Ars reviews the Intel iMac, and concludes: “For $1299, the iMac Core Duo is a good buy. You get a 160GB SATA hard drive, a dual-layer DVD burner, and a 1440×900 17″ LCD monitor along with a dual-core CPU. I’m always reluctant to play the cross-platform price comparison game, but I will note that this machine stacks up quite favorably in terms of price and features with Pentium D desktops. All in all, the iMac Core Duo is an impressive machine. If the rest of Apple’s product line moves to Intel this smoothly, the transition should prove to be a big success.”
Judging by XBench (www.xbench.com), a dual-core 2.0 Ghz G5 is 30% faster processor-wise than the Intel iMac. I’m in the market for a new PowerMac, but thought it was silly that the iMac was just as fast as a G5 – so I was going to wait for the Intel PowerMac’s. Guess I dont have to!
Still, this is an amazing machine and is well worth the money, especially the education price. I can’t wait for Windows to run on them. Gametime!
Um, why does anybody trust XBench? The XBench CPU test shows the Intel Developer Kit machines (3.8 GHz P4) being slower than the top iMac (1.5 GHz G4). It shows the user-interface performance of the Intel Mac to be much lower than the UI performance of the G5 iMac, even though the reviewer remarked that the former felt snappier. The review itself is deeply flawed — the Rosetta benchmarks in particular, which test the poor iMac with 512MB and compare it with a 4.5 GB PowerMac.
Um, why does anybody trust XBench?
Actualy, new IMac did equaly bad on all tests.
The review itself is deeply flawed
Not really. Results are just what it was to be expected. Centrino Duo actualy belongs right where it landed. Being really good for base desktop functions only.
Actualy, new IMac did equaly bad on all tests.
The Rosetta benchmarks, as I said, were flawed. Rosetta is a JIT. OS X is a memory hog. Photoshop is a memory hog. Running all three together, and then comparing a 512MB system to a 4.5GB system is just plain stupid. The XBench results are suspect, since as I said they peg the 3.8 GHz dev kit machines as being as fast as a 1.5 GHz Mac Mini. There is clearly something wrong with XBench on Intel. So what does that leave? Quicktime and iTunes. The Intel Mac does quite well in both of these benchmarks.
Not really. Results are just what it was to be expected. Centrino Duo actualy belongs right where it landed. Being really good for base desktop functions only.
Read some of the macnn forum threads about people who have actually tried the new Intel Mac, and then compare it to this review. The results of the observations are very different.
From looking at pictures of the chips, we were able to do some deduction from the chip numbers. The northbridge chip looks to be an 82945GM, which identifies it as the Mobile Intel 945GM Express chipset. According to the S-Spec number (SL8Z2), it’s a standard Intel chipset and therefore not an Apple-specific variant.
So it’s guaranteed to be compatible with Vista when it comes out. Which is nice.
A couple things that I noticed when comparing XBench between Intel iMac’s and G5…
1. The CPU Test has x86 way ahead on GCD Loop, a little lower on FP, and way way lower on the vecLib FFT test. I think it’s the lack of Altivec that is killing the CPU Score.
2. I think the UI test relies heavily on Altivec, which would account for the low XBench score for UI and would be consistent with the reported UI responsiveness of the Intel iMacs.
Also, the major issue I have w/ Ars testing of the x86 iMac is the memory being only 512MB. This should have been at least 1GB, as OS X is such a memory hog, not to mention the applications being used for the benchmark.
-Kelson
Until Apple ships the base box with 1gb, 512 is a more than reasonable test kit. And the results are scewed, as Xbench does appear to have some Intel issues with it’s methodologies, but the memory remains a reasonable baseline.
Are you seriously suggesting that 512MB is a reasonable baseline for running Photoshop and Cinebench? Apple ships 512MB because they cannot get away with shipping a machine with no memory. Even the Quad ships with 512MB, and I’d be surprised if there was a Quad in existence that actually run with so little memory.
Even if you do consider 512MB to be a reasonable baseline, then I should point out that the dual G5 in the comparison also shipped with a base 512MB of RAM. It was tested, however, with 4.5GB of RAM. Clearly, the testing methodology is flawed.
Edited 2006-01-17 19:12
The idea of 512MB base is this – you can do things, of a reasonable size, to that of a reasonable speed – if you require more, its up to you to specify, only you know what you’re going to use your computer for, so its for you to decide whether or not 512MB is enoug and whether you wish to use the memory upgrades provided by Apple or obtain your own.
