Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Originally posted by dethl
Any idea when the MPC7457 will be coming out? If they get it out before IBM does...moto may have a chance...but i wouldn't hold my breath.

Very perceptive.
there is a BIG difference between having an integrated DDR controller and simply having support for DDR. None of the x86 processors out have integrated memory controllers.
This article doesn't make any assertions as to whether or not the entire 7457 line will support DDR. I'd *guess* that it would, but only because I can't imagine anyone releasing a general purpose CPU in this day and age that didn't support DDR [note I said "releasing", as in future of present tense].

....ffakr
 
Re: Re: Moto in towers, 970 in servers?

Originally posted by Catfish_Man


The opposite is about 50 times more likely. The XServe is a 1U. It CANNOT use high power consumption/heat dissipation chips. There is a reason why there are no major-brand Pentium 4 based 1Us. The 970 isn't too high consumption, but it's a LOT higher than a .13 micron G4+ will be (42 watts, vs. <20 watts).

a .13 micron G4 should come in under 20watts (depending on L2 cache and speed of course)... the current G4s are under that now (though not by much).

There ARE P4s in 1U machines. I don't have time to check every major manufacturer (Dell doesn't sell them), but a search for "1U P4 Rack Server" will get you tons of hits.
there are tons of K7 1U servers also... and the Athlon puts out about as much heat as a P4 anyway...
 
Hmm ...

1833 Mhz its a good clock for a DDR RAM 266 or 333 good multiplicator for the BUS

or ?
 
Yes, this is just what we need: IBM and Moto competing against one another, just like Intel and AMD compete against each other and that leads to faster chips sooner. Yes, this will make the "MacWorld" so much brighter! This is just what we need. :D
 
Originally posted by e-coli
great...so moto won't actually support true DDR until 2004. By then PC's will have LudicrousSpeedRAM.

Ummm ... an integrated DDR memory controller in the chip is not required for "real" DDR.

Current Macs support "real" DDR, but lack the CPU-to-system controller bandwidth to take advantage of it. One would hope that the 970-competitor chip would have a dramatically improved FSB which would allow CPU-to-SC bandwidth that would take advantage of the high-speed DDR controller in the SC.

The only reason to support DDR on-CPU instead of in the SC chip is that it makes the SC chip that much simpler and should reduce latencies for memory accesses. The drawback to such a design is that you eliminate the possibility of other system components (AGP et al) talking directly to memory without interfering with the CPU-SC bandwidth, and you completely eliminate the possibility of a simple SC upgrade introducing faster memory technology (such as the current DDR update ... the CPU itself didn't change at all to support DDR memory, just the SC and memory controller).

IMHO, I'm a bit sceptical of moving the memory controller on-CPU. I'd have to see some conclusive performance evaluations to agree that it makes sense. You give up quite a bit of flexibility in such an approach, IMHO.
 
for the people that are talking about waiting
till next year and acting like it's a long time.

2003 IS LESS THEN THREE MONTHS AWAY.
 
check out the applestore -- they have fast machines for sale now.

what the heck are you doing that the powermacs can't handle?

i regularly record 8 simultaneous tracks of full-bandwidth audio on my 3 year old dual 500. it flies.

what are YOU doing that's so intensive?

fullbandwidth ?

44.1Khz, 24bit, 48Khz, 24bit or 96Khz 24bit ?

Stereo or mono ?

I can record 32 simulataneous tracks at 48Khz, 24bit on my 5 year old beige G3, it doesn't mean I can use many plug-ins or software synths with it, even an 800MHz G4 is only around 3 times faster for purely audio and plug-ins.

Software synths are another thing entirely, Altivec plays a large part in their performance and cpu overhead but for raw plug-in performance an Athlon 2200XP or a 2Ghz or more Pentium 4 Northwood running on a TRUE DDR system bus is up to 3 times faster comparing single cpu to single cpu performance in protools LE.

Imagine how useless a 1.8Ghz G4 will be in realworld terms.

