Intel Shows Off Conroe (Desktop) Processors

[Multimedia]

Merom Will Work In First Gen iMacIntels and Mac mini Intels - Cool

Merom will be compatible with the current iMac and MacMini !!!!!!!!
http://www.anandtech.com/tradeshows/showdoc.aspx?i=2715&p=2

Have to admit this is pretty amazing. I think I posted earlier that Merom wouldn't work with Yonah chipsets. I was mistaken. Too bad they didn't put a socket inside the MacBook Pro. That really blows for all the MBP buyers. But the iMac and Mac mini buyers are good to go for an easy off the shelf processor upgrade. All they have to do is figure out how to get inside those difficult to open up models. By the looks of that chart though, I am more inclined than ever to wait for the 2007 support chips so I can have faster 802.11n wi-fi and graphics for the long term.

Good find PITT.

 

[Multimedia]

Toshiba Core Duo 17" WUXGA Laptop with HD-DVD Drive

Now to observe what Apple may be aiming at with their first MBP 17" Model:

Toshiba Core Duo 17" Laptop with HD-DVD Drive. The HD-DVD drive can write 320 GB to one HD-DVD platter. WUXGA is 1920 x 1200

 

[AidenShaw]

You need 64-bit hardware *and* software

I don't really want a 32-bit processor in a portable when I am driving four 64-bit processors on the desktop already.

The only initial advantage to Merom will be future-proofing - you'll be "driving" that Merom MBP in 32-bit mode at 32-bit speeds until

1. Apple comes out with a true 64-bit OSx64 operating system (Apple has to say something about this at WWDC - look for the "Leopard x64" announcements)
2. Applications are ported to 64-bit for the extra speed

But, since probably all the applications that you run on the quad are 32-bit, "future-proofing" will be good enough.

Intel says that Merom is 20% faster than Yonah. 20% is also a typical figure for the speedup going from 32-bit x86 to 64-bit x64....

I wonder if Intel is comparing 64-bit Merom to 32-bit Yonah - and the speed boost is only seen in x64 mode?

 

[thunng8]

Spec comparisons

You don't know what you are talking about.

Back when the G5 actually showed up on spec.org it got trounced by the Pentium-M (Dothan). Something like 1300 vs 900 integer and 700 vs 900 (the other way) for floating point. Yonah is 2 generations newer than Dothan, is faster, has more cache, and also has 2 cores.

The Single G5 has NEVER been faster than a comparable pentium 4. It may have squeaked by in floating point, but the integer gap has always been humongous.

Not exactly true. We do have a new datapoint with the release of the 2.7Ghz 970MP in the IBM JS21 Blade system.

http://www.spec.org/cpu2000/results/res2006q1/

2.7Ghz 970MP
SpecInt2000:1706
SpecFP2000:2259

2-way 2.7Ghz 970MP
SpecIntRate:38.5
SpecFPRate:43.9

4-way 2.5Ghz 970MP
SpecIntRate:67.5
SpecFPRate:58.8

As a comparison, the 2.16Ghz Core Duo (2-way):
SpecIntRate:34.9
SpecFPRate:27.3

And best current P4 system:

3.46Ghz P4EE (2-way):
SpecIntRate:41.1
SpecFPRate:36.7

However, the most remarkable figures are the Power5+ based p5-575 system (up to 16 core 2U rack system)

8-way 2.2Ghz
SpecIntRate:200
SpecFPRate:382

16-way 1.9Ghz
SpecIntRate:314
SpecFPRate:571

Truly amazing numbers - nothing at the moment comes remotely close to this performance (with same number of cores).

 

[thunng8]

That is most likely not the case. The reason that Intel manages to get their Netburst chips to those high frequencies is mostly because of the very long pipeline. I don't know how familiar you are with pipelining, but, basically, the longer the pipeline you have the shorter the signal delay between each stage and the higher you can go with the fequency. Just look at the differences in max frequency between Yonah and Dothan compared to the P4 and Pentium-D. The former use a pipeline length similar to the 970 and they are also able to run at about the same frequencies.

