256MB cards are faster at high resolutions above 1600X1200 "generally" and allow more fsaa and artoscoping basically higher quality graphics, all fx on and quality settings at high or super at those higher resolution, at lower res its basically a wash under 1600x1200 as far as pc's are concerned and you almost say can say the same thing for 16 vs 32-32vs64-64vs128. the more memory the higher the resolution more fx, faster at high resolution generally although the extra overhead can slow it down at some resolutions its just a matter of time before games list 64MB as a min and then 128. seeing as a larger portion of macs-imacs/emacs unupgradable) or limited to 32MB and still sale well, that will be a ways off still for most games
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
PowerMac 1.6GHz Benchmarks
- Thread starter MacRumors
- Start date
- Sort by reaction score
Dave, might I suggest using punctuation.
Tests show that even with 256MB video cards, performance is pretty weak (less than 50 fps) when you start going above the resolution range where 256MB cards move ahead of 128MB cards (more than about 1600x1200 at 4xFSAA). Any resolution that is reasonably playable with 256MB gets the same frame rates on 128MB cards. Which means effectively a 128MB card is just as good as a 256MB version of the same card. Even for future games.
The reason for this is that the GPUs currently can't handle such high resolutions anyway no matter how much memory you throw at them--that is, the GPU is the limiting factor at those resolutions. So the extra memory is overengineered (for marketing purposes, really) and useless. When a faster GPU comes out, you'll have to get a new card to use it anyway--and that new card may well be able to actually effectively use the larger memory it comes with.
one last time...for the slow learner spewing propaganda as fact...
1. DDR RAM is NOT significantly faster than SDR RAM when compared to like bus speeds. DDR was originally used as a cost saving measure by hardware manufacturers. DDR RAM requires 1/2 the logic board traces from the memory controller hub to achieve a slightly higher level of performance.
DDR266 (PC2100) yields ONLY between 3-5% increase in performance over PC133 SDRAM.
Where the G5 will gain performance is through doubling the bandwidth through using a "Dual Channel" memory archetecture allowing memory to be strobed between two independant banks (similar to the RAMBUS Archetecture...but, unlike RAMBUS the machine will be functional with only one stick of memory not requiring a matched stick or continuity PCB placed in the opposite bank).
...and it is not the case of a 2" pipe vs. a 1" pipe...it's a 1" pipe vs. a 1"pipe insofar as bandwidth is concerned.
2. AGP was created to allow Palletizing Textures greater than 8MB to system memory, not because it allowed better performance but to keep the costs of Video Cards reasonable (Video Memory is ALWAYS faster than System Memory when considering the GPU's work)
Although theoretically higher bandwidth creates greater performance (which actually is the truly case between AGP and AGP2x and not so much between 4x and 8x) the greatest difference between AGP 1/2, 4 and 8 is Voltage. Each stepping of the AGP bus drops voltage levels (AGP 1/2 is 5v...AGP 4 is 3.3v...and AGP 8x is 1.6v) to allow for PC2000 Compliance and smaller process sizes that are energy saving and offer more features and performance for the same size piece of silicon.
Increases seen in graphics performance is through better "Northbridge" Archetectures...and through higher performing GPU's with greater amounts of onboard memory. The more memory a graphics card has the less dependant it is on the AGP bus (and system memory) for palletizing textures.
3. SATA's performance is not significantly higher than ATA100...typically even the fastest drives only send and receive data at a level slightly higher than ATA66. The technology behind IDE/ATA Hard Disks has not come close to catching up with that of the controllers (here's a hint...if you want HDD speed...fault tolerance...and reliability...the advice has been the same for many years...go SCSI).
The reason SATA has an advantage, even the most advanced ATA133 controllers are blind and stupid. ATA Controllers can send/receive only to one drive at a time per channel and must continually check for parity to make sure the data was handled correctly. Regardless whether that chain has one or two drives transactions are handled in the same manner. SATA has the benefit of having only one drive permitted per chain (and takes some of the guesswork out of communications) and making a slightly smarter drive (although not quite to SCSI standards...and Ultra320 isn't anywhere near 320...but it's faster than SATA...and can have 14+1 devices/channel). Did I forget to mention the reason there's even IDE/ATA on an Apple is because it's cheaper???
...as a friend mused: SATA150 drives will probably reach 150 right about the time SATA2 is released...he's probably right...
4. In your own quote:
...look back at your own link...find the 8bit ISA card...the connector is about the same length as a PCI slot...and no...the Enhanced Industry Standard Archetecture (EISA) was a flop (like VESA Local Bus)...but there were cards produced for the general market mostly HDD Controllers and Video Cards...and EISA (unlike MCA) was "Compatible" with existing 8 and 16bit ISA which is why it found it's way onto every x86 motherboard from 80386SX to PentiumIII's before obsoleted all together (not completely though...there's an Athlon board that was produced last year with one oddly enough)
AND I'll probaly get some hate and dscontent over this one...BUT...the G5 won't be and is not the fastest...it'll be the fastest Macintosh once they get optimized code out for it...and if it's reliable and stable probably the best Macintosh ever produced...but I don't want one...
