[Veltliner]

I wonder how hot the 4-core will get.

I read the 4-core processor was 100W. Is that correct?

And wouldn't the 27" iMac improve vastly from the newer, thinner multi-core processors that are about to appear within the next 6 months?

I had the impression that every new model Apple introduces, after 6 to 8 months you got a major improvement in the second generation.

I don't have to buy now, and I won't buy this year. But what I've seen of the 27" iMac - I have this date feeling when I look at it

[Eug]

I'm no expert, but here are my 2¢:

Originally Posted by Veltliner 

I wonder how hot the 4-core will get.

I read the 4-core processor was 100W. Is that correct?

TDP is 95 Watts. However, it has excellent power reduction capabilities for when it is used lightly, which is most office usage.

And wouldn't the 27" iMac improve vastly from the newer, thinner multi-core processors that are about to appear within the next 6 months?

I had the impression that every new model Apple introduces, after 6 to 8 months you got a major improvement in the second generation.

I don't have to buy now, and I won't buy this year. But what I've seen of the 27" iMac - I have this date feeling when I look at it

No. 4-core iMacs are likely going to remain Lynnfield until end of 2010.

Gulftown 6-core is coming, but that's for the Mac Pros.

The interesting part is for dual-core CPUs. Intel is revamping the dual-core lines and is claiming that its top-of-the-line dual-core Clarkdale processors will be faster than Core 2 Quad, at least in SPEC fpu.



Core i7 will still easily maintain the performance crown over Clarkdale Core 2 Quad however. CPU-wise, Clarkdale is essentially a 2-core version of Core i7. I just hope I get my Core i7 before Xmas.

The weird part will be the naming conventions, which will be potentially very confusing. Low end Clarkdale will be named Core i3 which makes sense but high end Clarkdale will be named Core i5. That makes Core i5 will have both dual-core and quad-core CPUs.

[Simon]

Originally Posted by Eug 

I'm no expert, but here are my 2¢:
...
CPU-wise, Clarkdale is essentially a 2-core version of Core i7.

I don't think "CPU-wise" has any meaning, but in terms of process nothing could be further from the truth. Lynnfield (the iMac's Core i7) is 45nm fab. Clarkdale OTOH belongs to an entirely new generation (Westmere generation) based on a 32nm process with integrated switchable graphics core. Performance-wise these CPUs might be similar, but in terms of design and fabrication they are inherently different.

In terms of marketing there is a similarity though. Lynnfield is a 'dumbed down' version of Bloomfield. And that is what Clarkdale is to Westmere.

[Eug]

Originally Posted by Simon 

I don't think "CPU-wise" has any meaning, but in terms of process nothing could be further from the truth. Lynnfield (the iMac's Core i7) is 45nm fab. Clarkdale OTOH belongs to an entirely new generation (Westmere generation) based on a 32nm process with integrated switchable graphics core. Performance-wise these CPUs might be similar, but in terms of design and fabrication they are inherently different.

In terms of marketing there is a similarity though. Lynnfield is a 'dumbed down' version of Bloomfield. And that is what Clarkdale is to Westmere.

You're right, but I was specifically talking about performance. My guess is that 2-core Core i5 Clarkdale is going to be similar to what 2-core Core i7 Lynnfield would have been. When released, clock speeds with Core i5 Clarkdale will be faster than quad-core Core i7, but I don't think there will be any significant inherent clock-for-clock speed boost compared to a hypothetical dual-core Lynnfield.

Back to iMacs... I don't think Apple will update the iMacs until Intel updates the quad-core CPUs. So I foresee an update to the iMacs in the second half of 2010:

27" iMac Core i7 quad-core
27" iMac Core i5 quad-core
27" iMac Core i5 dual-core
21.5" iMac Core i5 dual-core

ie. No spring iMac update, if Intel's public roadmaps are accurate.

[P]

Originally Posted by mduell 

If you want exactly the iMac specs (27" 1440p, i5, etc), it is the cheapest computer out there.

If you go with the 30" (with more screen area and pixels than the 27" iMac) you end up just shy of $2200 with i5, or 10% higher. Throw in Apple's i7 upgrade price and you're down to $100 (5%) apart. The competition will only get cheaper while Apple maintains price over the next half year.

IMO the insistence on IPS is a bit of a red herring; fine to exclude comparisons with TN, but PVA is on the same level IMO.

That's a fair comparison. I agree that PVA is an acceptable substitute (TN is not), but where did you spec that model?

[P]

Originally Posted by Simon 

I don't think "CPU-wise" has any meaning, but in terms of process nothing could be further from the truth. Lynnfield (the iMac's Core i7) is 45nm fab. Clarkdale OTOH belongs to an entirely new generation (Westmere generation) based on a 32nm process with integrated switchable graphics core. Performance-wise these CPUs might be similar, but in terms of design and fabrication they are inherently different.