That is generally the case - however IBM has revealed that their next generation POWER6 chip has the same number of pipeline stages the POWER5 but will clock 2X higher.

http://news.zdnet.com/2100-9584_22-6039293.html

"We doubled the frequency but held the pipe depth the same as Power5."

Shipping frequency will be 4-5Ghz and they have the POWER6 chip running at 5.1Ghz in the labs with standard voltages (and up to 6Ghz with higher voltages).

 

[thunng8]

People forget that an iMac really is of laptop form factor...

Yes, it is slightly larger than a laptop, (and does incorperate a much bigger screen) but it is still only a few inches deep. The heatsink and fan on my A64 3500 is deeper than the whole iMac!

And its the form factor that makes me want an iMac soooooo bad....

I think Apple will change to Conroe with the iMac when that chip is introduced.

The reason for putting a Core Duo now is that the dual core P4 is way too hot. With Conroe being much cooler, I don't see any more reason on putting a laptop chip in an iMac. The mac mini on the other hand will likley continue with laptop chips because it is indeed very small.

 

[janstett]

INTEL Duo does not beat the single core G5 clock for clock, hell the Intel duo has trouble vs the single core G4.

Sigh... It must be some strange anomaly, then, that my MacBook Pro D-E-S-T-R-O-Y-S my G4 Mac Mini by a factor of 5:1 on Handbrake. I mean, the Core Duo has trouble vs. a single core G4, right?

INTEL chips are not better than PowerPC chips, Apple is forcing them upon Mac users and everyone is just jumping on the bandwagon blindly. If Apple "must" use INTEL chips they should leave rinky dink X86 chips for their consumer lines (iMac, iBook, Mac mini) and PowerPC chips for the pro's (Powerbook and PowerMac)

Right, and you know this based on your personal use of an X86 Mac, whereas people like me who actually OWN a PPC and an X86 Mac, well, we're just cwazy.

 

[AidenShaw]

check the website again

http://www.spec.org/cpu2000/results/res2006q1/
Truly amazing numbers - nothing at the moment comes remotely close to this performance (with same number of cores).

Check http://www.spec.org/cpu2000/results/ - you're looking only at the results submitted in the most recent quarter. If you look at "all results", there are Itanium II systems that are close to the POWER 5 numbers.

It would be more interesting to see SPECint/SPECfp (single stream) numbers for all the chips as well.

"SPECxx_rate" runs separate copies of the bench on each core, it gives a good idea of the memory system scales to support the additional cores (Opteron does rather poorly in this regard), but few single applications will see anything close to that much speedup. (Both numbers are useful)

 

[DeathChill]

Intel says that Merom is 20% faster than Yonah. 20% is also a typical figure for the speedup going from 32-bit x86 to 64-bit x64....

Uh, this has nothing to do with anything. 64-bit support wouldn't make a chip any faster, it just allows it to address more RAM (besides supporting 64-bit of course!).

 

[~Shard~]

I had to go look up a Wikipedia article because I don't have these names memorized. According to that, Kentsfield will be the four core version of Conroe, while the four core version of Woodcrest will be called Cloverton, or Clovertown depending on which article you look at.

So I'm saying that since it appears Intel is developing the Glidewell platform for using two Woodcrest processors in a workstation, it sounds reasonable that it would be the foundation for a new PowerMac, or Mac Pro.

I was trying to find a source the $850 Woodcrest price but couldn't.

Ah, I see. Cool, thanks for the info - makes sense.

 

[AidenShaw]

please read up on the x64 architecture

Uh, this has nothing to do with anything. 64-bit support wouldn't make a chip any faster, it just allows it to address more RAM (besides supporting 64-bit of course!).

True for PPC, not true for x64.

When running in 64-bit mode, the Intel architecture chip has more than twice as many usable integer/pointer registers, twice as many 128-bit SSE/FP registers, and a few other tweaks.

See http://en.wikipedia.org/wiki/X86-64

Benchmarks typically show an average of around 20% speed improvement for 64-bit applications/OS vs. 32-bit applications/OS running on the same system. This has been discussed many times in this forum - PPC slower in 64-bit mode, x64 faster in 64-bit mode.