I've never bought a Macintosh for speed...my Laptop outperforms this Dual 1.25...and my boxed up AthlonXP 3000+ runs circles around the laptop...
...if I'd wanted speed...I'd have bought a Dual Opteron with SuSE8.1 (64bit).
...if I'd wanted to do graphics...I'd have bought from SGI.
...if I'd only wanted only stability...I'd still be using OS/2.
...BUT...I bought a Macintosh...I wanted another Mac...I like my Mac...buying a computer is not a rational decision...it's a personal one...you like the G5...power to 'ya...I think it's Industrial-Strength-Ugly...doesn't allow for the type of expansion I'm looking for...and is badly overpriced for what you get...but that's my opinion...and it's my money to spend...I chose to buy a G4 to replace my G3 that recently died (at a hair over 4 years old...not too bad...and I guess the next one I buy...if Apple is still around will probably be a G7 or whatever they'll call the one after the next when I have want for one to replace this).
...although I hope I'm wrong...I sincerely hope I'm wrong...but Apple may have just launched the iceberg that could sink it...if they're not careful...and don't attract the customers they expected with the "Switch" campaign...
...lastly...Googling is fun...but if you want hard facts...work in the industry...read the tech journals...read a book...
1. DDR RAM is NOT significantly faster than SDR RAM when compared to like bus speeds. DDR was originally used as a cost saving measure by hardware manufacturers. DDR RAM requires 1/2 the logic board traces from the memory controller hub to achieve a slightly higher level of performance.
DDR266 (PC2100) yields ONLY between 3-5% increase in performance over PC133 SDRAM.
Where the G5 will gain performance is through doubling the bandwidth through using a "Dual Channel" memory archetecture allowing memory to be strobed between two independant banks (similar to the RAMBUS Archetecture...but, unlike RAMBUS the machine will be functional with only one stick of memory not requiring a matched stick or continuity PCB placed in the opposite bank).
...and it is not the case of a 2" pipe vs. a 1" pipe...it's a 1" pipe vs. a 1"pipe insofar as bandwidth is concerned.
2. AGP was created to allow Palletizing Textures greater than 8MB to system memory, not because it allowed better performance but to keep the costs of Video Cards reasonable (Video Memory is ALWAYS faster than System Memory when considering the GPU's work)
Although theoretically higher bandwidth creates greater performance (which actually is the truly case between AGP and AGP2x and not so much between 4x and 8x) the greatest difference between AGP 1/2, 4 and 8 is Voltage. Each stepping of the AGP bus drops voltage levels (AGP 1/2 is 5v...AGP 4 is 3.3v...and AGP 8x is 1.6v) to allow for PC2000 Compliance and smaller process sizes that are energy saving and offer more features and performance for the same size piece of silicon.
Increases seen in graphics performance is through better "Northbridge" Archetectures...and through higher performing GPU's with greater amounts of onboard memory. The more memory a graphics card has the less dependant it is on the AGP bus (and system memory) for palletizing textures.
3. SATA's performance is not significantly higher than ATA100...typically even the fastest drives only send and receive data at a level slightly higher than ATA66. The technology behind IDE/ATA Hard Disks has not come close to catching up with that of the controllers (here's a hint...if you want HDD speed...fault tolerance...and reliability...the advice has been the same for many years...go SCSI).
The reason SATA has an advantage, even the most advanced ATA133 controllers are blind and stupid. ATA Controllers can send/receive only to one drive at a time per channel and must continually check for parity to make sure the data was handled correctly. Regardless whether that chain has one or two drives transactions are handled in the same manner. SATA has the benefit of having only one drive permitted per chain (and takes some of the guesswork out of communications) and making a slightly smarter drive (although not quite to SCSI standards...and Ultra320 isn't anywhere near 320...but it's faster than SATA...and can have 14+1 devices/channel). Did I forget to mention the reason there's even IDE/ATA on an Apple is because it's cheaper???
...as a friend mused: SATA150 drives will probably reach 150 right about the time SATA2 is released...he's probably right...
4. In your own quote:
...look back at your own link...find the 8bit ISA card...the connector is about the same length as a PCI slot...and no...the Enhanced Industry Standard Archetecture (EISA) was a flop (like VESA Local Bus)...but there were cards produced for the general market mostly HDD Controllers and Video Cards...and EISA (unlike MCA) was "Compatible" with existing 8 and 16bit ISA which is why it found it's way onto every x86 motherboard from 80386SX to PentiumIII's before obsoleted all together (not completely though...there's an Athlon board that was produced last year with one oddly enough)
AND I'll probaly get some hate and dscontent over this one...BUT...the G5 won't be and is not the fastest...it'll be the fastest Macintosh once they get optimized code out for it...and if it's reliable and stable probably the best Macintosh ever produced...but I don't want one...
I've never bought a Macintosh for speed...my Laptop outperforms this Dual 1.25...and my boxed up AthlonXP 3000+ runs circles around the laptop...