You are right, 512MB isn’t enough – personally for me, 1GIG, minimum for computers these, regardless of whether it runs Windows, Linux, MacOS X or some other niche OS.
Is a 512MB performance comparison useful? Should comparisons between different computers try to be as even as possible?
I think as long as the standard configuration is 512MB it’s obvious that reporting information about such a comparison is quite useful.
You should also standardize configurations as much as possible for comparisons. At least one of those results should have included more memory, simply to see how increasing the RAM would compare.
Then again, one would ask what the purpose of the benchmark is for – is it to benchmark the WHOLE computer or simply to demonstrate how fast the CPU is?
For me, if we’re looking at the ‘over all performance’ of the machine, out of the box – if 512MB isn’t enough to perform at a reasoanble standard, then the problem lays with Apple and their tightness in the amount of memory that comes standard with their machines.
It isn’t so much a matter of whether 512MB is “enough.” Or at least it isn’t to me. Whether 512MB is “enough” is dependent upon the user’s requirements. One can more readily decide if it’s enough or not if given sufficient information about the iMac CD with different memory configurations for different tasks. Then the variable for performance comparison on the same machine becomes memory.
If you just want to compare the different machines and increase the utility of the results to “slightly less useless,” you should minimize the amount of variance between machine configurations. If they were interested in showing how the PowerMac and the iMac CD compared then they should have both had an equal amount of RAM for the testing. It’s not that knowing how the PowerMac G5 performs with that memory isn’t interesting, it’s just not as immediately informative for comparison as also including results with 512MB of RAM.
I think the idea with the PowerMac comparison was to say, “this is what it like, performance wise, in regards to the other extreme end of the G5 spectrum”.
If they were benchmarking the whole computer, then they’d test the dual G5 with 512MB as well. You can’t upgrade one computer in the test and not the other!
huh? The reports from Ars say that the intel macs have instant UI response.
And the XBench testing has a lousy score for the UI test on the Intel iMac.
– Kelson
If you threw a Radeon X1600 sitting on a PCIe with 128MB, I’d expect the damn thing to atleast provide some ‘teh snappy”.
Yeah, the problem with the XBench CPU test is a single thing: they take 3 different CPU benchmarks and average them (Integer, Floating Point, and Vector Unit).
The problem with this, is it gives an end result which does not reflect reality in many situations, they should keep these 3 benchmarks separate. Clearly, the new Intel-using Macs are much better in integer, which is the most important in terms of apps today.
It is important to notice that lack of Altivec does kill the vecLib benchmark (and it seems SSE3 is not comparable in performance, since Apple themselves say that vecLib uses the best unit available on the CPU for the vector stuff). This is obviously seen in the Quicktime encoding benchmarks, which are both native, yet the new iMac doesn’t perform anywhere near the “2X speed” that Apple advertises, simply because Quicktime (and many other Apple apps) has been heavily optimized for Altivec.
This is obviously seen in the Quicktime encoding benchmarks, which are both native, yet the new iMac doesn’t perform anywhere near the “2X speed” that Apple advertises, simply because Quicktime (and many other Apple apps) has been heavily optimized for Altivec.
Could be that QuickTime isn’t multithreaded, in which case it’s straight core-to-core combat.
Closed-source application benchmarks are great for people who want to decide what to run those very applications on.
But they’re not much good for comparing different computer architectures, because you just can’t tell how well the code was adapted to each arch.
And a closed-source synthetic benchmark like XBench is just laughable for comparisons across architectures.
It’s important to note that XBench is not multi-cpu aware for any of it’s tests, excluding the thread test.
Excellent review getting excited about future intel Mac releases! Its strange thought that apple chose such a slower 666Mhz FSB speed you would think they would have at least went 800Mhz. Who knows perhaps they are reserving that for the power Mac line.
Apple didn’t choose that bus speed, Intel did. All the Core Duo chips us a 667MHz FSB. Although, it should be noted that unlike the G5, the Yonah FSB is a single 64-bit bidirectional link instead of two unidirectional 32-bit links. That means the G5 can only approach its maximum speed on interleved reads and writes, while Yonah can approach its maximum speed on burst reads and burst writes.
and they didn’t even open the thing up, or drive over it with a car.
From the article: “I took a shot at getting inside by removing the Torx screws at the bottom of the case along with memory door covering. No dice.”
You need to slip a thin piece of firm, slightly bent plastic into the upper corners of the bezel to disengage the latches. After that you can pull it off and disconnect the wires. He did say that he liked his warranty, though.