Call it "core architecture-wise", then. Note that the memory controller is not integrated on the CPU die of Clarkdale - it is in the package, but off the die. When this was the case for L2 cache back in the day (The G3 and Pentium III both had models with the cache on and off the die), it made a huge difference. How much of a difference this makes for Clarkdale remains to be seen. I also don't know on which die the L3 cache is.

Clarkdale (Core i5 670) seems to peak at 3.46, with 2 bins of Turbo with one core active. Lynnfield (core i7 870) reaches 3.46 with two cores active, but only 1 bin higher with a single core active. Given that Lynnfield has twice the L3 cache and a likely lower latency to main memory, it should remain superior even on single and dualthread tasks - excluding AES-related operations.

In terms of marketing there is a similarity though. Lynnfield is a 'dumbed down' version of Bloomfield. And that is what Clarkdale is to Westmere.

That's a bit rough. Lynnfield has higher turbo, a cheaper motherboard design and comes in units with much lower TDP. Lynnfield is an evolution of Bloomfield, losing the third memory channel while increasing memory bus speed to compensate. Bloomfield really only makes sense for Xeons that need dual sockets.

Also, Westmere is process step name, comparable to Nehalem - not an actual processor.

[P]

Originally Posted by Eug 

Back to iMacs... I don't think Apple will update the iMacs until Intel updates the quad-core CPUs. So I foresee an update to the iMacs in the second half of 2010:

27" iMac Core i7 quad-core
27" iMac Core i5 quad-core
27" iMac Core i5 dual-core
21.5" iMac Core i5 dual-core

Given the cost of things, I think that two lower models will remain 21.5" for now. The question is how far up Apple will leave us using the Clarkdale integrated graphics. Of course, they might put a Core i5 6x0 (ie, Clarkdale dualcore with turbo) in all the three lower models but give us discrete graphics in the middle two anyway. OTOH, Intel might release a slower, cheaper Core i5 740 or 730 or something.

[Eug]

Originally Posted by P 

Given the cost of things, I think that two lower models will remain 21.5" for now. The question is how far up Apple will leave us using the Clarkdale integrated graphics. Of course, they might put a Core i5 6x0 (ie, Clarkdale dualcore with turbo) in all the three lower models but give us discrete graphics in the middle two anyway. OTOH, Intel might release a slower, cheaper Core i5 740 or 730 or something.

Yeah, I was just referring to the CPUs, not the iMac models per se. To flesh that out:

iMac 21.5" with Core i5 dual-core and integrated graphics.
iMac 21.5" with Core i5 dual-core and slower discrete graphics
iMac 27.0" with Core i5 dual-core and slower discrete graphics
iMac 27.0" with Core i5 quad-core and faster discrete graphics
iMac 27.0" with Core i7 quad-core and faster discrete graphics <-- CPU upgrade of Core i5 quad model.

ie. 3 dual-core models and 2 quad-core models. I would have hoped for more quad-core models, but judging by Intel's roadmap, I'm not counting on it.

[Pierre B.]

Originally Posted by Eug 

I just hope I get my Core i7 before Xmas.

Hmm... see what happened here.

[P]

Originally Posted by Eug 

Yeah, I was just referring to the CPUs, not the iMac models per se. To flesh that out:

iMac 21.5" with Core i5 dual-core and integrated graphics.
iMac 21.5" with Core i5 dual-core and slower discrete graphics
iMac 27.0" with Core i5 dual-core and slower discrete graphics
iMac 27.0" with Core i5 quad-core and faster discrete graphics
iMac 27.0" with Core i7 quad-core and faster discrete graphics <-- CPU upgrade of Core i5 quad model.

ie. 3 dual-core models and 2 quad-core models. I would have hoped for more quad-core models, but judging by Intel's roadmap, I'm not counting on it.

That looks more like what I'd expect. The issue is that with Clarkdale being 32 nm dualcore, it should be a quarter the size of Lynnfield excluding GMCH, and even Gulftown should be smaller. Size=cost, so Intel has no real incentive to sell Lynnfield unless AMD manages to light a fire under them. It could happen - all they need to do is to sell quadcores that outperform Clarkdale without running too hot - but I don't see it.

Originally Posted by Pierre B. 

Hmm... see what happened here.

Mine is still on order, no cancellation.

[Eug]

Originally Posted by P 

That looks more like what I'd expect. The issue is that with Clarkdale being 32 nm dualcore, it should be a quarter the size of Lynnfield excluding GMCH, and even Gulftown should be smaller. Size=cost, so Intel has no real incentive to sell Lynnfield unless AMD manages to light a fire under them. It could happen - all they need to do is to sell quadcores that outperform Clarkdale without running too hot - but I don't see it.

