merged together: http://openbenchmarking.org/result/1...hgv=2600k-3400

The fx-8150 and i7-2600k are roughly equal in well threaded tests (whenever fx-8150 easily beats fx-4100). When no more than 4 threads are used, i5-2500k is roughly equal to the i7-2600k. The choice is between:

Low-threaded performance - 2500k
High-threaded performance - 8150 (or 8120)
Both - 2600k

[fmmjqtft]

I accidentally reposted (I didn't realize there was moderation).

[raj7095]

Two tests merged together: http://openbenchmarking.org/result/1...hgv=2600k-3400

The fx-8150 and i7-2600k are roughly equal in well threaded tests (whenever fx-8150 easily beats fx-4100). When no more than 4 threads are used, i5-2500k is roughly equal to the i7-2600k. The choice is between:

Low-threaded performance - 2500k
High-threaded performance - 8150 (or 8120)
Both - 2600k

Thanks a lot bro. That benchmark was really helpful. Interesting to see that 2600k beat the 8150 in 7zip compression? at default clock speed? 2600k for sure now.

[Kano]

The problem with bulldozer is that it is only a 8 core for apps that dont use the fpu, but a 4 core for apps that do (well it always shows 8 cores, but internally). 7zip seems to use the fpu much, so it does not scale much, doing simpler stuff like parallel compiling you can see some better results. Also amd did the false decision in advertising 8 full cores instead of 4 cores with ht to the os, so you would need a different sheduler as workasround, for win you will need version 8 to gain a little bit. well it will not be the case that the cpu will fly away then in benchmarks for me the whole design is crap, they should never have build that 8/4 core mix thing just to get a 8 core for the masses.