A few months back, when there was some discussion of the DRM built into Intel chips and I talked about my concerns about how this would be implimented, etc. (I have my Mac set up to be a TiVo.)
At which point some zealot mounted up and went on and on and on about Open Firmware and how Open Firmware meant that DRM was worthless and Apple computers used Open Firmware because Apple was committed to Open Firmware.
As if Apple couldn’t switch?
Oh no, once you’ve got Open Firmware, why would anybody switch?
From the review: doesn’t support Extensible Firmware Interface, which is what the new iMac uses in lieu of Open Firmware.
;/
Is The Register correct –
http://www.theregister.co.uk/2006/01/16/intel_core_duo_vt/
No VT for the T2400/2500/945GM and I assume iMac ?
I got a 24 inch LCD and a Dell Pentiun D 820 for 1299 from Dell. Much better than the mac.
I got a 24 inch LCD and a Dell Pentiun D 820 for 1299 from Dell. Much better than the mac.
How? The 2405FPW is a $680 build to order option on Dell.com. That means the base computer price was $619.0. No Dell with a base price of $619.0 has a dual core processor.
The Dell comparison is interesting, though. The closest Dell I can find to the iMac is the the XPS200 (slimline model). I’ve specc’ed it out here:
http://configure.us.dell.com/dellstore/config.aspx?c=us&cs=19&l=en&…
That system costs $1483. The $1700 iMac Core Duo 20″ has a faster CPU and a faster graphics card. According to the SPECrates available at aceshardware.com, the Pentium-D 820 at 2.8 GHz is roughly equivalent to a Core Duo 1.67 in integer, and a Core Duo 2.0 GHz in floating-point. The X1600XT in the Intel Mac has 4x the pixel fill rate and 2x the memory bandwidth of the X600SE in the Dell, and supports Pixel Shaders 3.0 while the X600SE only supports Pixel Shaders 2.0. In this particular comparison, the “Apple Tax” is almost nothing.
EDIT: Sorry, the link doesn’t save changes to the cart. To get the $1483 figure, add the $29 remote, the $80 graphics upgrade, the $50 hard drive upgrade, the $40 DVD burner upgrade, the $260 2005FPW upgrade, and the $25 speakers. These make the system match the iMac specs as closely as possible.
Edited 2006-01-18 04:33
Grandparent poster probably got in on a deal like this one:
http://www.techbargains.com/news_displayItem.cfm/56496
Although that one’s only $1199. Personally if I had the cash I’d get it, sell the computer and keep the monitor…
That’s not even close to an equivalent computer, though. X300 SE? No dual-DVI? No DVD burner? No speakers? And its still got a slower CPU! It’s not even a slimline form-factor (which costs more even at Dell). It’s a completely irrelevent comparison…
I agree, it’s a typical Dell deal. They lure you with a great LCD and a decent (if you like Intel) cpu, but everything else is average to meh and with overpriced upgrade options…
There are a lot of serious mistakes in the benchmarking process, many mentioned in Ars’ forum discussion of the review. IMHO, these stand out:
* First, the Intel iMac had 512MB, yet the iMac G5 used had 1GB, and the Power Mac had freaking 4.5GB. That alone makes many of the results meaningless.
* The CD encoding test is flawed. My old G4 can encode faster that what the iMac G5 did. There was probably something wrong with the DVD drive, which is entirely possible considering that:
* The Intel iMac was a brand new machine, with a clean System, whereas the iMac G5 and PM G5 weren’t, as the author acknowledges. This makes hard disk & “general system responsiveness” comparisons very uneven (in this case, unfair to the G5 models).
* Xbench. The app is not a serious benchmarking tool, but it is even worse in this case: the graphic test results for the Intel iMac are ALL flawed, since apparently Xbench for X86 was linked against OS X 10.4, which forced it to beam sync. For full details, including updated score results, check this post at Ars:
http://episteme.arstechnica.com/groupee/forums/a/tpc/f/174096756/m/…
No wonder he found the graphic results “odd and inexplicable”.
Let’s hope for a revised review that addresses these and other issues raised in the Ars forum.
You do have some good points but I have ot disagree with the first point you make about memory. The author himself acknowledges you need a boatload of memory to run OS X in the article. I think you can look at it as the author saying a lot of people are gonna go with 512 mb anyway when they buy the system instead of 1 gig so here it is a review that will be based on a spec that most people will most likely choose to get.
The people who get 512MB won’t be running Photoshop. So what’s the point of making that comparison?