...if I'd wanted speed...I'd have bought a Dual Opteron with SuSE8.1 (64bit).
...if I'd wanted to do graphics...I'd have bought from SGI.
...if I'd only wanted only stability...I'd still be using OS/2.
...BUT...I bought a Macintosh...I wanted another Mac...I like my Mac...buying a computer is not a rational decision...it's a personal one...you like the G5...power to 'ya...I think it's Industrial-Strength-Ugly...doesn't allow for the type of expansion I'm looking for...and is badly overpriced for what you get...but that's my opinion...and it's my money to spend...I chose to buy a G4 to replace my G3 that recently died (at a hair over 4 years old...not too bad...and I guess the next one I buy...if Apple is still around will probably be a G7 or whatever they'll call the one after the next when I have want for one to replace this).
...although I hope I'm wrong...I sincerely hope I'm wrong...but Apple may have just launched the iceberg that could sink it...if they're not careful...and don't attract the customers they expected with the "Switch" campaign...
...lastly...Googling is fun...but if you want hard facts...work in the industry...read the tech journals...read a book...
Do you read the responses to your posts? How many times has the fallacy of this been pointed out to you?
It's not all that bad. Certainly not as god-awful bad as the G4's were just a couple months ago. Yes, it's probably overpriced for the performance, but that is unfortunately how it is going to be: minus=price, plus=OSX. I'd like to see them drop prices and increase supply, but I'm not holding my breath for it.
unfortunately...many people buy into the propaganda and myths of performance...your so called performance increases are mostly due to improvements in chipset and processor...
Sorry for the bad penmanship.Originally posted by soggywulf
Dave, might I suggest using punctuation.
The reason for this is that the GPUs currently can't handle such high resolutions anyway no matter how much memory you throw at them--that is, the GPU is the limiting factor at those resolutions. So the extra memory is overengineered (for marketing purposes, really) and useless. When a faster GPU comes out, you'll have to get a new card to use it anyway--and that new card may well be able to actually effectively use the larger memory it comes with.
I checked a coulple of sites months ago on the 256MB vs 128MB question, and just about all of the sites said that currently the main benefit other then bragging rights is:
1. better quality graphics:
meaning all fx on high 6xfsaa etc:
better frame rates but not by much at higher res.
2. the main difference you will see is when you turn everything up to super high quality, 6xfsaa etc 1600x1200 or higher;
3.Of course it depends on the game involved but the difference can be dramatic, but often is not.
4. AGP8x has only been shown to give up to a 3% increase in frames per second sometimes slower with double the bandwidth but in the future it will come into its own.
5.SATA 150 drives offer very little speed benefit then ata100/133, indeed they are sometimes slower, but more expensive. Theoretically, they offer 17MBS more bandwidth, but in test on many pc sites, sometimes in practice they give less bandwidth, because it is still in its infancy. But the advantages of smaller cables and less cpu use and no master-slave are worth it to some people.
I wish thier was a slot that you could add memory to later, but then who would buy new cards?
htmlcheck this links for more info
http://www.anandtech.com/video/showdoc.html?i=1821&p=22
http://hothardware.com/hh_files/S&V/radeon_9800_256mb(2).shtml
http://www6.tomshardware.com/graphic/20030604/index.
John Q Public
If you want something that is fast, stable, and Apple (partly), then look no further than here. Considering even SGI is moving away from Irix to SuSE, why not Yellow Dog?
Tidbit of the day: Yellow Dog is a term coined back in the days of the staunch and loyal Democratic states of the south, such as if the Democratic Party put forth a yellow dog as their candidate, it would win the state over the republican.
Hence, vote Linux.
Jaedreth
If you want something that is fast, stable, and Apple (partly), then look no further than here. Considering even SGI is moving away from Irix to SuSE, why not Yellow Dog?
Tidbit of the day: Yellow Dog is a term coined back in the days of the staunch and loyal Democratic states of the south, such as if the Democratic Party put forth a yellow dog as their candidate, it would win the state over the republican.
Hence, vote Linux.
Jaedreth
Re: John Q Public
...But I like SuSE...choice of distro is as personal as choosing a 'puter or car...
...guess you learn something new everyday...thanx for the trivia...
...But I like SuSE...choice of distro is as personal as choosing a 'puter or car...
...guess you learn something new everyday...thanx for the trivia...
Daveg5
Heya, isn't it sad in this day and time you're talking about *graphics cards* coming with 128 vs 256 MB RAM, when Apple still ships *computers* with a base of 128 or 256 MB of RAM?
Apple's use of 32 and 64 MB graphic cards is the laughing stock of the computer industry, even though it's a lot better than what we had a short time ago: 8 and 16 MB.
Apple needs to get out of the Dark Ages.
On so many things, Apple is first, or at least claims to be.
You want a portable game machine? Easy. Apple, ship a high end laptop with a built in 256 MB graphic card and 1 GB RAM standard. Don't make it a bto or cto.
Alas, my please of sanity fall upon deaf ears, don't they?
Apple is too entrenched in their corporate image to create truly kick ass machines.
