Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

ajkst1

macrumors member
Original poster
Nov 11, 2003
43
0
I'm taking a class at my university and my professor claims that 800 mhz (AMD Opteron) is the fastest front side bus in production. I know better than that because the G5's FSB is equal to half the clock speed of the processor. What is the actual front side bus speed, i.e. what are people actually getting in real world use vs. laboratory testing. He doesn't want Apple marketing nonsense (I showed him that already), he wants technical details and proof (screenshots? :)) that the FSB of a G5 is equal to half the clock speed of the processor.

HELP ME WIN THIS ONE GUYS!
 
it a really good question because it where the number is taken from can greatly effect what it is

For example I have a AMD64 3000 in my computer. The mobo is telling my the front side bus is at 200mzh (so does AMD stating that is there FSB number) but that number is taken earily then where it is recored in a P4 chances are the apples does the same so I dont really buy apple number. AMD number is taken from where the data is going both to the ram nad processor so it is running at 200mhz (PPC3200 ram before it his the mulitplyer to up it to 400mhz).

Now if you take the number from the same spot apple and Intel take it the AMD64 are running at 1.6ghz fsb but the AMD64 dont have a standard frount side bus since there memory control modle is integreted into the CPU it self. Something only they have which really up the speed. If the Optron is running on sock 754, 939 or 940 (sorry I can not remeber what the ops run on off the top of my head) then I think it has the ingteted memory modle as well and in reallity it running at 1.6ghz if you want to ajusted it to the what apple and Intel use.

I am sorry but in relaielyt you going ot loose this one. AMD optron and 64 line have a much faster memory control (FSB more or less) than any one else and it faster than apple and Intel.
 
I don't think its actually half the FSB, but you also need to remember that FSB doesn't always translate directly to more performance.

Anyway, in the 939 socket chips, the 200Mhz "FSB" is multiplied by the HTT multiplier (defaulting at 5x), while 754 socket chips are 200Mhz "FSB" with a 4x HTT multiplier.

Comparing P4's to the 939's or the 754, they both could have a theoretical "800FSB" or "1000FSB", but the 754's, even at a much lower clock speed, the AMD chips will beat Intel chips with the same "FSB". That is, if you are comparing CPU performance from a FSB perspective.

That said, you could then move to comparing memory bandwidth as another method for comparing FSBs, for example, but then A64 will win hands down b/c of its inherently reduced latencies. Is there a way of comparing FSBs fairly? If you had a I/O chipset benchmark, that is compiled and tested fairly on both sides, perhaps, but comparing PC to PC is hard enough. Comparing PC to Apple only makes this even more of a grey area.
 
also remember how much BS there is in marketing the FSB. and everything else. Realtiy is Apple is starting to get into the famus Mhz myth and starting to almost validated it now. Remeber there is the thing call the mhy myth and there currently is really only one complain not playing into it. Yes it is AMD. they dont brag about there Mhz of there chips. Apple promised 3ghz chips (failed but they are trying) and now are trying to brag about there mhz.

There is a lot more stuff that factors into the speed of a computer
 
Why are you looking here, you should be looking at the IBM site for documentation.

Specifically one of the in depth manuals on the PPC970...

970FX_user_manual_v1.41.pdf

But it does run at half the clock, and is a variant of IBM's elastic bus -- using two 32-bit unidirectional buses.
 
I don't think when Steve said they would be at 3GHz it was a case of bragging. Of all the CPU makers it has only ever been Intel who have had the figures plastered around. Any PC owner with sense and some idea of what really goes on in the industry would buy AMD processors. Even AMD however were forced to call there processors after the equivalent speed compared to the P4 ie; Athlon 2400 etc. Now Intel have backed themselves in to a corner and have had to move away from naming their CPU's after their speed because they just aren't getting the gains they used to.
Current P4 designs have a FSB's of 800 and 1066MHz, these DDR buses are quad pumped so are actually only 200MHz (in the case of the 800MHz). If someone can explain the 'pumping' principal please do.
The Apple DDR buses on the G5 are double pumped so the true FSB of the dual 2.5 is 625MHz far faster than that of the Pentium 4.
Working out the AMD Athlon 64 FX's FSB is rather harder as someone has pointed out already. AMD love quoting figures in MTps or mega transfers per second. All I can tell you there fore is that AMD were scheduled to release a 1GHz FSB CPU at the start of 2004. From the technical sheets I've managed to see I think they are still at the 1GHz mark and by the looks of it it is double pumped like the G5 so the actual non DDR speed is 500MHz.
The major advantage the with the Athlon 64 design is the on chip memory controller negating the need to have such extreme bus speeds as RAM content can get to the CPU's faster than in the G5 design.
In many ways, memory controllers asside, the AMD and IBM/Apple designs are actually very similar both using Hypertransport technology as a way of increasing throughput.
Strictly speaking then I would say that the G5 does indeed have the fastest FSB, but the award for most efficient and there fore over all fastest would go to the AMD Athlon.
 