For example, the following chart from BareFeats shows one comparison - the three PD lines show XP 32/64 bit with a 32/64 application:

32/32 = Windows XP 32-bit with 32 bit application
64/32 = Windows XP 64-bit with 32 bit application
64/64 = Windows XP 64-bit with 64 bit application
PD = Pentium Dual Core (Pentium-D)

http://www.barefeats.com/dualcore.html

 

[ksz]

Intel says that Merom is 20% faster than Yonah. 20% is also a typical figure for the speedup going from 32-bit x86 to 64-bit x64....

I wonder if Intel is comparing 64-bit Merom to 32-bit Yonah - and the speed boost is only seen in x64 mode?

The new Core architecture in Merom has these additional enhancements (click here):

1. Ability to issue and retire 4 instructions at once (4-wide architecture) compared to 3 for Yonah.

2. Macro-Op fusion in addition to micro-op fusion that is present in Yonah. Load and Compare operations, for example, are two macro-ops that can be combined and executed as if they were one op. Because Load/Compare occurs very often, macro-op fusion can provide improvements across a wide range of applications.

3. 4MB L2 cache versus 2MB on Yonah.

4. "Substantial enhancements have been made to the SSE engine. SSE registers are 128 bits wide, but in the past, data paths within the CPU were 64 bits wide. That meant that SSE instructions would take two cycles. In Conroe, the entire SSE engine is 128 bits, not just the registers. So SSE instructions can now execute in a single cycle."

There is no difference between Conroe and Merom other than:

1. "Conroe shares the architecture with the upcoming mobile equivalent, codenamed Merom. The main differences between the two revolve around implementation of power management and subtle differences in the manufacturing process."

 

[ksz]

Thanks for the reminder. I remember seeing $850, but forgot it included a link. I agree with your point there that Apple would want the new PowerMac to be priced about the same as the current edition. Checking the link gives me hope for a lower Mhz version at a lower price.

Two 3.0 Ghz Woodcrest chips = $1700
Two 2.0 Ghz Woodcrest chips = $ 660

I could be happy with a 2.0 Ghz Woodcrest "Quad."

From the link:

"The 5160 clocks at 3GHz, the 5150 at 2.66GHz, the 5140 at 2.33GHz, the 5130 at 2GHz, the 5120 at 1.86GHz and the 5110 at 1.6GHz.

The price cashcade when you buy loads of these is $850, $700, $470, $330, $270 and $230."

Agreed. At those prices, why not! Intel claims an 80% performance improvement over the current 2.8GHz Dual-Core Xeon (2x2MB Cache). Apple never had choices like this before...

 

[DavidCar]

Agreed. At those prices, why not! Intel claims an 80% performance improvement over the current 2.8GHz Dual-Core Xeon (2x2MB Cache). Apple never had choices like this before...

I wonder if Apple could create one motherboard with socketed processors that could be sold with 2Ghz Woodcrest dual core chips, and upgraded by either Apple or the user to 3 Ghz chips, or Cloverton quad core chips. Those would be great choices.

All I need now for my dream system is for EyeTV 500 to go MacIntel, and an Apple Cinema Display with an HDMI/DRM connector for next generation DVDs.

 

[DavidCar]

Benchmarks typically show an average of around 20% speed improvement for 64-bit applications/OS vs. 32-bit applications/OS running on the same system. This has been discussed many times in this forum - PPC slower in 64-bit mode, x64 faster in 64-bit mode.

I didn't realize that. So universal applications will have to come in PPC, Intel 32, and Intel 64 flavors?

 

[Anonymous Freak]

I didn't realize that. So universal applications will have to come in PPC, Intel 32, and Intel 64 flavors?

Unfortunately, true.

 

[thunng8]

Check http://www.spec.org/cpu2000/results/ - you're looking only at the results submitted in the most recent quarter. If you look at "all results", there are Itanium II systems that are close to the POWER 5 numbers.