Jaedreth
Heya, isn't it sad in this day and time you're talking about *graphics cards* coming with 128 vs 256 MB RAM, when Apple still ships *computers* with a base of 128 or 256 MB of RAM?
Apple's use of 32 and 64 MB graphic cards is the laughing stock of the computer industry, even though it's a lot better than what we had a short time ago: 8 and 16 MB.
Apple needs to get out of the Dark Ages.
On so many things, Apple is first, or at least claims to be.
You want a portable game machine? Easy. Apple, ship a high end laptop with a built in 256 MB graphic card and 1 GB RAM standard. Don't make it a bto or cto.
Alas, my please of sanity fall upon deaf ears, don't they?
Apple is too entrenched in their corporate image to create truly kick ass machines.
Jaedreth
Untrue.
Untrue.
True.
Possible, but I doubt it. I suspect that the firmware won't allow boot with an odd number of DIMMs installed.
Untrue.
Bullspit. (Ever watched How to Lose a Guy in Ten Days on the airplane? It's pretty weird with all the dubbing...) Anyway, did you even notice when Apple announced they'd gotten over 100,000 preorders on the (highly profitable I might add) PMG5? And they've undoubtedly continued to sell G4's and will sell even more G5's once they hit store shelves.
As for the first part of your post, you continue to spout untruths without citing any sources whatsoever. You linked to a Barefeats article that is completely irrelevant to the current discussion, yes, but other than that we're supposed to take your word for it--when you've already demonstrated in this thread your tendency to insist that you're right when you're clearly not (see: G5 architecture discussion).
You also accuse those of us who are calling you on your misinformation of "spewing propaganda". I don't see how saying that 266 > 133, 333 > 167, and 400 > 200 is propaganda. If we're missing something, please explain. But judging by your past record, I doubt you will.
I'm losing patience, and I apologize for that.
WM
I understand what you are saying. My point is this. Even with 256MB cards, the super-hi-quality settings you mention above lead to unplayably low framerates (according to the tests I have seen comparing 128MB vs 256MB cards at tomshardware etc). Sure the frame rate will be higher than the 128MB cards at these settings...but it's still too low, because the GPU can't keep up even if the card had 700 terabytes of memory. So the question becomes "what's the point of the extra memory".
Basically, what I am trying to say is that Apple made a wise, non-marketing-driven decision to get only the 128MB version of the ATI 9800 graphics card. The 256MB version is $100 more on the PC, and gives no effective benefit at playable framerates. (Aside--kudos to ATI for pricing the retail Mac version of the 9800 at the same point as the PC version! Woohoo!)
It's actually difficult to find meaningful DDR vs. SDR benchmarks. You CAN'T use the Mac as a valid source to pull out a 3-5% difference since the Processor is the limiting factor (as has been mentioned a billion times before)
When you look to the PC, you find that direct comparisons between SDR and DDR chipsets generally compare the KT133A vs. the KT266. Unfortunately, the KT266 was immature when released and the KT266A performed much better. Also, the KT133 benchmarks that I've seen have been performed with rather slow CPUs (typically around 1GHz). The faster the CPU, the more the memory speed would be a factor.
What I've found though:
amd 760/PC2100/1GHz
655 CachMem memory write score
KT 133a/PC133/1GHz
338 CachMem memory write score
(KT266 was 479)
amd 760/PC2100/1GHz
1084 CachMem memory bandwidth
KT 133a/PC133/1GHz
960 CachMem memory bandwidth
(KT266 1027)
CachMem Memory latency
AMD 760/PC2100/1GHz
184
KT133a/pc133/1GHz
210
(lower is better)
In some cases, the DDR machine had much better performance, in some cases they did have better performance, but it wasn't very large (under 10%). In a few application benchmarks I quickly found, the speed difference (all on a 1GHz CPU) wasn't significant, BUT...
As I've said, it isn't easy to compare SDR and DDR since they coincide with a rapid change in CPU speeds, so the SDR benchmarks done with robust SDR chipsets are compared to first gen DDR chipsets. It's difficult to compare a 2nd gen SDR chipset against a 2nd gen DDR chipset with a fast processor because people stopped benchmarking SDR chipsets by the time the DDR/fast CPUs were available (and most SDR chipsets on Athlon weren't available with support for chips over 1.4GHz).
.... looking at P4 now...
Ah, The P4s are a much better comparison.
They clock higher and the transition from SDR to DDR was derailed by a run with RDRAM so there are more recent SDR-DDR comparisons.
Check this out...
Tom's Hardware compares SDR and DDR P4 chipsets
Doh! most of the application benchmarks are generally 10-20% faster and the synthethic Memory sysoft Sandra benchmarks are nearly double for DDR.
As with any benchmark, some tests are not very memory dependent and DDR does have minor performance gains over SDR.. but some apps (like encoding streaming data) tend to benefit more. 20%, 30%... or more.
I'm not aware of any 'strobing' between banks. You make it sound as if there are two 400MHz memory channels that are used alternately. Can you provide back up for this?