If he wants to make a valid comparison against the Intel iMac from the G5 based systems, he should have increased the memory in the Intel iMac or decreased the memory in the G5 based systems.
If you want to argue that most people will buy it w/ 512MB, that’s fine, but the argument still stands that when doing comparison testing, the RAM quantities should have been equal.
This doesn’t matter really for XBench, but for all of the application style testing, it is critical.
– Kelson
“Judging by XBench (www.xbench.com), a dual-core 2.0 Ghz G5 is 30% faster processor-wise than the Intel iMac. I’m in the market for a new PowerMac, but thought it was silly that the iMac was just as fast as a G5 – so I was going to wait for the Intel PowerMac’s. Guess I dont have to!”
“Um, why does anybody trust XBench? The XBench CPU test shows the Intel Developer Kit machines (3.8 GHz P4) being slower than the top iMac (1.5 GHz G4). It shows the user-interface performance of the Intel Mac to be much lower than the UI performance of the G5 iMac, even though the reviewer remarked that the former felt snappier. The review itself is deeply flawed — the Rosetta benchmarks in particular, which test the poor iMac with 512MB and compare it with a 4.5 GB PowerMac.”
You guys are amazed one of the best and most modern 64 bit cpus are beating a turbo loaded but power efficient P3 which are just a relative to the i386? Of course a dual PPC970 will beat the crap out of a dual-core x86. Have you noticed how Apple have went from Photoshop performance to SpecFP/SpecInt? That is of course because they chooses the benchmark which gives the best numbers for their statistics, they would probably not be as impressive compared to an Altivec optimized program. Xbench DOES matter because that is how many people will be running their programs atm, emulated.
I’m surprised by the result because it doesn’t jive with my experience. The “best and most modern” 64-bit CPU is actually a fairly mediocre chip when you get down to the details. The 970 has a lot of potential, but the implementation is flawed in many ways (not enough integer units, high-latency memory controller, overly long pipeline, etc, etc). I use one on a daily basis, and for the things I do (from browsing the web to compiling code), it is a slower chip.
As for SPEC versus Photoshop — Photoshop is and always has been a terrible benchmark. It makes no sense to compare CPUs by running Photoshop, because Photoshop runs different code on each platform. Also, it only tests one type of code. SPEC is a suite of benchmarks that tests a wide range of real-world code. It’s results are also very consistent with what I’ve observed regarding the G5 versus my Athlon64.
Further, the P3 is not a ‘relative’ to the i386 any more than the G5 is a relative of the PPC601. They use the same ISA, but that’s about it.
Regarding the benchmarks: take a look at the actual breakdown of the CPU benchmark. The Core Duo is significantly faster in integer (expected), a little slower in floating-point (expected), but its overall score is much lower because the vector score is *really* low. Look at the other xbench results available on their website. They show the 1.25 GHz G4 in the Mac Mini being faster in the veclib test than a 3.8GHz P4! No way! If an experiment tells you that the sky is purple, there is clearly something wrong with it!
Also, while you’re right that many people will be running apps emulated in Rosetta, its still stupid to compare one machine with 512MB of RAM and one with 4.5GB of RAM. Nobody is going to be running Photoshop and Cinebench on 512MB of RAM!
Edited 2006-01-19 01:11
The 970 is generally a bit better than the Opteron at floating point code, but gets totally slaughtered by Intel (and AMD) at integer code. So while the 970 is great for workstations and professional uses, most people on the desktop will be better off with the Intel chips.
As for Apple’s choice of benchmarks – they’ve always chosen the benchmarks that make their current computers look best. Why do you think they were using Photoshop in the first place? So I don’t see either Photoshop of SPEC tests as being very impartial.
XBench: Clearly the Altivec code used in the CPU test was not ported to SSE very well. It should be slower, but not that much slower. And because the Intel version was compiled against a later version of OSX, the results are quite different.
Testing with different amounts of RAM: I actually don’t mind that they tested against the Powermac with 4.5 GB of RAM. There’s nothing wrong with testing a typical iMac against a top of the line Mac. You’re just testing the entire platform and not specifically the processors. What was bad was the different amounts of memory in the 2 iMacs. Both should have been equipped with 1GB, or at least the same amount. Those 2 machines were clearly meant to be compared against each other to evaluate the new processors, and it is hard to do that with different amounts of memory, particularly when the testing involves programs that are likely memory constrained at 512MB.
All in all, I thought this was a very disappointing review from Ars, although my expectations were very high. Hopefully they’ll make a new one that will be up to their usual standards.
Seems Xbench where a universal binary and so on.