Not even close:

8-way 2.2Ghz Power5+
SpecIntRate:200
SpecFPRate:382

8-way 1.67Ghz Hitachi IT2 (best IT2 result)
SpecIntRate:134
SpecFPRate:222

It would be more interesting to see SPECint/SPECfp (single stream) numbers for all the chips as well.

"SPECxx_rate" runs separate copies of the bench on each core, it gives a good idea of the memory system scales to support the additional cores (Opteron does rather poorly in this regard), but few single applications will see anything close to that much speedup. (Both numbers are useful)

Opteron does pretty good, it is the Intel Xeons that do very poorly:

Current best 8 way results:

Opteron 2.6Ghz 8-way
SpecIntRate:147
SpecFPRate:150

Xeon 3.0Ghz 8-way (HP DL580)
SpecIntRate:104
SpecFPRate:48.2

Note the extremely bad scaling of FP in the Xeon. This does not bode well for high core count Intel Powermac systems. The upcoming Bensley platform will help the Xeons out with FB-DIMM, independent buses (this will help fp_rate a lot) and Woodcrest. Intel has quoted a 80% increase in performance in specINT_rate of Woodcrest over current Xeons (not sure about core count of that comparison though).

Also note that the Power5+ system is almost 8X faster at specfp_rate with the same number of cores over current Xeon for 8 core system.

 

[AidenShaw]

Not even close:

Since you said "not even remotely close" originally, you've changed your position a bit

This was my bad - that one 2.2 GHz POWER5 line is a bit out-of-place - I missed it earlier.

Note the extremely bad scaling of FP in the Xeon.

I can't note "bad scaling" with a single data point.

IMO, it makes much more sense to look at SPECint_2000/SPECfp_2000 (base) than the peak rate numbers. This reflects more on what an application with a couple of active threads will see.

It doesn't seem that reasonable to compare enterprise systems in the $50,000 to $200,000 range - and then extrapolate to what that means for some future chips in a PowerMac....

 

[thunng8]

Since you said "not even remotely close" originally, you've changed your position a bit

This was my bad - that one 2.2 GHz POWER5 line is a bit out-of-place - I missed it earlier.

I can't note "bad scaling" with a single data point.

OK .. here are the results of that particular xeon system (specfp_rate):

2 core: 24.5
4 core: 32.1
8 core: 48.1

Do you admit the extremely bad scaling of the Xeon system on fp?

IMO, it makes much more sense to look at SPECint_2000/SPECfp_2000 (base) than the peak rate numbers. This reflects more on what an application with a couple of active threads will see.

Going for the base results doesn't change performance that much (~5% for POWER5 and very little difference for Xeon)

It doesn't seem that reasonable to compare enterprise systems in the $50,000 to $200,000 range - and then extrapolate to what that means for some future chips in a PowerMac....

You are right, the system I quote is rather expensive, however there are much cheaper Power5 systems.

POWER5 systems start from $3,399 (1.65Ghz P5-505) and $4,485(1.9Ghz P5-510). At the moment no entry level systems use the 2.2Ghz chip.

BTW, here are the results of the P5-510 1.9Ghz (2 core):

SpecIntRate:39.9
SpecFPRate:67.1

 

[AidenShaw]

Quit It With The Spec "rate" Citations

OK .. here are the results of that particular xeon system (specfp_rate):

2 core: 24.5
4 core: 32.1
8 core: 48.1

Do you admit the extremely bad scaling of the Xeon system on fp?

Do you honestly believe that SPECfp_rate measures floating point performance?

It doesn't, and nobody buys an MP Xeon system for what it does measure. Yes, it's obvious to everyone who really understands SPECfp_rate that large numbers of Xeons don't scale. Dell stopped selling 8-way Xeons - there's a reason.

Which is exactly why an 8-way PowerMacIntel will have two quad-core chips with a huge shared cache and a 1333 MHz FSB. Do you think that Intel isn't concerned about this issue, and isn't working on improvements for the next generation chips?

What is your point, anyway? That a quarter of a million dollar POWER 5+ system is faster than a $50K Xeon server? I'd hope so.... What about a 64-way SPARC system vs a Celeron?

