Slashdot Mirror
G5 vs Opteron, Finally
metfoo writes "It's been months since the G5 and Opterons have been available for purchase. When the G5 systems were first released, many Mac bashers and AMD nuts discredited the G5's performance. They always ended their comments with 'Wait until its compared to an Opteron, then we'll talk.' Well, it's finally time to talk. Barefeats has posted an article comparing the two systems. The G5 line was compared to a Dual 2GHz Opteron and the results are impressive. In gaming, the Opteron system proved to be superior, which is partly due to the superior 9800XT over the base Radeon 9800. The G5 spanks the Opteron in many of the non-gaming tests, except for the Photoshop tests."
same could be said for the G5, as OS X is not fully 64-bit yet, and neither are many of the programs (with the exception of a few high-level apps, like Photoshop, etc)
Viva La Revolucion! Buy a Mac!
Agreed. I especially wonder how the Photoshop Tests, AfterEffects tests, and Bryce tests would have done. None of these have a Linux version available. I don't know if any but Quake 3 do.
I suggested running 64-bit linux on both, just to be supremely fair, but I think that it wouldn't be much different from 64-bit amd64 and 32-bit darwin.
:)
It should be noted that 32-bit darwin runs 64 bit apps just fine (and fast!
No, I was stating that it WAS A good comparison. Two 64-bit processors (that are 32-bit compatible, natively) running 32-bit code (along with "optimized" code as I'm sure Adobe's optimized for both CPUs) aka a "test under real world conditions" because both were run with what is available today.
The desktop iteration of the Opteron is the Athlon FX-51. Maximum PC has run multiple benchmarks on the two systems to compare them, and the athloh whoops up on the G5. As a matter of fact, the G5 lost in all but one or two tests to the two frontrunners, the Athlon FX and the P4 Extreme Edition. This was obviously in 32-bit mode. I don't hate macs, but in this race (the desktop race), it certainly comes out under the other two major chip manufacturers.
More here. I suspect that Apple's developer pages have more info on the trickery.
Try reading the article. He addresses this, but I'll save you the time. OSX is not a 64-bit OS either, so neither really has any advantage there. Hopefully, in the next few months, each system will have a fully 64-bit OS, so we can really see some full speed comparisons.
had feature called "The Race Is On" by Jonathan Seef. The comparison was between G5 and PC's with opteron. The PCs seemed to fare better in most of the tests (photoshop, word, quake, premiere, mp3-encoding, mpeg-2 encoding). Mac seemed to be better only with the DVD creation. By the way, I use Powerbook G4. Anyone's got a link for the article ?
Science as a way of life.
Except for nVidia cards, which have always had excellent OpenGL drivers, D3D renders faster than OGL on practically all current consumer level 3D hardware for Windows. Even on the nVidia hardware the speed difference is practically negligble.
That since they are running the Opteron in 32-bit mode, it's not taking advantage of it's full potential. Guess we'll wait until "round 2" like he says, but it still looks bad that he kind of dodges this. If it were me I'd be running the benchmarks on 64-bit linux versus 64-bit linux.(gentoo?)
The G5 is similarly crippled. Mac OSX is a 32 bit OS running in 32 bit mode. All 32 bit. The only "piece" of OSX that is 64 bit is the memory access - to allow a G5 to use up to 8 GB of ram.
While the G5 chip is fully 64 bit, OSX is taking each 64 bit instruction (from 64 bit apps like Photoshop, for example) and breaking it into two 32 bit instructions which it sends one after the other to the CPU. I imagine 32 bit Windows on the Opteron is similar.
The "real" benchmarks will be the ones that compare 64 bit performance on each platform. We'll be waiting a while for those, as OSX will not be fully 64 bit for another year (or more).
Linux would level the playing field for the short term, because that would leave hardware as the only variable thus providing a true comparison of hardware performance.
Last thing to keep in mind is that OSX is built using GCC - which doesn't build the fastest PPC970 binaries. IBM is working to provide Apple with a much optimized PPC970 compiler tailored for the G5 - this is expected to give a large performance gain for this platform.
I'd rather be a conservative nutjob than a liberal with no nuts and no job.
First, lets examine the statement The G5 spanks the Opteron in many of the non-gaming tests, except for the Photoshop tests. I see five tests in this review, and there are two wins for G5, two wins for Opteron, and one tie. So I really don't see either chip spanking the other. On the second page Opteron wins two, G5 wins one.