A 1.7Ghz Pentium 4 with a PC133 SDRAM chipset was actually around 10% slower than a 1 Ghz Pentium 3 with PC133 SDRAM. Sure there was the little matter of anaemic cache sizes, bloated pipelines and a host of other inefficiencies in the name of clockspeed on the Pentium 4 compared with the Pentium 3 but all the same, PC133 SDRAM on a 1Ghz chip isn't as bad as PC133 on a 1.8Ghz chip no matter how efficient the chip might be.

Baring in mind 133Mhz = 133.333 and 167Mhz = 166.666, Here's the cpu on a 1Ghz Quicksliver G4 :

CPU (s) : 1000Mhz
L1 : 32K + 32K 1000Mhz SRAM
L2 : 256K 1000Mhz SRAM
L3 : 2Mb 250Mhz DDR SRAM (1 quarter cpu speed, double pumped to 500Mhz)

Bus Speed : 133Mhz (cpu speed / 7.5)
System RAM : PC133

Now the 1.25Ghz G4

CPU (s) : 1250Mhz
L1 : 32K + 32K 1250Mhz SRAM
L2 : 256K 1250 Mhz SRAM
L3 : 2Mb 312.5 Mhz DDR SRAM (1 quarter cpu speed, double pumped to 625Mhz)
Bus Speed : 167Mhz (cpu speed / 7.5)
System RAM : 333Mhz DDR SDRAM (irrelevant because the cpus don't benefit from the extra bandwidth)

That's an equal bottleneck to a 2Ghz Pentium 4 running on a chipset supporting 266Mhz DDR SDRAM, except the mac is half the clockspeed and has the large L3 cache to act as a buffer to the relatively slow system RAM.

In basic terms for either the 1Ghz QS G4, 1.25Ghz Mirror Doors G4 or 2Ghz Pentium 4 on a DDR chipset, the cpu is being fed 2 15ths of the bandwidth it really needs from main system memory across the system bus.

Now for the 1.833Ghz G4.

1.833Ghz isn't possible on a 133Mhz bus so that would make it a 167Mhz bus like the 1 and 1.25Ghz G4 in the mirrored doors G4.

From the specs in the article and from previous specs of the G4s and apple's system controller here's how it would stack up :

CPU (s) : 1833.3Mhz
L1 : 32K + 32K 1833.3Mhz SRAM
L2 : 512K 1833.3 Mhz SRAM
L3 : 2Mb 458.3 Mhz DDR SRAM (1 quarter cpu speed, double pumped to 916.6 Mhz)
Bus Speed : 167Mhz (cpu speed / 11)
System RAM : 333Mhz DDR SDRAM (irrelevant because the cpus don't benefit from the extra bandwidth)

So a 1.8Ghz G4 will only be able to feel 1 11th of the bandwidth the cpu needs because the cpu still doesn't support a DDR system bus.

The bus speed isn't going to be anywhere near fast enough to feed enough to the cpu for it to perform efficiently unless 333Mhz DDR SDRAM can interface directly with the cpu. It's still going to have the same bottleneck the G4s have had since their introduction, a bottleneck that's really holding the cpu back from achieving it's true potential.
 
Look, all I am saying is that there is some sort of theory, or rule, that processor speeds increase exponentialy. I am sorry I'm not that big a techo-geek I don't know the exact name or premis. Either way Motorola processors have fallen behind this slope and probably will never catch up. Software comes and go's While Winodws XP is a piece of crap today, there is nothing to say that Winodws 2006 won't be brilliant. However, it is guarenteed that by that time Motorola processors will be so far behind Intel/AMD that we'll be sticking them in our refrigartors and alarm clocks.

Now keep in mind I am a Mac user and fan. I feel that their software and hardware DESIGN is second to none. However, there can no longer be any serious argument to Mac hardware being even close to the Intel/AMD world when it comes to everyday PC tasks. I am hopefull that the news of IBM re-entering the game will give us some new bragging rights, but I am still not convinced of the immediate benefits of 64bit processing for everday tasks.

Now please don't go yelling and screaming about Photoshop benchmarks and so forth, I, like most Mac users, use my mac for everday tasks like web browsing, office apps, simple photo editing, mp3, etc. If the Mac has truely become so specialized that it's only advantages are in these high-end pro apps then it is truely a dark day for Apple.