Lynnfield is listed only to 2010 Q3 in the last roadmap I've seen. I'm wondering if an updated roadmap will have a 32 nm quad-core Westmere/Nehalem-C desktop chip for 2010 Q4. (I don't see an iMac getting Gulftown since it's a 130 Watt part.)

I don't completely rule out the possibility of a Core i7 45 nm update either. Expensive in terms of area, yes, but IIRC, Intel has done stuff like that before.

Mine is still on order, no cancellation.

Same here. I'm getting a bit annoyed though as we're already into the second week of November (and all those 1st-week-of-November predictions by stores, etc. have proven to be false). I just hope it's nothing like those several month iMac G5 delays, which it shouldn't be, since all indications are that Intel has no problems with Core i7 Lynnfield yields.

[Simon]

Originally Posted by Eug 

I'm wondering if an updated roadmap will have a 32 nm quad-core Westmere/Nehalem-C desktop chip for 2010 Q4.

The original Nehalem-C was 'officially' canned a while ago. There's no quad-core Westmere expected, even now. Sandy Bridge is still quite far out, but it should bring us really nice low-power desktop quads (beyond the first stepping that is).

Some i5 iMacs have shipped. However, other i5/i7 iMac orders have been canceled.

[Eug]

Originally Posted by Simon 

The original Nehalem-C was 'officially' canned a while ago.

I thought Westmere was just a name change for Nehalem-C.

There's no quad-core Westmere expected, even now.

I know, but I was saying it was always still possible Intel could release one. Alternatively, Intel could just release more Lynnfields at faster clock speeds, to bridge the gap. Not that I really care that much though, since I'll have my 2.8/3.46 GHz Lynnfield, and I plan on keeping it for a few years.

The only two things I'm wanting in the iMac at this point are Blu-ray and USB 3, but I'm perfectly happy to get an external Blu-ray drive, and USB 3 is something I don't I'll really care about until 2011 anyway.

Some i5 iMacs have shipped.

Oooohhh... I didn't know that. Nice to hear - Not quite shipped but imminent. I started a shipping thread here.

I guess the hold up for Core i7 must partially be because of the Configure-To-Order nature of the Core i7 machines. There is no standard Core i7 machine. Still this suggests to me that my Core i7 could possibly ship by the end of the week. I (re-)ordered the afternoon of the first day. My CTO is i7 plus 2 TB and wired keyboard.

[P]

A Lynnfield/Clarksfield die-shrink might yet happen if:

1) Sandy Bridge is 2H 2011 rather than 1H 2011
2) AMD starts showing some decent performance
3) Intel can't rev either Clarkdale/Arrandale any further

Intel is run based on quarterly sales numbers. Say that AMD can cook up something that hurts the Core i5 600 series and combine it with good integrated graphics of their own for 2010, and then start talking about Fusion. Nehalem will still rule the server space, but an attack on the midrange mobile and desktop will make Intel sit up. The way out then is a pricecut on Lynnfield, and Intel might simply decide that an investment in a 32nm Lynnfield might be worth it. I would be surprised if Intel doesn't have that as a backup option.

The reason this won't ever happen is that I think Clarkdale will be able to rev at least 2 bins over what it launches at.

[downinflames68]

Originally Posted by Simon 

Umm, no. Actually you were ranting a something Apple already put in there. Presumably because you prefer to rant and use foul language to appear as the tough guy rather than read specs. Shame.

Uh, I've been bitching about this for YEARS, not just in this thread. DUHHHHHHHHHH.

[Simon]

And continued bitching about it even after they solved the issue made sense why again?

[downinflames68]

Originally Posted by Eug 

I think that's one big reason why Apple has done this. iMacs were more and more becoming ginormous laptops, but more and more people were just getting external screens instead to go along with their laptops, esp. since with the iMac, you weren't actually getting any real benefit performance-wise.

While I don't expect a socketed iMac will be upgradable for as long, I do expect to see a few Mac repair shops doing the upgrade (assuming it's actually possible). The all-in-one-ness makes that unlikely to be a mainstream option though.

Only if you're talking about base model iMacs. The top of the line iMac has had a pretty awesome video card option for a while now, which makes it easily kick the snot out of MBP.

As for it being socketed... again... even the ORIGINAL 24" iMac was socketed. The problem isn't upgrading the CPU, RAM or HD, its' been the GPU. Always.

[mduell]

Originally Posted by Eug 

Intel is revamping the dual-core lines and is claiming that its top-of-the-line dual-core Clarkdale processors will be faster than Core 2 Quad, at least in SPEC fpu.

No surprise in a single threaded benchmark.