Second, the configuration notes section was pathetic. It doesn't really give a very good description of the real configuration of the systems. Anyone that views benchmarks regularly knows that the devil is in the details. Still, this is equally a problem for Opteron and G5.
Third, I wonder what kind of comparison is really valid. Anyone familiar with the AMD/Intel world knows that you can't just grab two 2Ghz chips and run them head to head. The architectures are not the same, it wouldn't be a valid comparison. So with two entirely different ISAs, what chips should be run head to head? The only obvious comparison would be each manufacturer's fastest...in this case I believe the 2GHz G5 is Apple's current fastest, but AMD does have a 2.2Ghz part that is available (see pricewatch) and that wasn't tested.
Lastly, let me address the importance of compilation. I can't speak for G5, but you would get a substantial boost in performance on most applications just with a recompile for AMD64 chips. This is because Opterons have 32 GPRs instead of 16, which can make a big difference (especially in multimedia apps like photoshop). Obviously these products aren't commercially available, but people should be aware that a substantial performance boost for AMD64 could come just from optimized releases of software once it reaches a wide enough audience to make it worthwhile for software vendors.
I guess this has turned more into a "notes about AMD64 architecture" post than anything else. It looks to me like this review is interesting but doesn't really settle much. Both Opteron and G5 performed well.
I know this isn't quite on topic, but I wonder how the latest Alpha design would fair. The alpha was the first mass produced 64 bit chip that had any commercial success. It was introduced in the early nineties. IN fact Linus had one. Basically the curret EV78 is a 6 or 7 year old design, but in most serious tests of processor power it has done quite good. It's amazing that such an "old" design still works so well. The last SPEC numbers I can find are here. Considering the platorm has been ignored and basically orphaned, it's suprising that this chip still powers many of the worlds top rated super computers.
How does all this relate to the G5 and Opteron? Well AMD gets it's bus design from the Alpha lineage. The G5 is built by IBM, who I believe is building the alpha cores as well (I could be wrong, I can't keep up). The irony? Every current intel pentium chip is quality control checked by machines with alpha processors. Funny world huh?
AngryPeopleRule
"Science is about ego as much as it is about discovery and truth " - I said it, so sue me.
I can't figure out the pricing in that review anyway, I just went to the xicomputer.com website and configured the same system and came up with $3236, not $4107. I'm not sure if I missed an option or something, but maybe they just bought their system a long time ago before prices dropped (maybe before AMD released the Opteron 248, the 246 which was tested isn't even their fastest chip anymore.
Here are some benchmarks that show comparitive performance under Maya and Mental Ray. The G5 does not due so well in this, atleast compared to Intels and AMDs offerings.
t al.htm
_ maya.htm
An item of interest in both benchmarks are
the stats from an SGI Tezro workstation with 4 procs.
Mental Ray for Maya
http://www.zoorender.com/html/benchmark_men
Standard Maya Renderer
http://www.zoorender.com/html/benchmark
Also there are alot better reviews already published that cover these cpus respective performance in more detail.
Cheers
The key difference is that the opteron is /faster/ in 64-bit mode. There are more registers. The same is not true of a G5. OSX doesn't "break up" instructions - instructions are instructions. OSes don't interpret each one. 32-bit windows on opteron just uses 32-bit instructions, just like 32-bit darwin on a G5. Photoshop is not a 64-bit app in the case of OS X IIRC - it uses a 64-bit math library as its G5 optimization. This is fine and works. As I mentioned in another post, darwin can't run "64-bit apps" yet, because there are no 64-bit interfaces to system calls (think about it - if the kernel expects a 32-bit FILE * and you send it a 64-bit one, you're going to have trouble doing I/O). I think I missed something with what you said there. The main 64-bit part of darwin is the math library since they can throw some 64-bit ASM in there plus code to convert back-and-forth to the 32-bit bindings.
I would liek to say teh test is unfair and biased so i can sound like most of you. Tne tes should be more standard.
1) Why would you argue this and waste your valueable time.
2) Both systems have advanatges and disadvanatages, liek everything else in yoru life
3) Why don't you spend the hours you essentialy waste on this riddiculous debate by doing something constructive liek volunteering for community service or collecting stamps
4)Have A Good New Year be you RISC, CISC, MIPS or x86, ppc, x86-64 Itanuium, SPARC MIPS,
just have fun being constructuve; you people are smart
Are the moderators on crack? That's not informative, it's wrong!
