Slashdot Mirror
Preliminary OS X & PPC 970 Benchmarks
Dixie_Flatline writes "Macbidouille.com is reporting that they have
preliminary benchmarks involving PPC970 hardware. The results are seriously impressive. We're looking at a single processor PPC 970 1.4GHz machine quite strikingly outperforming a dual G4 1.42GHz machine. Don't worry, there's an English translation embedded in the page so you don't have to try to muddle through the French." Update: 05/05 19:58 GMT by T : Thanks to Eric from macbidouille.com, above link updated to a static page; hopefully you'll get better response this way.
It seems the benchmarks they ran all favour SIMD FPU performance. I'd be much more interested in integer (and integer-SIMD) performance, as this is used much more in mainstream video and audio compression work.
How does it compare to the AMD/Intel/Via processor families?
Food not Bombs is a nice platitude but it breaks down when you notice that the Bombees are usually well fed
Remember that MacBidouille has a history of inaccurate rumors... remember their AMD rumor earlier this year. Check out their rating at www.macrumors.com
I find in hard to believe that MacBidouille actually have been able to benchmark a computer not announced by Apple, based on a chip that's not available before the end of the year according to it's manufacturer IBM.
(Of course, IBM may have been willing to enter Steve Jobs' reality distortion field this time, and have been misleading us all this time - but personally I find that unlikely)
The roadmap is also interesting, though still just a rumeur, of course.
Macbduille must either:
a) have some really good contacts nobody knows about
or
b) is trying to cash in on money from adversiting on it's site and such, but is going to burn it's reputation to do so (come end of may we will know more).
I honestly hope thier reports are true. If they are, macbiduille will be given much status among mac rumor sites, if not, they will be ignored for a long time to come.
64 bits is mostly silly for 99% of applications. Sure its nice to have wider data paths, but that doesn't require any code changes. And sure, as with any radically different procressor implementation the code generator optimization rules need to be changed in the compiler...
But the 6 applications that could actually benefit from a process address space greater than 4GB, or were simulating 64 bit integer math are the only ones that need to be recoded. (Lets see, oracle server comes to mind, nothing like caching the whole database for performance. We do that regularly when possible. Is it on os-x yet? I'm not sure anything else comes to mind. I suppose the computational fluid dynamics folks and other simulations might appreciate it. In general it is the people who do a little bit of processing to large amount of data on a repetitive basis that benefit most from larger address spaces.)
Still, don't underestimate the importance of that code generator rework I mentioned before. I would presume that the applications benchmarked are the regular old 'optimized for motorola g4' versions and a recompile with the new code generator will result in 5-25% improvements. (You might wipe that number off before you use it anyplace else. It came out of my ass.)
I think the heavy duty Photoshop MAC vs Intel machines would be more useful if someone ran the same benchmarks with a Xeon with a (at least) 1 meg onboard cache, even an inexpensive PIII 550.
/1M for a video enthusiast 2 years ago; it was able to do video conversion ( create VCD from videos ) in about real time (ie 60 min to transform a 60 min video). That was an order of mag faster than non-xeon machines.
Usually the bottleneck is in fast memory access..
I remember re-building a Dell 610 Xeon 550
-------------------------------------------------
Apple must have had at least some "same machines that will be sold" at that time; they just did a really good job at keeping them under wraps.
These days the beans are typically spilled at least a few days prior to the announcement of new Apple products. It would be very difficult to have such a high-profile (and high-tech) company entirely leak-proof.
Take this article to mean what you want it to, but I think it is quite possible that it is true.
One thing that the new iTunes Music Store does for apple is smooth out the income curve.
Apple sell computers seasonally for back to school and Xmas. They also sell when they announce the availability of a new model (or is it when they ship it? -- not the same thing unfortunately).
Whether these benchmarks are true or not they are going to depress sales of G4's even further (tho the author rationalizes this by saying pretty much they can't go any lower).
Personally, I was thinking about specifying a refurbed dual Xserve for a customer, which are a really great deal right now if you can find them, but this makes me think that I'll be happy if I wait.
Tho this still hurts Apple it's not as bad as it could be because of the iTunes music store they can get income all year long in a fashion that follows neither back-to-school or holiday seasonality nor is it tied to product announcements.
> IBM may have been willing to enter Steve Jobs' reality
> distortion field this time, and have been misleading us
> all this time - but personally I find that unlikely
I've a couple of uncles who recently retired from IBM. And today's IBM isn't the "Big Blue" of the '80s. Things have changed.
For starters, the engineers, at least, don't wear suits anymore!
But that's not the important bit. The important bit is that ever since bill gates fucked them over, back in the early '90s, in the OS/2 incident, IBM has had an institutional hatred of microsoft the likes of which mere mortals can barely comprehend. They're nowhere near as rabidly vocal about it as the likes of Ellison, McNealy, or a big segment of the Linux community, of course. But, then, IBM has always been rathar understated. They don't bluster. But they *DO* remember!
Catch an IBM'er and have a frank discussion sometime. And you'll find that the prevailing attitude towards microsoft there is: "One day, maybe not soon, but one day... we WILL bend gates and his minions over a barrel and assrape them HARD. And as they say: 'Revenge is a dish best served cold'".