Apple is marketing this as a 128bit memory implementation. I wouldn't describe this a 'strobing' between two 64bit memory banks.
I've NEVER seen any mention that the memory in the G5 can run with one bank. That's what it looks like you are saying. I've heard Apple specifically state that memory must be installed in pairs though.
I don't see how you've refuted the analogy
*** got to run so I'll skip some***
HUH??
OK, look at YOUR quote...
You say 'No, EISA...was a flop' (which is my rebuttal to you), then you say that EISA was on every motherboard from 386 to Pentium.
I think you are confusing the architectures.
Yes, there is an 8bit ISA that is PCI sized. There is ALSO a 16 bit ISA card which has the same tab, but two of them.
EISA has a different tab/slot. It is compatible, but it is not the same as the 16bit ISA slot you are thinking of. Look back at the link I posted earlier. It shows 8 bit ISA, 16 bit ISA, and a very different looking EISA card.
Long ISA slots were put on all PC motherboard until very very recently, but NOT EISA!
I worked on a dual 2GHz at WWDC, running on a beta OS and it is the fastest Mac, right now. It will only get faster as the OS and Applications get revised. Don't use early benchmarks on 1.6GHz G5s to claim that it isn't currently the fastest Mac (the implication being the dual 1.42s are faster.. I think that's what you are saying).
and once again... why is it at all revelevent to this thread that you don't want a G5. Who cares? Don't buy one! There are consumer lines that are a lot cheaper and more suited to your needs.
ha, ha. That's rich. I run a department responsible for the support of thousands of PCs. I do work in the industry. I do read books.
Googling is fun, but it's also a way to BACK UP your arguments instead of making off the wall claims without any support.
I'll assume I'm one of these "slow learners" you are referring to and will take a slight offense to the parts of your posts that allude to mine. (I'll continue to kindly gloss over the historical footnote that is EISA because I fail to see the relevance of 486-era standards to a discussion today.) I suppose I shouldn't have bitten on the flame bait, and should have held back when you tried to imply that Apple discontinued the FW800 in favor of the older MDD as some great conspiracy to avoid a price/performance embarrassment. Oh well, sometimes I can't help myself
Please cite your source. So far you've shown only one example (G4 PowerMacs) and I, as well as many others, pointed out the error using that report. In fact, if you read the article you cited, the conclusion of that article pretty much sums up our argument: the DDR RAM is much faster, but when you attach it to a SDR shared CPU bus, there is no benefit.
No, you do not get double the performance because of latency issues (name one area that hasn't been the case?), but you do get double the bandwidth which leads to a significant gain in performance. Nearly every PC benchmark bears this out.
I'm going to ignore the RAMBUS analogy since it is irrelevant. I will point out however, that there is no evidence that you can put unmatched or unpaired sticks in a G5 like you can in dual-channel PCs. I would imagine Apple would strongly discourage such a thing perhaps to the point of having the computer refuse to boot because first guess would be that a 2x2Ghz G5 with independant 1Ghz FSBs would certainly saturate a 800Mhz (effective) RAM connection, let alone a 400Mhz (effective). If you have proof otherwise, I'd like to see the cite and clear that uncertainty up in my mind.
There are many benefits to SATA beyond performance and independant channels which you have conveniently ignored (less voltage, smaller cable for airflow, integrated power, hot swappability, lower voltage, no interference issues, jumperless). Also, while I've pointed out multiple times in MacRumors forums that no hard drives (even your much vaunted SCSI drives) come close to saturating a ATA133 in regular disk throughput, there are two things glossed over: 1) The RAM buffer is still bandwidth limited (a minor issue, but it does show on the proper configurations SATA150 can burst up to 150 right now), and 2) hard drive manufactures have no plans of ramping ATA drives beyond 7200RPM while you can buy a 10K RPM SATA drive (WD Raptor) right now with other companies soon to follow. The 10K RPM SATA, while priced much higher than a regular IDE, is still well below that of their or any competitors 10K SCSI.
Yes, it is true there is no reason why you can't have a 10K IDE drive, but that ignores the practical reality. If the trend continues (which I personally feel is inevitable), SATA drives will replace even SCSI on the high end. The Apple XServe RAID is the first shot in a eventual deluge. Your "advice" is fast becoming outmoded as a very large array of cheap IDE (and now SATA) disks made available to servers via Fibre Channel are fast replacing the archaic (though technically superior) SCSI. The personal computer was technically inferior to a lot of products (workstations, "business" computers a la Wang, minicomputers, mainframes) which they either replaced or relegated to niche status.
I, for one, am glad that Apple got its head out of its collective a** and adopted certain practical but inferior standards such as IDE/SATA.
(Note to the other poster quoting PC benchmark sites. You have to be careful because in the PC world, on-board SATA wasn't available until recently. This meant you were bandwidth limited by the PCI bus that your expansion card was on. SATA also has the capability to scale far beyond it's current rating (I can't understand why that might be needed yet, but it's nice to know it's there.)