Originally Posted by P 

That's a fair comparison. I agree that PVA is an acceptable substitute (TN is not), but where did you spec that model?

I went to Dell and picked some random machine from their gaming lineup.

[Eug]

Originally Posted by mduell 

No surprise in a single threaded benchmark.

I guess you didn't bother reading the link. They were referring to SPECfp_rate2006.



Clarkdale does well in SPECint_rate2006 too, with dual-core 3.33 GHz Clarkdale almost matching 2.66 GHz Core 2 Quad.

[P]

Weird comparison - far from the top C2Q, and not quite the top Clarkdale... Ah wait, no: all three processors cost the same, adjusting for the integrated MCH. That says something about how far up the chain Intel places Clarkdale.

Note that a Phenom II X4 955 also costs the same. That's the second fastest chip AMD sells. If you ever wondered why Intel doesn't push Lynnfield harder - that's why.

[mduell]

Originally Posted by Eug 

I guess you didn't bother reading the link. They were referring to SPECfp_rate2006.


Clarkdale does well in SPECint_rate2006 too, with dual-core 3.33 GHz Clarkdale almost matching 2.66 GHz Core 2 Quad.

True. Really hard to read the text on those scaled down slides.

[P]

Originally Posted by mduell 

I went to Dell and picked some random machine from their gaming lineup.

I guess I fail, then, because I can't make that website show me anything over 24". Could be that Dell.com detects that I'm in Sweden every now and then, with the result that they kick me out to the Swedish frontpage rather than letting me keep working in the country version I manually selected. That's almost enough reason right there not to buy Dell.

[P]

Originally Posted by downinflames68 

Only if you're talking about base model iMacs. The top of the line iMac has had a pretty awesome video card option for a while now, which makes it easily kick the snot out of MBP.

At the same resolution, sure. At the native resolution of each machine?

The top iMacs have switched to decent upper midrange cards, but they're no graphics monsters. Never were.

Originally Posted by downinflames68 

As for it being socketed... again... even the ORIGINAL 24" iMac was socketed. The problem isn't upgrading the CPU, RAM or HD, its' been the GPU. Always.

The model just before this one was not socketed.

Note that this will likely be of very minor importance in this case. With Sandy Bridge moving to a new socket again - LGA1155, as opposed to LGA1156 for Lynnfield - the limits of your upgrades in Core i5/i7 iMac are likely to be a Core i7 870. That's a minor boost at best for a i5 750, and minuscule one for a Core i7 860. For the Core 2s, there will be no faster Duos released ever. A Core 2 Quad 9550s might fit, but is overpriced for what you get, and there is no BIOS to let you undervolt a regular 9550 to the correct voltage. You RAM truly is maxed at 16 gigs - Intel reports the same max, so it's no Apple limitation - so the one significant thing you might upgrade down the line is to squeeze in an SSD, as I've mused about around here before.

[mduell]

Originally Posted by P 

I guess I fail, then, because I can't make that website show me anything over 24". Could be that Dell.com detects that I'm in Sweden every now and then, with the result that they kick me out to the Swedish frontpage rather than letting me keep working in the country version I manually selected. That's almost enough reason right there not to buy Dell.

I already knew the display price, so I was configuring the box (Studio XPS 8000) with no display.

But you're right, they don't offer the 27" with a the XPS 8000. They do offer it as an "additional display" with the XPS 9000. I guess the volume is so low they don't even try to push it.

[P]

Why? They obviously make them, why not offer them as an option with everything. With that model, they offer 5 displays: 21.5", 23", 20", 23" and 24". Awesome. Either you reduce the options because you think options scare people - in which case 5 is way too much - or you just offer everything.

[Eug]

Originally Posted by P 

so the one significant thing you might upgrade down the line is to squeeze in an SSD, as I've mused about around here before.

Might be a problem:

AppleInsider | Review: Apple's 27" big screen iMac (late 2009) [Page 2]

According to Other World Computing, Apple currently appears to be using hard drives from only two vendors: Western Digital and Seagate. System Profiler, which can be launched from the "About this Mac" in the Apple menu, will identify your installed hard drive by serial number, and designates its manufacturer by the first two letters of that number, either WD or ST. Because each vendor supplies a unique thermal sensor connector, upgrading users of the "Late 2009" iMacs will need to buy the same type of hard drive in order to use the same included cable installed inside their iMac. If another drive mechanism is installed, the system will not be able to sense the drive's temperature and will subsequently send the system's fans spinning into overdrive.