The only "piece" of OSX that is 64 bit is the memory access - to allow a G5 to use up to 8 GB of ram.
No, the math libraries and kernel support 64 bit goodness too. While the G5 chip is fully 64 bit, OSX is taking each 64 bit instruction (from 64 bit apps like Photoshop, for example) and breaking it into two 32 bit instructions which it sends one after the other to the CPU.
No, it runs as a single 64 bit instruction through the chip!IBM is working to provide Apple with a much optimized PPC970 compiler tailored for the G5 - this is expected to give a large performance gain for this platform.
This I agree with, in fact it should be pointed out that the compiler is available now as a beta on IBM's web site.
according to them:
Athlon 64 vs. Apple G5 Systems: Not Even Close
Now i can't say whether these tests are any less or more objective, but they do draw a completely different conclusion.
-Jon
this is my sig.
there is no emulation in either chip, it is all native.
I am the Alpha and the Omega-3
..unless you know enough about PCs to assemble from components.
Shop here if you know what you're doing. With components you'll spend about $1500 building an absolutely top-end Pentium 4 system that would be $3000 plus from Dell.
Apple have no business trying to sell to people who know what they're doing since you can't build an Apple G5 from scratch and avoid their exhorbitant markups.
On the latest Top500 list Virginia Tech's Mac cluster is number 3 with 2200 2GHz G5 processors, and Los Alamos National Laboratorys machine, with 2816 2 GHz Opteron processors is number 5.. I didn't look at the topology, or connection medium, but I am certain that the Mac cluster was cheaper, and is faster running the SAME benchmarks...
why don't you use a mac running their new OS, fucktard. it has built in support for scroll wheels, 3 buttons, and it's all setup to work by default with most mice.
plug just about any multi-button scroll mouse and it works fine.
right click on the desktop (or jsut about anything) and you get a contextual menu. and it's not just emulating a ctrl click. it's real multi mouse button support.
don't spew your ASM coding garbage trying to sound smart. the support is there, it's been there since it was called NeXT.
They are having trouble. When Solaris was moved to 64-bit, they compiled all the system software in 64-bit mode, and added 32-bit compatibility libraries and 32-bit compatibility system calls to the kernel. Users got a full 64-bit OS on a 64-bit machine, and all their 32-bit apps worked just fine. No convincing required, just a regular upgrade, like from OS X 10.2 to 10.3. The fact that 10.3 isn't 64-bit implies that there are some problems, namely that not all the code is 64-bit clean.
A deep unwavering belief is a sure sign you're missing something...
There is no breaking down of instructions as you mention - there is an extended instruction set which is either available or not on the PPC (so PPC processors which support 64 bit instructions support the full instruction set, whilst the 32 bit processors throw an illegal instruction).
The measure of the processor 'bitness' is when addressing memory, whether you are limited to a 32 or 64 bit address space. In effect the size of a pointer will tell you. Now the impact of moving to a 64 bit address space is usually detrimental to the speed of the application (I would expect that the G5 will be slower in 64 bit mode) due to the increased memory bandwidth required (if pointers are wider, you need more stack/memory to store them, so you have, say, a 10% increase in memory requirement for an application). If I remember correctly, the HP PA RISC systems were slower running 64 bit code than 32 bit. I'm thinking back a couple of years here, so it may be different for the latest generation of their processors.
It isn't all bad for the Opteron though - the instruction set has been altered to provide more processor registers. This may mean the compiler can do a better job of optimising code, reducing the number of load/store instructions, so increasing the IPC.
So, just don't assume that if the Opteron gets a performance benefit from 64 bit code, so will the G5. As other people have mentioned, you need to test both with a 64 bit OS to really know the results - extrapolating won't get you the answers.
Your comment about the IBM compiler is spot on. With any luck the IBM changes will make it into the gcc code for other non-apple PPC platforms.
I don't know if the author of the article was just trying to make the mac look cheaper, or if they just weren't looking very hard (I suspect the latter), but they could have EASILY shaved $82 off the cost of the Opteron system and got 1394b support for free too boot!
For some reason they configured the Opteron with an add-in Serial ATA RAID controller, supposedly in order to better match the configuration of the Mac (which doesn't support RAID, but I digress). This added $117 to the price. However they completely ignored the motherboard upgrade option that added SATA support (no RAID) and 1394b support together for only $35.