-Stu Koch
 
Originally posted by stukoch
Now please don't go yelling and screaming about Photoshop benchmarks and so forth, I, like most Mac users, use my mac for everday tasks like web browsing, office apps, simple photo editing, mp3, etc. If the Mac has truely become so specialized that it's only advantages are in these high-end pro apps then it is truely a dark day for Apple.

Stu, you just refuted your own point. Most people /do/ use their computers for simple stuff like you mention, and for them all that matters is that there is good software for those things, and that it runs responsively on their computers. Apple has theoretically cornered the market in these areas with the iApps and a little help from the MS MacBU (and the OmniGroup, IMO :)) If the hardware in a Mac is fast enough to handle "high-end pro apps" then it darned well ought to be satisfactory with desktop publishing and the like, don't you think? So if all you care about is whether you can use iTunes and Word and a browser at the same time, what is there to be concerned about?
 
Re: great...

Isn't competition beautiful.......?

If it weren't for IBM, Moto wouldn't budge an inch in processor speeds!!

I hope thry are through getting back at Apple about the clone shutdown.

Then again, perhaps they need the extra CPU income from Apple.

:D
 
Re: Moto in towers, 970 in servers?

Originally posted by markomarko
I hate to mention it, but isn't it a possibility that the 970 will be an xserve-only processor and that the powermacs will be stuck with G4?

No. People would bit– complain too much and cause apple to use it in their pmacs. I know I would.


--

Moto just announced a high tech trash can. You don't even have to use your hands to open it! You just press on a pedal at the bottom of the unit, and the lid will open up – welcoming you to throw away all the crap moto is making. :rolleyes:
 
Originally posted by Shrek
Yes, this is just what we need: IBM and Moto competing against one another, just like Intel and AMD compete against each other and that leads to faster chips sooner. Yes, this will make the "MacWorld" so much brighter! This is just what we need. :D

exactly. I look forward to competition. I hope IBM wins though, because if moto can't make chips unless they have competition then they shouldn't be making chips at all.
 
Originally posted by jg3


Stu, you just refuted your own point. Most people /do/ use their computers for simple stuff like you mention, and for them all that matters is that there is good software for those things, and that it runs responsively on their computers. Apple has theoretically cornered the market in these areas with the iApps and a little help from the MS MacBU (and the OmniGroup, IMO :)) If the hardware in a Mac is fast enough to handle "high-end pro apps" then it darned well ought to be satisfactory with desktop publishing and the like, don't you think? So if all you care about is whether you can use iTunes and Word and a browser at the same time, what is there to be concerned about?

Perhaps you haven't done any of these tasks on a modern PC lately but the speed differences are very noticable. For example, I am sitting next to my OLD 1.2ghz AMD right now and I can tell you that it loads this forum's page in less then half a second, on my NEW 800hmz iMac this page takes exatcly 2.5 seconds. While this may not seem like a big difference it is very noticable when you are used to zipping around the internet. Also, maybe you have not use iPhoto or iTunes but if you have you would know that they are insanley slow. Just to open up iPhoto on my machine takes over 30 seconds, and don't get me started on ripping CDs in iTunes. The indisuptable fact is that the Mac OS would be an entirly different and better expereince on machines with greater processing power, and Motorola is not even close to fulfiing that need.

-Stu
 
Originally posted by barkmonster


fullbandwidth ?

44.1Khz, 24bit, 48Khz, 24bit or 96Khz 24bit ?

Stereo or mono ?


i've got a digi001 and protools LE. i typically record 16-bit, 44.1k, mono. w/ the digi, i'm limited to 8 tracks of simultaneous recording. and w/ PTLE, i think it's 24 tracks of playback to which i'm limited.

my point being: i have use of a pmac, and my "ancient and slow" machine handles it just fine. you echoed my point w/ what your g3 can handle.

so my question to macrules929 is: what are YOU doing? i suspect he's mostly writing email and posting to macrumors. NOT doing digital audio, NOT running an avid system.

if he is, then i'll ****.

what percentage of mac users actually could max out a dual gig machine? it's pretty small, i'd guess. mostly those doing big renders.

i get tired of people screaming about how slow the macs are, when they're really not even using the machines.

i rarely max out mine. okay, i did the other day when i was simultaneously importing 14,000 emails into Mail.app and having iTunes volume smooth 2800 songs. both cpus were maxed for a while. it was neat to see how efficiently osx distributed the workload (i was watching w/ top).
 