[P]

Originally Posted by Eug 

Might be a problem:

AppleInsider | Review: Apple's 27" big screen iMac (late 2009) [Page 2]

According to Other World Computing, Apple currently appears to be using hard drives from only two vendors: Western Digital and Seagate. System Profiler, which can be launched from the "About this Mac" in the Apple menu, will identify your installed hard drive by serial number, and designates its manufacturer by the first two letters of that number, either WD or ST. Because each vendor supplies a unique thermal sensor connector, upgrading users of the "Late 2009" iMacs will need to buy the same type of hard drive in order to use the same included cable installed inside their iMac. If another drive mechanism is installed, the system will not be able to sense the drive's temperature and will subsequently send the system's fans spinning into overdrive.

Interesting - thanks for the warning. I don't plan on replacing the HD, though - the plan is one extra drive or replacing the DVD, and in either case it won't happen now.

[Eug]

Core i5 iMac benches

We've also noticed that the quad-core processor doesn't seem to have a significant bearing on the noise of the system. Even in CINEBENCH R10, a 3D rendering test that taxes both the processor and graphics, the system didn't become noisy.

[Simon]

Originally Posted by Eug 

Might be a problem:

AppleInsider | Review: Apple's 27" big screen iMac (late 2009) [Page 2]

So in practical terms the iMac just became even less upgradable? Too bad Apple couldn't wait until HDD manufacturers settled on a standard connector type for thermal sensors.

[Eug]

iMac 11,1 GeekBench scores
Core i7 64-bit - 9638 (int 8537, fp 14895)
Core i7 32-bit - 8410 (int 6857, fp 14006)
Core i5 64-bit - 7131 (int 6584, fp 10126)


iMac 10,1 GeekBench scores
3.33 GHz 64-bit - 5131
3.33 GHz 32-bit - 4707
3.06 GHz 64-bit - 4694
3.06 GHz 32-bit - 4324

[P]

That Core i7 iMac is almost as fast as the last gen octocore Mac Pro @3GHz - or put another way, according to Geekbench the 8 virtual cores in the iMac do as much work as the 8 actual cores in the octocore. That sounds...excessive.

The scores are wildly different, though: the MP has higher scores for both integer and float, but is absolutely toast in the memory benchmarks.

[Eug]

Originally Posted by P 

That Core i7 iMac is almost as fast as the last gen octocore Mac Pro @3GHz - or put another way, according to Geekbench the 8 virtual cores in the iMac do as much work as the 8 actual cores in the octocore. That sounds...excessive.

The scores are wildly different, though: the MP has higher scores for both integer and float, but is absolutely toast in the memory benchmarks.

Mac Pro with Xeon X54xx





iMac with Core i7



Summary
Core i7 iMac is as fast as the 2008 3.2 GHz 8-core Mac Pro for floating point.
Core i7 iMac is as fast as the 2008 2.8 GHz 8-core Mac Pro for integer.

[mduell]

The figures for i7 appear credible based on published results, but the 5400 Xeon int figures are low compared to published results.

[Eug]

One factor to consider is that the Mac Pro results would have been done on 32-bit 10.5 Leopard.

The 32-bitness vs 64-bitness may make a significant difference (as evidenced by the Geekbench scores), and the Linux SPEC benches you posted are with 64-bit Linux. If I can get all my drivers to work, I might run exclusively in 64-bit mode.

Either way though, the iMac Core i7 is a computational monster.

[Veltliner]

Is there so much difference between an 4-core i5 and an i7?

To the naked eye it looks like the difference between 2.66 and 2.8 Ghz.

Or are there more differences between these two processors than the Ghz rate?

If not, then the i5 should be an equally "computational monster".

[Eug]

Originally Posted by Veltliner 

Is there so much difference between an 4-core i5 and an i7?

To the naked eye it looks like the difference between 2.66 and 2.8 Ghz.

Or are there more differences between these two processors than the Ghz rate?

If not, then the i5 should be an equally "computational monster".

Core i7 supports Hyper-Threading, which means the OS sees it as an 8-core machine. Obviously, it's not twice as fast, but for most usage, HT gives an additional boost to performance.

Core i5 is the better bang for the buck but Core i7 is on average noticeably faster.

Intel Core i7 965 Extreme
Benchmark Hyper-Threading
Crysis -0.1%
Unreal Tournament 3 -2.1%
World in Conflict -2.7%
Supreme Commander 0.3%
AVG Anti-Virus 8 6.6%
Winrar 3.80 14.8%
Winzip 11 -1.0%
Acrobat 9 Professional -1.0%
Photoshop CS 3 -1.1%
iTunes 0.0%
Lame MP3 0.0%
Studio 12 -1.9%
DivX 18.9%
XviD 0.7%
MainConcept H.264 18.3%
Premiere Pro CS3 HDTV 7.8%
Cinema 4D Release 10 8.5%
3D Studio Max 9 10.3%
Fritz 11 23.8%
Nero 8 Recode 23.2%
Overall: 6.2%

As you can see here, games don't benefit from HT, but stuff like video encoding and 3D modelling benefit significantly from HT.