It seems you're the one denigrating other peoples purchasing choices while defending your own. Go read my post and you'll see that isn't the case with me. If it helps, I am not purchasing a G5 because I have no need for it, but I have a friend who has a serious need for it. Yeah, I own an AthlonXP box too, I fail to see how such ad hominem defenses validate your position or mine.
Hmm, if this is a pissing contest and if you wanted speed you'd have bought a IBM Power4 (13GB/s bandwidth) or an Alpha (the most work/clock ever). The Opteron is a great chip, but it is by neither the first nor the best in its class--it does currently have the best price/performance in its class (until Q1 2004 when IBM introduces 2 or 4-way PPC970 (G5) blades). We'll have to wait for the Athlon64 before we talk about great desktop chips, that was only slated for release back in what? Q4 2002... *sigh*
I don't mean to flame you very hard. It's just somewhat shocking when you go on implying that the G5 PowerMac may be Apple's Titanic in a forum about an article with two benchmarks: one of which show that the single processor 1.6Ghz G5 is slightly slower than a double processor 1.4Ghz G4 in a benchmark that has a strong bias toward multiple CPUs, and another benchmark which shows it besting the 2x1.4Ghz G4 in an unoptimized benchmark that is heavily SIMD-dependant--not bad for a VMX core that was widely regarded as "hacked" on!
(And in spite of that, I still believe the MDD dual core G4s still the better machine than the PowerMac G5 in certain instances.)
Take care,
terry
DDR RAM runs at the same speed of standard SDRAM...but moves data twice during a cycle...but does NOT equate to double performance...these are I've found on a quick search to back up the performance issue...I'm sure you can find more if you look for them...I'm not about to waste time looking for references through books that I'm reasonably sure none of you have or will ever have access to...and yes...my patience too is running extremely thin...some of you people are as hard-headed as some of the people I've trained over the years...
http://www.anandtech.com/chipsets/showdoc.html?i=1451
http://www.anandtech.com/chipsets/showdoc.html?i=1373
http://www.anandtech.com/chipsets/showdoc.html?i=1402
http://www.anandtech.com/chipsets/showdoc.html?i=1414&p=10
http://www.anandtech.com/chipsets/showdoc.html?i=1344
http://www.anandtech.com/chipsets/showdoc.html?i=1386
http://www.tomshardware.com/motherboard/20001030/index.html
http://www.tomshardware.com/motherboard/20010104/amd-133fsb-07.html
http://www.sharkyextreme.com/hardware/reviews/mainboards/gigabyte_ga-7dx/index.shtml
http://www.sharkyextreme.com/hardware/reviews/mainboards/asus_a7m_266/index.shtml
http://www.sharkyextreme.com/hardware/articles/iwill_ka266r_benchmarks/index.shtml
...every one of these are tests performed by reputable sites...comparing DDR Based Chipsets to SDR Based Chipsets...if you want to dispute this issue further...don't waste my time...I'm right...
You seem to do a wonderful job of shooting yourself in the foot:
1) All of them date back between late 2000 to early 2001 back when the Athlon CPUs (no P4s were out yet) were ~1Ghz and thus not bandwidth limited. These CPUs are literally drowning in data.
2) Many Athlon chipsets of that generation (if you remember) had an FSB of 133Mhz. I couldn't find a single mobo that actually benchmarked a 266Mhz FSB. A few benchmarked a 200Mhz effective FSB--I should add here that a 200Mhz FSB would, most likely, have a DDR200 RAM paired with it so it would be clock-doubling 100Mhz SDRAM). Thus, many of these run into the exact same FSB issues that the G4 currently has (delta the shared bus issue since few mobos were dual CPU, OTOH they have no workaround like an L3 cache either).
3) All of them were benchmarking the AMD Athlon which I mentioned in one of my first posts in this thread has a narrow internal memory access bus which means that its quad integer/fpu units become data starved, not because of the FSB, but because of the datapath internal to the chip. This is why the P4 beats the Athlon pretty easily on the top end (even the reputable sites you mention now admit this).
4) The DDR chipset was in its infancy. A couple of the articles allude to the horrible performance of early chipsets. In fact, the fastest board at the time was AMDs own reference implementation, which shows that nobody had yet gotten a handle on it. Even then, some of the performance gains in your article were as high as a 20% overall increase in performance (not memory benchmark)!
I bet if you went on anandtech, tomshardware, or sharkeyextreme right now and posted that DDR represents only a 5-7% memory performance gain over an SDR at the same clock speed on today's top-end CPU and chipsets you would be flamed much harder than the kid gloves these MacRumors folk have treated you with. And nobody has yet claimed the double bandwidth means double performance, you misread that... just as you have misread the articles you cite!
Oh I noticed you're no longer claiming that it was all about latency. I guess that is because as a self-proclaimed assembler of PCs you must have finally looked at your BIOS settings and noticed that latency timings and RAM clock speed are two separate settings!
Hannibal wrote that before had had all the answers. IBM wasn't talking yet. He has backed off on his criticism of the vector unites on the G5 somewhat. The biggest problem for now is programmers changing the way they write and compile apps to work better with the vector units on the G5.