Originally posted by barkmonster

A 1.7Ghz Pentium 4 with a PC133 SDRAM chipset was actually around 10% slower than a 1 Ghz Pentium 3 with PC133 SDRAM. Sure there was the little matter of anaemic cache sizes, bloated pipelines and a host of other inefficiencies in the name of clockspeed on the Pentium 4 compared with the Pentium 3 but all the same, PC133 SDRAM on a 1Ghz chip isn't as bad as PC133 on a 1.8Ghz chip no matter how efficient the chip might be.

Reference, please. My understanding that the P4/P3 difference was on similar clock speeds, not 1.7x the speed of P4 being slower than the P3. Please provide a reference for the 10% speed difference not being based on equal-clocked CPUs.


That's an equal bottleneck to a 2Ghz Pentium 4 running on a chipset supporting 266Mhz DDR SDRAM, except the mac is half the clockspeed and has the large L3 cache to act as a buffer to the relatively slow system RAM.

In basic terms for either the 1Ghz QS G4, 1.25Ghz Mirror Doors G4 or 2Ghz Pentium 4 on a DDR chipset, the cpu is being fed 2 15ths of the bandwidth it really needs from main system memory across the system bus.

Umm, no. A 2GHz P4 has a quad-pumped 100MHz FSB. That's 400MHz FSB, (the P4-2.8GHz has a 533MHz FSB, but the 2.0GHz did not). The system bus on a P4 is 8 bytes (64 bits), which gives the 2.0GHz P4 an effective input bandwidth (not counting a small overhead) of 3.1 GB/s. The 2.8GHz P4 has a bandwidth of 4.2 GB/s (ref: http://www.intel.com/design/Pentium4/prodbref/index.htm?iid=ipp_dlc_procp4p+prod_brief&#bus )

Given that DDR memory is much less bandwidth than 3.1GB/s, much less 4.2 GB/s, the P4 is hardly bandwidth-starved.


Now for the 1.833Ghz G4.

1.833Ghz isn't possible on a 133Mhz bus so that would make it a 167Mhz bus like the 1 and 1.25Ghz G4 in the mirrored doors G4.

From the specs in the article and from previous specs of the G4s and apple's system controller here's how it would stack up :

CPU (s) : 1833.3Mhz
L1 : 32K + 32K 1833.3Mhz SRAM
L2 : 512K 1833.3 Mhz SRAM
L3 : 2Mb 458.3 Mhz DDR SRAM (1 quarter cpu speed, double pumped to 916.6 Mhz)
Bus Speed : 167Mhz (cpu speed / 11)
System RAM : 333Mhz DDR SDRAM (irrelevant because the cpus don't benefit from the extra bandwidth)

So a 1.8Ghz G4 will only be able to feel 1 11th of the bandwidth the cpu needs because the cpu still doesn't support a DDR system bus.

Unless the CPU-to-SC bus is (a) Wider or (b) double- or quad-pumped as in the P4.


The bus speed isn't going to be anywhere near fast enough to feed enough to the cpu for it to perform efficiently unless 333Mhz DDR SDRAM can interface directly with the cpu.

Wrong. Neither the P4 nor the AMD Athlon embed the DDR memory controller in the CPU. They super-pump the FSB, then let the north bridge of the chipset determine what kind of memory is going to be supported. The circuitry to quad-pump the FSB is both simpler and takes up less die space than the DDR memory controller.

Again, the ONLY thing one would gain by putting the DDR controller on the CPU, aside from simplifying the SC, is that you remove a latency from memory accesses. This is a valid goal (typical PC latencies are 2 or 3 cycles per memory access), but the equally troubling tradeoff is that you lose flexibility in design and a massive amount of CPU die space. The smaller the CPU, the fewer rejects you will get in manufacturing, and the faster you can ramp up to higher-frequency-stable production.