[Eug]

Originally Posted by Eug 

One factor to consider is that the Mac Pro results would have been done on 32-bit 10.5 Leopard.

The 32-bitness vs 64-bitness may make a significant difference (as evidenced by the Geekbench scores), and the Linux SPEC benches you posted are with 64-bit Linux. If I can get all my drivers to work, I might run exclusively in 64-bit mode.

Either way though, the iMac Core i7 is a computational monster.

Hmmm... I just ran Geekbench 32-bit and 64-bit versions on my MacBook Pro. Yes the 64-bit version is much faster, but it doesn't matter what Snow Leopard kernel is loaded. The 64-bit scores for Geekbench are identical regardless if you're running 32-bit SL or 64-bit SL.

Nonetheless, it still could be that 64-bit Snow Leopard 10.6 is much better optimized for SPEC than 32-bit Leopard 10.5 from Jan 2008.

[Eug]

Haha... For comparison to this Core i7 Cinebench result, the 1.7 GHz Cube I'm typing on right now gets just 806 for the CPU rendering.



That makes the Core i7 iMac 19X as fast as my 1.7 GHz G4.

P.S. Veltliner, check out the number of cores listed. The multiprocessor speedup is still less than 4X though.

[mduell]

Originally Posted by Eug 

One factor to consider is that the Mac Pro results would have been done on 32-bit 10.5 Leopard.

The 32-bitness vs 64-bitness may make a significant difference (as evidenced by the Geekbench scores), and the Linux SPEC benches you posted are with 64-bit Linux. If I can get all my drivers to work, I might run exclusively in 64-bit mode.

The 64-bit i7 benchmarks I linked to were about the same as Apple's figures. I wouldn't put much stock in Apple's estimates since they're unpublished.

But yea, Nehalem is a monster. The current Mac Pros are well into the 200s.

[Veltliner]

Originally Posted by Eug 

Core i7 supports Hyper-Threading, which means the OS sees it as an 8-core machine. Obviously, it's not twice as fast, but for most usage, HT gives an additional boost to performance.

Core i5 is the better bang for the buck but Core i7 is on average noticeably faster.

Intel Core i7 965 Extreme
Benchmark Hyper-Threading
Crysis -0.1%
Unreal Tournament 3 -2.1%
World in Conflict -2.7%
Supreme Commander 0.3%
AVG Anti-Virus 8 6.6%
Winrar 3.80 14.8%
Winzip 11 -1.0%
Acrobat 9 Professional -1.0%
Photoshop CS 3 -1.1%
iTunes 0.0%
Lame MP3 0.0%
Studio 12 -1.9%
DivX 18.9%
XviD 0.7%
MainConcept H.264 18.3%
Premiere Pro CS3 HDTV 7.8%
Cinema 4D Release 10 8.5%
3D Studio Max 9 10.3%
Fritz 11 23.8%
Nero 8 Recode 23.2%
Overall: 6.2%

As you can see here, games don't benefit from HT, but stuff like video encoding and 3D modelling benefit significantly from HT.

It looks a little bit like those extra $200 would give you some real value here.

If hyperthreading gets more and more implemented in complex applications, then the i7 would even gain more on the i5 in a near future (and how about hyperthreading in 10.7? - looks like hyperthreading will be the point of attack in all software development today and in the near future).

Basically: if you already spend $2000, here it really pays to spend $2200 on the i7?

I wonder why the i5 doesn't have this hyperthreading feature... (hardware or firmware related?)

[mduell]

Originally Posted by Veltliner 

If hyperthreading gets more and more implemented in complex applications, then the i7 would even gain more on the i5 in a near future (and how about hyperthreading in 10.7? - looks like hyperthreading will be the point of attack in all software development today and in the near future).

There's no application implementation needed, and not a lot you can even do to optimize for hyperthreading. It helps when you stall, so the best you can do is not stall which... eliminates the upside of hyperthreading.

Originally Posted by Veltliner 

Basically: if you already spend $2000, here it really pays to spend $2200 on the i7?

Probably, although spending that $200 on memory may be a better tradeoff depending on your workload.

Originally Posted by Veltliner 

I wonder why the i5 doesn't have this hyperthreading feature... (hardware or firmware related?)

Market segmentation.

[Veltliner]

So you're saying that hyperthreading features isn't that big a thing?

[Eug]

Originally Posted by mduell 

The 64-bit i7 benchmarks I linked to were about the same as Apple's figures. I wouldn't put much stock in Apple's estimates since they're unpublished.

I would. IBM's SPEC scores are often unpublished. So are Intel's.