They could also have shaved another $37 off the price by using a software modem instead of a hardware modem (though the HW modem might be a good idea for Linux users that need dial-up) or $72 off the price by not including a modem at all for those of us with broadband connections.
In the end though, the Mac is still a bit cheaper. Macs are not expensive for what you get, the problem is that you don't have much choice but to get top-end. To price out a dual-processor Opteron with similar specs to a dual-processor Mac, you'll be easily over $3000 and possibly up closer to the $3938 of the Xi computer system. However, if you don't need all those features you can easily configure yourself an Athlon64 system for SIGNIFICANTLY less.
I have absolutely no need for a modem (got an old external kicking around in case of emergancies) and have never owned any 1394b devices. Therefore, if I were configuring a PC for myself I would never bother adding either of those two options. I might also configure a cheaper video card and I probably wouldn't bother with a DVD-RW drive, though I would prefer to have two optical drives (one CD-RW and one DVD). These are all easy options on most PC configurations, but often they aren't on Mac configurations. Simply put, you have more choices on PC configurations than on Macs. If you desired setup matches that of a Mac closely, then they often offer good value for your money. If not, then they can be quite expensive for what you want.
We've had benchmarks for months, actually meaningful benchmarks. They show that the G5 is a nice, competitive chip, but it's merely keeping up with AMD performance-wise. And G5 systems are behind Opteron systems in terms of bang-for-the-buck and features.
If you check the published SPEC benchmarks for the Opteron 148 against Apple's claimed SPEC results for the G5, you'll see that a dual G5 is not faster than the Opteron. It is pretty telling, incidentally, that Apple still has not actually submitted official SPEC results for the G5's--they really don't seem comfortable with the comparison on a real benchmark.
Of course, a dual Opteron will have other advantages for many users: you can get it in 1U rack mounts, it runs a lot more application software, and it's cheaper.
Running five application programs does not constitute a meaningful benchmark of the CPU. We don't know how those applications are written, what CPUs they are compiled for, what compilers they used, etc. Most likely, none of those applications have been tuned for Opteron, wherease they have received extensive tuning for PPC and AltiVec over the years. The differences may be something as trivial as cache conflicts. All those "benchmarks" tell you is that if you must run the current version of Bryce and AfterEffects, you may get more bang (but not necessarily more bang-for-the-buck) out of a G5 for the time being.
The Opteron and the G5 (IBM PowerPC 970) are two disparately different chips meant to serve two different purposes. The Opteron is AMD's server chip designed to handle for the most part, 64 bit high performance database applications and applications which require large memory models in which the 64 bit memory addressing is needed-- NOT 64 bit desktop applications or games. That's reserved for the Athlon 64 which is clocked significantly higher and has a much smaller L2 cache than the Opteron. On Gaming and desktop content creation benchmarks the Athlon64 is a much better match for the 970.
Actually, the cache size of the Athlon64 3200+ and the Opterons are exactly the same, 1MB. The frequency difference is a mere 200MHz (A64 3200+ vs Opteron x46, the x48 chip is clocked exactly the same), but that might seem significant as it is 10%. The most significant difference is that the Opteron has a dual channel memory interface, which the A64 hasn't. Enter the Athlon FX, which is exactly the same as the Opterons but without the ability to run MP. The cores are identical, so there's no real optimisation benefiting the Opterons over the A64s.
Sun and SGI (and now HP-UX with Itanium) systems are still messy with dual ABIs for backward compatibility. On Solaris, gcc still defaults to 32-bit mode.
The only proprietary system that did a clean 64-bit transition was OSF/1, because it simultaneously switched architectures and the OS.
Actually, the G5 is Apples Pro Line. Their consumer line consists of the iBook, iMac, and eMac. Their Pro line consists of the PoweBook and Power Mac. Maybe the naming gives it away too... POWER Mac, POWER Book.
It is my understanding that a number of these changes have been rejected by the gcc team on the basis that they would require some major changes to the internal structure of gcc which could have a detrimental effect on the code generation on other platforms. This is a valid point, since the primary purpose of gcc is to be a cross platform compiler, while XLC is a pure PowerPC/Power4 (where 4 is any integer) compiler. I wouldn't be surprised if Apple or IBM (or a collaboration of the two, and possibly Motorola) were to fork the gcc code base and and make a more heavily optimised version, periodically syncing the non-platform-specific parts with the GNU project version.
I am TheRaven on Soylent News