I haven't seen anything stating how the FSB on the upcoming Motorola chip is implemented. Unless we know for a "fact" that Motorola is keeping its old single-pumped, 52-bit width FSB for this new chip, I'm inclined to give Apple the benefit of a little bit of foresight and note that the DDR support next to a slow FSB is something that even they would have thought of. (Note on bit-width of G4 FSB: I'm not really sure about this. Moto claims 640 MB/s bandwidth peak and sustained on a 100MHz bus, which gives 51.2 bits, and 851 MB/s on a 133MHz bus, which gives 51.18 bits ... but 52 bits seems an odd number; maybe someone else can enlighten me :) ). I mean, the G4 only recently (166 MHz FSB) gained the FSB bandwidth that the P3 enjoyed several years ago (~1GB/s)! I hope that Motorola has noticed this!

Since I've mentioned the P3, note that Intel did something very similar, with a higher memory bandwidth than the processor's FSB could support. The P3 started supporting RDRAM long before it had the bandwidth to take advantage of the more advanced memory. This is a chief reason why RDRAM gained such a crappy reputation early on (people were paying an extra $1000 per PC just for RDRAM and seeing absolutely no performance improvement, then Intel went to RDRAM=only and people were more than a bit ticked). Intel at the time knew that the P4 would have both the FSB and the underlying design to take advantage of the massive bandwidth, and by the time the P4 debut'd RDRAM prices were at least in the acceptible range.

Of course, Apple isn't pushing a whole new memory technology (thankfully!), and I don't see a "master plan" behind introducing DDR prior to a chip that can take advantage of it, which leads me to suspect that the timing of the two releases was simply off.
 
Umm, no. A 2GHz P4 has a quad-pumped 100MHz FSB. That's 400MHz FSB, (the P4-2.8GHz has a 533MHz FSB, but the 2.0GHz did not).

I said on a DDR chipset, nothing about the quad pumped FSB on the cpu itself. A 2Ghz Pentium 4 with a 4 x 100Mhz FSB has the same bottleneck as a 2Ghz Pentium 4 Northwood with a 4 x 133Mhz FSB if they're both using a chipset that supports 266Mhz DDR SDRAM.

Crap diagram below :

|---------| ...... |------------|
| 266Mhz .|------->| Pentium 4 .|
| DDR ....| ...... | with ......|
| Chipset |<-------| 100Mhz FSB |
|---------| ...... | x 4 (quad} |
.................. |------------|


This for the Willamette 423 and Prescott 478 versions of the Pentium 4 available in speeds of 1.3 - 2 Ghz. All supporting 256K of L2 and a 100Mhz x 4 FSB. It also applies to the Northwood Pentium 4 with 512K on L2 and a 100Mhz x 4 FSB available in speeds of 2, 2.2, 2.4, 2.5 and 2.6Ghz.

|---------| ...... |------------|
| 266Mhz .|------->| Pentium 4 .|
| DDR ....| ...... | with ......|
| Chipset |<-------| 133Mhz FSB |
|---------| ...... | x 4 (quad} |
.................. |------------|


This for the updated Northwood Pentium 4s with a 512K L2 and a 133Mhz x 4 FSB. available in speeds of 2.26, 2.4, 2.53, 2.66, 2.8 and soon to be 3.0Ghz

Either way, an effective bus speed of 400 or 533Mhz on a chipset with 266 or 333Mhz DDR is still not getting the full 400 or 533Mhz of bandwidth the cpu demands.

I wasn't implying the cpu should have a DDR controller onboard or anything like that, I was meaning supporting double or quad pumping of the existing FSB speed in the new PowerPC instead of only supporting 133 or 167Mhz. I agree with you're points on double and quad pumping the bus and I realise the pentium 4 still has a 133Mhz bus when it comes down to the numbers but I don't see how a G4 with a 133Mhz or 167Mhz bus in the 1.8Ghz could possibly cope with the bandwidth needs unless some kind of 2 or 4 x FSB was supported.
 
but...

The athlon xp only has a 133mhz fsb and it runs faster with 266ddr than it does with 133 sdram. how come the g4 doesnt increase its speed like this? cause the chip isnt designed for ddr. but the 7457 is. this chip placed into the current apple mobo set up will have more power, clock for clock.
 