I suspect that Apple's unpublished scores were the absolute best they could achieve at the time, with their own OS. Apple just doesn't officially publish their SPEC results. At least it's better than before, since before in the G4 days, Apple didn't unofficially publish any SPEC results either, presumably because the G4 SPEC scores sucked royally.

Anyways, overall I still think that Core i5 is the best bang for the buck, even though I bought the faster Core i7. In fact I usually say when buying a computer, the slowest CPU is often the best bang for the buck. However, this time around it isn't by a long shot. The Core 2 Duo 3.06 and 3.33 are just soooooo much slower than the Core i5/i7. Quite frankly, while I can sort of understand why somebody might consider getting the 3.06 27-incher, I don't understand why anyone would buy the 3.33 for the 27" model. It actually costs more than the Core i5 if you get the same Radeon 4850. The only reason the 3.33 can be cheaper is if you get the much slower 4670.

Originally Posted by Eug 

3.06 GHz Core 2 Duo with Radeon 4850 - US$1849
3.33 GHz Core 2 Duo with Radeon 4850 - US$2049
2.66 GHz Core i5 Quad with Radeon 4850 - US$1999
2.80 GHz Core i7 Quad with Radeon 4850 - US$2199

However, even if you get the 4670, getting the 3.33 GHz Core 2 Duo means an $1899 price, just $100 cheaper than the Core i5 with 4850.

Originally Posted by Veltliner 

So you're saying that hyperthreading features isn't that big a thing?

Not having HT is not a dealbreaker, but it's still significant. The question is if it's worth the extra cash. If you're on a budget, it can make sense to save the $200 and put it to something else like more memory or a bigger hard drive, or an external Firewire 800 hard drive.

[P]

Apple has published some SPEC results, but they're all terrible for some reason. I don't think that a single Lynnfield beats 2 Yorkfields on a majority of tasks, but it is clear that it is a major step forward.

[Eug]

It looks like someone out there was reading this thread.

Originally Posted by Eug 

iMac 11,1 GeekBench scores
Core i7 64-bit - 9638 (int 8537, fp 14895)
Core i7 32-bit - 8410 (int 6857, fp 14006)
Core i5 64-bit - 7131 (int 6584, fp 10126)

iMac 10,1 GeekBench scores
3.33 GHz 64-bit - 5131
3.33 GHz 32-bit - 4707
3.06 GHz 64-bit - 4694
3.06 GHz 32-bit - 4324

P.S. Here are a few previous gen and current gen Mac Pro octocore benches:

Mac Pro 2.80 GHz (early 2008) 8-core 32-bit - 10334
Mac Pro 2.80 GHz (early 2008) 8-core 64-bit - 11265
Mac Pro 2.92 GHz (early 2009) 8-core 64-bit - 17248

[mduell]

Eug can you tell Cinebench to only run 4 threads and see what the score is?

Originally Posted by Veltliner 

So you're saying that hyperthreading features isn't that big a thing?

Hyperthreading can be a big thing, but you don't have to do anything for it. That's part of what makes it great.

Originally Posted by Eug 

I would. IBM's SPEC scores are often unpublished. So are Intel's.

I suspect that Apple's unpublished scores were the absolute best they could achieve at the time, with their own OS. Apple just doesn't officially publish their SPEC results. At least it's better than before, since before in the G4 days, Apple didn't unofficially publish any SPEC results either, presumably because the G4 SPEC scores sucked royally.

IBM has published over 3000 SPEC CPU2006 scores for everything from x64 servers to Cell to i Series to RS/6000. Intel has published nearly 1000. Apple did publish scores for an 8 core 3.0Ghz box, but those aren't the ones you cited from their website. Estimates are useful for prerelease hardware, but once it's available anything other than published results is marketing bullshit.

Originally Posted by P 

I don't think that a single Lynnfield beats 2 Yorkfields on a majority of tasks, but it is clear that it is a major step forward.

I'd suspect it does for most uses, thanks to the integrated memory controller and non-FB RAM.

[Eug]

Originally Posted by mduell 

Eug can you tell Cinebench to only run 4 threads and see what the score is?

No you cannot. However, on Windows you can shut off Hyper-Threading and then re-run Cinebench. The speedup is around 8-10% on Bloomfield Core i7 with HT active.

IBM has published over 3000 SPEC CPU2006 scores for everything from x64 servers to Cell to i Series to RS/6000. Intel has published nearly 1000. Apple did publish scores for an 8 core 3.0Ghz box, but those aren't the ones you cited from their website.

They don't have a Mac Pro listed, but yes they do have an Xserve.

int_rate - 106
fp_rate - 74

These results are roughly similar to the 2.8 and 3.2 GHz (early 2008) Mac Pro benches that Apple used in their advertising... and also roughly similar to Apple's iMac Core i7 scores (which we already know are in the right ballpark by comparing to published non-OS X Core i7 SPEC results).