Re: G5 Memory
You are correct. I've already bought my first upgrade of memory, and after shopping around, ever single vendor (not to mention Apple has said it too) says that you have to put RAM in pairs.
Not to mention that when you read the specs it says 128 bit memory data path (or something like that). The chipes are 64 bit in singles.
What REALLY blows me away is all this theorectical falderal and not much of a mention of the Apple developed chipset that holds all the components together as a whole.
NOBODY knows what the performance of the 1.8 or dual 2GHz machines are going to turn out to be. Way too many holes in the arguments here - the biggest one is nobody has tested the big boys.
You are correct. I've already bought my first upgrade of memory, and after shopping around, ever single vendor (not to mention Apple has said it too) says that you have to put RAM in pairs.
Not to mention that when you read the specs it says 128 bit memory data path (or something like that). The chipes are 64 bit in singles.
What REALLY blows me away is all this theorectical falderal and not much of a mention of the Apple developed chipset that holds all the components together as a whole.
NOBODY knows what the performance of the 1.8 or dual 2GHz machines are going to turn out to be. Way too many holes in the arguments here - the biggest one is nobody has tested the big boys.
Re: Daveg5
Thia obaervation is the one thing that keeps me from going out and buying an Apple right now. The laptop configurations are laughable, you have to wonder if they expect that most of their customers are still running OS9. OS/X is an operating system that loves memory you would think that a PowerBook at least would ship with enough memory to at least make OS/X look good.
The video configuration is just as bad. You have to wonder about Apples commitmant to Quartz and other optimizations. I'm not talking aobut configuraitons for gaming specifically, just reasonable performance systems that are reasonablly state of the art.
Lets face it the G5 is not demonstrating that Apple has a clue here.
Thanks
Dave
Thia obaervation is the one thing that keeps me from going out and buying an Apple right now. The laptop configurations are laughable, you have to wonder if they expect that most of their customers are still running OS9. OS/X is an operating system that loves memory you would think that a PowerBook at least would ship with enough memory to at least make OS/X look good.
The video configuration is just as bad. You have to wonder about Apples commitmant to Quartz and other optimizations. I'm not talking aobut configuraitons for gaming specifically, just reasonable performance systems that are reasonablly state of the art.
Lets face it the G5 is not demonstrating that Apple has a clue here.
Thanks
Dave
Re: Re: Daveg5
In what way is the G5 graphics selection lacking? Apple has the same performance range of video cards available on the G5's as on PC's...from the low end nvidia 5200 to the top-of-the-line ATI 9800 Pro. I don't see that there is a problem here.
In what way is the G5 graphics selection lacking? Apple has the same performance range of video cards available on the G5's as on PC's...from the low end nvidia 5200 to the top-of-the-line ATI 9800 Pro. I don't see that there is a problem here.
Re: Re: Re: Daveg5
There aren't any professional video cards (Quadro, FireGL, Wildcat) for the Powermacs although this could change now that we have a decent workstation class system.
There aren't any professional video cards (Quadro, FireGL, Wildcat) for the Powermacs although this could change now that we have a decent workstation class system.
Re: Re: Re: Re: Daveg5
Yes, perhaps. I am not too familiar with hi-end 3D packages, but I know that these GPU's (whether consumer, gamer, or pro) are not used for non-real-time renders. They can only benefit the low-detail GUI design and proofing displays--so it seems to me that this is a pretty marginal gap in capabilities.
I am still curious as to what Dave means when he says the graphics options are so poor.
Originally posted by Cubeboy
There aren't any professional video cards (Quadro, FireGL, Wildcat) for the Powermacs although this could change now that we have a decent workstation class system.
Yes, perhaps. I am not too familiar with hi-end 3D packages, but I know that these GPU's (whether consumer, gamer, or pro) are not used for non-real-time renders. They can only benefit the low-detail GUI design and proofing displays--so it seems to me that this is a pretty marginal gap in capabilities.
I am still curious as to what Dave means when he says the graphics options are so poor.
tychay seems to know quite a bit more about this area than I do, so I'll let him deal with this particular post.
But I'm noticing an interesting pattern here in this thread: you make a disparaging statement about the G5, or one that's simply not true; then I and sometimes a few other people call you on it and use *gasp* logic and some simple math to try to prove you wrong. I'm perfectly aware that what intuitively seems right may not hold up to scientific testing, or that things may be more complex than they seem. With adequate references to back up your unintuitive contentions, I'd believe you. But in this pattern, you respond with outdated and/or irrelevant references (if any), telling us all how stubborn we are and how wise and experienced you are all the time. Then, at least in the architecture discussion, you admitted defeat and I tried to respond with an understanding post that put that discussion to rest, at the same time pointing out the irrelevance of your first reference in the DDR/SDR debate. That launched us into another cycle that seems to be following the same pattern.