You're looking at server benchmarks on high-end enterprise systems vs. low-end servers, and trying to make some statement about what the next generation of desktop chips will deliver on desktop apps? The next Apple isn't going to have a POWER 5+ chip, nor is it going to have a Netburst Xeon chip. Stop the mental masturbation about these SPEC rate scores.

 

[ksz]

I wonder if Apple could create one motherboard with socketed processors that could be sold with 2Ghz Woodcrest dual core chips, and upgraded by either Apple or the user to 3 Ghz chips, or Cloverton quad core chips. Those would be great choices.

I hope we're witnessing the birth of the Era of Replaceable Processors in desktop Macs and eventually in laptops as well.

All I need now for my dream system is for EyeTV 500 to go MacIntel, and an Apple Cinema Display with an HDMI/DRM connector for next generation DVDs.

It does not seem likely that a single next-gen DVD format will emerge. BluRay and HD-DVD are likely to coexist just as DVD+R and DVD-R. All I hope is for inexpensive (relatively speaking) dual-format drives to be available when the Mac Pros are released.

 

[thunng8]

Do you honestly believe that SPECfp_rate measures floating point performance?

What does it measure then?

It doesn't, and nobody buys an MP Xeon system for what it does measure. Yes, it's obvious to everyone who really understands SPECfp_rate that large numbers of Xeons don't scale. Dell stopped selling 8-way Xeons - there's a reason.

Which is exactly why an 8-way PowerMacIntel will have two quad-core chips with a huge shared cache and a 1333 MHz FSB. Do you think that Intel isn't concerned about this issue, and isn't working on improvements for the next generation chips?

What is your point, anyway? That a quarter of a million dollar POWER 5+ system is faster than a $50K Xeon server? I'd hope so.... What about a 64-way SPARC system vs a Celeron?

What are you talking about?, I've pointed out a $4-5k Power5+ system that beats the 50k Xeon system in specfp_rate. And the p5-575 does not cost a quarter of a million dollars .. it is closer to $100k.

You're looking at server benchmarks on high-end enterprise systems vs. low-end servers, and trying to make some statement about what the next generation of desktop chips will deliver on desktop apps? The next Apple isn't going to have a POWER 5+ chip, nor is it going to have a Netburst Xeon chip. Stop the mental masturbation about these SPEC rate scores.

It was just something I noted in the original post , and then you challenged me on it (firstly by saying an itanium 2 system would be close to Power5+ performance), not once, but twice (asking me to provide more data points for bad specfp scaling) - nothing more. I think this can stop now.

 

[Mechcozmo]

blahblahblahblahblahblahblahblahblahblahblahblah<funny>Stop the mental masturbation</funny>blahblahblahblahblahblahblahblahblahblahblahblah

Read the 'tar.

 

[rhashem]

Hmm...well the G4 was dead on the mobile mac platform and mini...but how will Conroe really compare to the Quad G5? Given that the Pentium M architecture performs at around clock-for-clock the same as a G5 equivalent, we would need two Conroes at at least 2.8Ghz to match or beat the Quad G5..right?

The P-M is considered clock-for-clock equivalent to the G5 because of poor floating-point performance. The Merom design fixes that, so the SPEC scores for a 3 GHz Conroe are 2800/2500, versus 1438/2076 for a 2.5 GHz 970MP. So a 2.5 GHz Conroe would match the 970MP in SPECfp and outdo it by a large margin in SPECint. Considering Woodcrest (the CPU you'll need for a Quad-like machine, because of SMP) will be available up to 3.3 GHz, Apple doesn't even have to ship Intel's fastest parts to outdo the old Quad.

 

[rhashem]

64-bit just for the sake of being 64-bit doesn't make it faster. It can actually be slower.

On x86, this generally isn't true. x64 gets rid of most of the performance limitations Intel and AMD haven't been able to exorcise out of x86 already. Generally, it leads to a 10-20% speedup on a processor that implements x64 natively (eg: the Opteron, and hopefull Conroe).