BTW, the bench is compiled with icc, not gcc. I wonder how much of OS X and Apple software is compiled with icc vs. gcc.

Estimates are useful for prerelease hardware, but once it's available anything other than published results is marketing bullshit.

Yet, these companies continue to do it. Anyways the "marketing bullshit" comment seems rather odd, considering that most of the "estimates" from Apple are actually either comparable to Windows, or else worse performance. Given that observation, it'd be hard to argue Apple is lying about the numbers.

[Eug]

It's interesting to compare Core 2 Duo power usage vs Core i5/i7

At idle, Core i5/i7 Lynnfield wins by a long shot:



Under load, Core i5/i7 uses significantly more power:



We already knew Core i5/i7 would use a lot more power, as it has a 95 Watt TDP, vs. 65 Watts for the 3.33 GHz Core 2 Duo. However, the idle power measurements just go to show you how well Lynnfield's power reduction mechanisms are.

At idle, the Core i5 system uses a whopping 100 Watts less. (Yes the TDP is only 95 Watts, but there are other aspects of the system that draw power.) At idle, the Core 2 Duo uses 35 Watts less than it does at full tilt. Furthermore, at idle, the Core 2 Duo system uses 40 Watts more than the Core i5 system. That's a pretty huge difference.

Thus for most usage, it seems a 95 Watt TDP Core i5/i7 would overall use less power than a 65 Watt TDP Core 2 Duo, since most of the time our computers are at or near idle.

P.S. Electronista's full Core i5 review is up.

----

Here are the numbers for Apple's Adjusted Peak Performance (APP) in Weighted Teraflops (WT) for export compliance:

Mac Pro 3.2 8-core: 30.7 Gigaflops <-- 2008 model
Mac Pro 3.0 8-core: 28.9 Gigaflops <-- 2008 model
Mac Pro 2.93 8-core: 28.1 Gigaflops
iMac Core i7 4-core: 28.1 Gigaflops
Mac Pro 2.8 8-core: 26.8 Gigaflops <-- 2008 model
Mac Pro 2.66 8-core: 25.5 Gigaflops
iMac Core i5 4-core: 25.5 Gigaflops
Mac Pro 2.26 8-core: 21.7 Gigaflops
iMac C2D 3.06 2-core: 7.3 Gigaflops <-- Ouch

These numbers don't make much sense to me. However, here's Wikipedia's page on Adjusted Peak Performance, and here are some numbers from Intel.

[P]

Originally Posted by mduell 

Originally Posted by P

I don't think that a single Lynnfield beats 2 Yorkfields on a majority of tasks, but it is clear that it is a major step forward.

I'd suspect it does for most uses, thanks to the integrated memory controller and non-FB RAM.

And why weren't you in that thread over in the MP forum where we compared Lynnfield to two Woodcrests?

The more I read about the dual Xeon 5300/5400 setups, the less I like them. They were severely bandwidth limited on any reasonable task, and there was a lot of snoop traffic as soon as several cores were working on the same dataset. If Intel had forgotten about quads, SSE4 and hyperthreading, Nehalem would still be a monster based on the integrated memory controller alone. I wonder how good the bargain basement Core i3 and Pentium G6xxx will do?

[P]

Originally Posted by Eug 

Here are the numbers for Apple's Adjusted Peak Performance (APP) in Weighted Teraflops (WT) for export compliance:

Mac Pro 3.2 8-core: 30.7 Gigaflops <-- 2008 model
Mac Pro 3.0 8-core: 28.9 Gigaflops <-- 2008 model
Mac Pro 2.93 8-core: 28.1 Gigaflops
iMac Core i7 4-core: 28.1 Gigaflops
Mac Pro 2.8 8-core: 26.8 Gigaflops <-- 2008 model
Mac Pro 2.66 8-core: 25.5 Gigaflops
iMac Core i5 4-core: 25.5 Gigaflops
Mac Pro 2.26 8-core: 21.7 Gigaflops
iMac C2D 3.06 2-core: 7.3 Gigaflops <-- Ouch

These numbers don't make much sense to me. However, here's Wikipedia's page on Adjusted Peak Performance, and here are some numbers from Intel.

I played with some OpenCL benchmarks yesterday, and I got better than that in actual, real world numbers. Either they're reporting only for the CPU, or the tests I was playing with are single precision.

That 8 C2D cores is better than i7 here is nothing to care about - i7 is mostly about the memory latency and bandwidth, and the APP numbers are not adjusted for that. The C2D might easily have a higher theoretical number, but it will have a much lower number in real world tests.