It wouldn't be suprising at all if the RDF had nothing to do do with it; and IBM sped up production, and got prototypes to Apple early, JUST to spite gates.
cya,
john
Imagine all the people...
This strikes me as being odd, considering that an IBM chip shouldn't have an "Altivec" unit (Altivec is a Motorola brand name.) I know the 970 is supposed to have a vector processor, maybe the author's just screwed up. I'd certainly like to believe this article.
This site is wrong a majority of the time and their specs/benchmarks do not ring true.
Go to http://www.macrumors.com/ for a detailed analysis in the forums of why these are fake benchmarks.
Beyond that, the release dates they give are insane, apple is still producing G4 desktops.
Call me when the G5 desktops stop rolling off the line and apple starts depleting inventory.
What do you mean, not needed?
:D
How about Photoshop, which could *easily* swallow 4gb of RAM?
Or VirtualPC running Windows XP + some program?
Or Classic running OS 9 + some program?
Or a combination of all three of the above at once?
Sure, only *some* applications can use the 64bit data paths, but every program can take advantage of the faster bus
GPL Deconstructed
Um, maybe they didn't have one? At any rate, the main point of the benchmarks (making the large assumption that they're legit) was to compare the PPC 970 to the G4, not x86. Although since the low-end single 970 beat the 3.0 P4, I'd imagine a high-end dual 970 would beat a dual P4/Xeon. We'll just have to wait a few more months to see.
How to solve most of our problems: 1.Lots of nuclear plants. 2.Cure aging.
I've worked corporate meetings for IBM, and they would quite happily do quite a lot to frustrate Microsoft. Keeping the 970 secret would be right in line with their corporate attitude. You have to remember that Microsoft's screwing of IBM didn't stop with the selling of DOS to Compaq - it's been a habit for the last 20 years.
As posted by some on macrumors, the benchmarks claim a performance increase in Bryce 3D with dual processors. Bryce 3D does not take advantage of dual CPU's. Don't trust these numbers. I think this website is just making some cash off of the banner ads on the site.
It says exactly what it should say about SIMD speed: "how fast does regular C/Fortran code run when I compile it with a regular compiler". If the compiler can figure out how to transform, say, Fortran array code into SIMD, it will help the SPECmarks. If not, then whatever SIMD features the chip has are useless for most applications, in particular scientific applications. The exception are a few specially crafted apps like Photoshop plugins and MP3 encoders.
(quoted from "you" here)
Well back you working on you're article, love me
Who is this "Poster" guy and why does he own all of my comments?!?
Yes, this is probably true. The PPC 970 is awesome. It's RISC, it's 64-bit and 32-bit backward compatible so all of your old apps will work, but there's one little problem... cost.
Apple isn't known for being inexpensive, and these chips will not be cheap! Intel has cost down pat. Sure, Intel may not be quite as fast as this chip at first, but they'll still be cheaper, much cheaper and w/i a year of the 970's release, Intel will equal or exceed its performance for 1/3 to 1/2 the cost.
How many of you can afford a 5,000 to 6,000 dollar apple/ibm ppc 970???
Why not buy a 3,000 dollar Intel Xeon and run Linux on it?
1. Bryce does not support multiple processors. So the MP results should NOT be significantly different.
l
2. The Pentium 4 and 1.42Ghz DP G4 numbers were lifted directly off of Barefeats website!!! The odds of them using theEXACT combination of hardware and software setup to receive exactly to the second numbers is *HIGHLY* suspicious.
See this page to see where they got some of the numbers...
http://www.barefeats.com/pentium4.htm
3. In general they have been hit and miss on rumors.
I wouldn't believe these numbers at all. Although, I would love for them to be true.
Honestly, I am surprised that you don't know this, Jon!
/ ar chitecture/speccpu.html
Even if SPECint2000 were allowed to run on multiple cores, my understanding is that it is not multithreaded (and you're not allowed to run the different sub-benchmarks in parallel!), so clearly it wouldn't matter.
And in fact we can prove that it doesn't matter, because there IS such thing as a single core Power4! I don't know if IBM is still shipping them, but they certainly used to. Those single core Power4 chips performed just as well per clock cycle in SPECint2000 as the dual core Power4 chips do, so clearly the Power4 is not "cheating" and using the second core. For example, see:
http://www.hp.com/products1/itanium/performance
Note that the 1 Ghz single core Power4 scores 613 in SPECint2000, vs. 804 for a (presumably dual core) 1.3 Ghz Power4. This clearly disproves the hypothesis that the Power4 is getting its good integer performance from having two cores. And I certainly hope that you are not going to claim that *HP* faked results to make the Power4 look better than it is!!!
I respect your analysis in general, but at the end of the day, for whatever reasons, the Power4 core clearly performs MUCH better in integer code than you seem to predict that it would. Remember, these are real numbers, not simulations.
Ok, I went back and looked at the docs and I think you can in fact disable one of the cores, so this is probably a single-core benchmark, WITH A 32 MB L3 CACHE AND 8GB RAM!!!!
(The 970 has no L3.)
That is all.
Senior CPU Editor | Ars Technica | http://arstechnica.com/