FWIW
WM
I don't know why I'm even bothering with writing this...maybe boredom...maybe massochism...BUT...it does seem people have misconstrued and twisted the intent of some of things I've posted
this would be the comment that started it all...not to mention there were a few cockles raised when I attempted to use an analogy to illustrate that point...
by using comparisons between PC1600 v. PC100 and PC2100 v. PC133...I've tried to explain that there is not a major jump in speed or performance (regardless of theoretical bandwidth) between DDR and SDR...when comparing to it's same speed counterpart...
and as tychay astutely pointed out:
All of them date back between late 2000 to early 2001...These CPUs are literally drowning in data.
all those tests were performed in 2000-2001...because there is no PC200 SDR... and you can't perform a fair or accurate comparison between PC3200 and PC100 or PC133...unless you're trying to prove how far chipsets and processors have progressed in the last couple years...PC3200 DDR can only be fairly judged against PC200 SDR...which does not exist...and the CPU's in question aren't "Drowning in Data"...they suffer from the same problems of every generation of PC...being choked by bloated and badly written code...namely Windows...
...but also there have been some who've actually read what was intended...
and
...my issue is not that the G5 is a bad machine...nor do I believe the G4 is faster (although...for now...many synthetic tests paint a differing picture)...when the G5 finally gets an OS (with more than a few optimised libraries) that is written specifically to address it's unique archetecture...and when apps are finally coded to take advantage of the PPC970...it'll be one of the fastest PC's available...BUT as I've stated...ad nauseum...I bought a G4 (to replace a dead G3)...and I found the G4 to be a much better "Fit" for what I wanted/needed in a desktop machine...and seeing that I paid $150 under MSRP it gave me more than enough money to add a few minor upgrades...I'm not defending a purchase...I'm not bashing the G5...the G4 (for what I paid...and the cost of my extras) was more fiscally responsible and had more "bang for the buck"...
...and the comments at the close that were taken WAY out of context...the sincere hope that Apple hasn't shot themselves in the foot with the G5 is warranted...the G5 is a big step...and although they've 100,000 units preordered (credit to WM. for the info)...Apple will need to move far more units than that to make it viable...
...if Itanic (ITANIUM, Merced or IA64...or whatever name Intel's calling it this week which has again been delayed...this time to Q3 2004) fails...they can easily take the hit...
...if AMD's Athlon64 (which still won't have a Microsoft OS until Q2 2004) stumbles...no problem...they've very deep pockets...
...BUT...if the G5 fails to draw enough "Switchers" and new users into the fold, it could mean financial trouble to an already unstable company...and personally that's the last thing I want to see happen...
...last and certainly not least...my apologies to WM. and tychay...the "Slow Learner" comment was aimed at one person...not either of you...
this would be the comment that started it all...not to mention there were a few cockles raised when I attempted to use an analogy to illustrate that point...
by using comparisons between PC1600 v. PC100 and PC2100 v. PC133...I've tried to explain that there is not a major jump in speed or performance (regardless of theoretical bandwidth) between DDR and SDR...when comparing to it's same speed counterpart...
and as tychay astutely pointed out:
All of them date back between late 2000 to early 2001...These CPUs are literally drowning in data.
all those tests were performed in 2000-2001...because there is no PC200 SDR... and you can't perform a fair or accurate comparison between PC3200 and PC100 or PC133...unless you're trying to prove how far chipsets and processors have progressed in the last couple years...PC3200 DDR can only be fairly judged against PC200 SDR...which does not exist...and the CPU's in question aren't "Drowning in Data"...they suffer from the same problems of every generation of PC...being choked by bloated and badly written code...namely Windows...
...but also there have been some who've actually read what was intended...
and
...my issue is not that the G5 is a bad machine...nor do I believe the G4 is faster (although...for now...many synthetic tests paint a differing picture)...when the G5 finally gets an OS (with more than a few optimised libraries) that is written specifically to address it's unique archetecture...and when apps are finally coded to take advantage of the PPC970...it'll be one of the fastest PC's available...BUT as I've stated...ad nauseum...I bought a G4 (to replace a dead G3)...and I found the G4 to be a much better "Fit" for what I wanted/needed in a desktop machine...and seeing that I paid $150 under MSRP it gave me more than enough money to add a few minor upgrades...I'm not defending a purchase...I'm not bashing the G5...the G4 (for what I paid...and the cost of my extras) was more fiscally responsible and had more "bang for the buck"...
...and the comments at the close that were taken WAY out of context...the sincere hope that Apple hasn't shot themselves in the foot with the G5 is warranted...the G5 is a big step...and although they've 100,000 units preordered (credit to WM. for the info)...Apple will need to move far more units than that to make it viable...
...if Itanic (ITANIUM, Merced or IA64...or whatever name Intel's calling it this week which has again been delayed...this time to Q3 2004) fails...they can easily take the hit...
...if AMD's Athlon64 (which still won't have a Microsoft OS until Q2 2004) stumbles...no problem...they've very deep pockets...
...BUT...if the G5 fails to draw enough "Switchers" and new users into the fold, it could mean financial trouble to an already unstable company...and personally that's the last thing I want to see happen...
...last and certainly not least...my apologies to WM. and tychay...the "Slow Learner" comment was aimed at one person...not either of you...