Re: but...

Originally posted by Bradcoe
The athlon xp only has a 133mhz fsb and it runs faster with 266ddr than it does with 133 sdram. how come the g4 doesnt increase its speed like this? cause the chip isnt designed for ddr. but the 7457 is. this chip placed into the current apple mobo set up will have more power, clock for clock.

The AlthonXP has a Double Data Rate 133Mhz FSB, the G4 has a single data rate 167Mhz FSB.

167Mhz SDR* 64bit = 1.3GB/s
133Mhz DDR* 64Bit = 2.1GB/s

And the 2700+ and 2800+ have a 166Mhz DDR bus, or 2.7GB/s

IOW, the G4 bottlenecks on it's frontside bus.
 
Re: Re: but...

Originally posted by Chryx


The AlthonXP has a Double Data Rate 133Mhz FSB, the G4 has a single data rate 167Mhz FSB.

167Mhz SDR* 64bit = 1.3GB/s
133Mhz DDR* 64Bit = 2.1GB/s

And the 2700+ and 2800+ have a 166Mhz DDR bus, or 2.7GB/s

IOW, the G4 bottlenecks on it's frontside bus.

True, but to worsen the difference: the FSB on the G4 is 52 bits, not 64. Not sure why. But product literature from Motorola claims 640MB/s on a 100MHz bus, 800 on a 133MHz bus, and by extension (unless the FSB width changed with the 167MHz bus), 1.085GB/s.

So, revised figures:

Mac FSB = 1.085GB/s
AMD FSB = 2.1GB/s
P4 FSB = (133MHz * 4 * 8B/cycle (64bit) = ) 4.256 GB/s
IBM PPC 970 FSB = (900MHz * 8B/cycle = ) 7.2 GB/s

All figures using GB=1,000,000,000 bytes, so will be slightly off from "official" numbers using GB=1,073,741,824 bytes.

The P4 and AMD buses are wide enough to take advantage of the additional memory bandwidth given by DDR. The Mac bus is filled even with slow SDR RAM.

Note also that AMD is reported to be ready to transition to a 333MHz (167MHz*double-pumped) FSB. That would change the above numbers to:

AMD FSB3 = (167MHz * 2 * 8B/cycle = ) 2.672 GB/s

While AMD and Intel chips have adequate FSB bandwidth for any current memory systems (DDR-3 might cause further problems for all involved, however), the G4 is fundamentally memory starved and that starvation can not be fixed on the cheap by upgrading the SC to DDR memory. The FSB has to be widened or multiply-pumped to get more memory to the chip. The IBM PPC 970 stands as potential salvation at the end of a very long and dark tunnel, but that's a year away.

Side note: THe P4 is also memory starved, although not nearly to the degree the G4 is. When the P4 2.4GHz went from a 400MHz FSB to a 533 MHz FSB, benchmark numbers increased measurably. That means a core/FSB ratio of 6 was too high (and even 4.5 might have been too high, but we can't know that outside of Intel's labs), leaving the FSB as a bottleneck in at least some situations. The 533 MHz FSB will hit a ratio of 6 when the core is 3.2GHz. In other words, early Spring.
 
Re: Re: but...

Originally posted by Chryx

And the 2700+ and 2800+ have a 166Mhz DDR bus, or 2.7GB/s

Oops, I missed that line. So AMD has moved to the 333 FSB? I hadn't heard that. Thought it was still a future development ...
 
Re: Re: Re: but...

Originally posted by jettredmont

True, but to worsen the difference: the FSB on the G4 is 52 bits, not 64. Not sure why. But product literature from Motorola claims 640MB/s on a 100MHz bus, 800 on a 133MHz bus, and by extension (unless the FSB width changed with the 167MHz bus), 1.085GB/s.


Could be that they were taking into account any behind the scenes overhead the bus has?
 
Re: Re: Re: but...

Originally posted by jettredmont


Oops, I missed that line. So AMD has moved to the 333 FSB? I hadn't heard that. Thought it was still a future development ...

2800+ on Nforce2

They aren't a "future" development, but they haven't hit the market yet.

They aren't a rumour of something AMD might do anymore though.
 
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.