A bit more research;
http://www.amd.com/us-en/Processors/ProductInformation/0,,30_118_8826_9014,00.html
In the Opteron the memory FSB is the same as the processor as AMD are quick to point out. But the FSB for other data throughput to the processors is much slower as discussed above. So really it depends on what your teacher meant by FSB, if he meant from RAM then he is right, if he meant all data then you are correct.
The interesting thing about the the Opteron is how it scales up to 4 CPU systems. The bandwidth is dedicated for each CPU so while comparing a 1 CPU Opteron to a single G5 gives pretty similar results when you start scaling with 2 or more the bandwidth figures for the AMD system take a huge lead. Where the G5's FSB does excel is CPU to CPU transfers here there isn't a main stream system to touch it.
 
I can't really elaborate much more than they have already, just simplify it.

AMD had DDR FSBs. So the first Athlon was 100MHz double pumped to 200MHz. Like with DDR RAM, rising and falling. The latest was 400, so 200 DDR. The current 64-bit CPUs do have integrated, on chip FSBs... but as said how fast they talk to the memory is a different matter. If you're talking just regular FSB, it is full speed. Up to 2.6 I believe.

The P4 is quad-pumped. 1066 is actually 266 x4. Like RAMBUS.

The G5 is half the CPU speed on the high end. It's not marketing speak, that's how fast it is. The G5 in the eMac uses a slower FSB for lower heat. But the 2.5 uses a 1.25GHz FSB. Whether that makes it faster than what AMD has to offer is debatable. The G5's bus is what it is, and the AMD's is full speed, but there are caveats as to how fast it talks to the RAM.

I'm afraid it's an arguement neither of you can win because there are too many other variables, so you might want to just let it go. Since the fastest standard RAM at this point is dual-channel 400MHz (200 DDR x2, x2 for an effective 800), DDR2 notwithstanding, it probably doesn't matter either way anyway.
 
solvs said:
I can't really elaborate much more than they have already, just simplify it.

AMD had DDR FSBs. So the first Athlon was 100MHz double pumped to 200MHz. Like with DDR RAM, rising and falling. The latest was 400, so 200 DDR. The current 64-bit CPUs do have integrated, on chip FSBs... but as said how fast they talk to the memory is a different matter. If you're talking just regular FSB, it is full speed. Up to 2.6 I believe.

The P4 is quad-pumped. 1066 is actually 266 x4. Like RAMBUS.

The G5 is half the CPU speed on the high end. It's not marketing speak, that's how fast it is. The G5 in the eMac uses a slower FSB for lower heat. But the 2.5 uses a 1.25GHz FSB. Whether that makes it faster than what AMD has to offer is debatable. The G5's bus is what it is, and the AMD's is full speed, but there are caveats as to how fast it talks to the RAM.

I'm afraid it's an arguement neither of you can win because there are too many other variables, so you might want to just let it go. Since the fastest standard RAM at this point is dual-channel 400MHz (200 DDR x2, x2 for an effective 800), DDR2 notwithstanding, it probably doesn't matter either way anyway.

Apparently there is DDR 500 now as well so dual channel would allow for 1000MHz. DDR2 is faster but currently you pay through the nose for it.
Maybe I misread the AMD documents but I thought it was the RAM which got the full processor speed FSB while the rest of the components got a much slower bandwidth around 1000MHz.
 
The Frontside Bus is actually the speed in which you multiply the multiplier to get the clock speed.

example: 200MHz FSB times 10x multiplier equals 2GHz processor

when labeling the FSB, if they are double or quadruple-pumped they will put that instead. It should be actually labeled as MT/s or megatransfers.

On Intel's and AMD's PDF spec sheets they will label it as 800 MT/s or 2000 MT/s

AMD has the fastest bus at 2x1000 MHz or 2000 MT/s
 
also rememeber the FSB is taken from the apple after it hit the first multipleir. Because if you take a look at the mobo speed for the FSB of intel is really at 200mhz (limited by the ram speed). Hmm who would of known stuff about overclocking would come in handy

Since the apple computer are using PC3200 ram it fsb speed is going to be set at 200mhz. PC3200 ram is only designed to take a fsb speed of 200mhz the ram it self has a 2x mutiplier built into it to up it to 400mhz. So other the fsb on the apple computers is after the first mulitply that ups it to half the reported clock speed. then it hits anthere 2x mulitplier to go up to the cpu reportec clock speed.
 
Wow I learned something, but like they said, too many variables to calculate in. Now who said that AMD would beat Intel, even if they were the same FSB speed? - He obviously knows his stuff then. Everyone thinks Intel's faster because it has Hyper-Threading Technology (HTT), but AMD is because of its Hyper-Transport Technology (HTT) and lets not forget the 64KB & 128KB L1 Cache... hmm... who else does that.... oh yeah Apple =D. While Intel only has a meezly (I know I spelled that wrong) 32KB (12 for data, 20 for instruction or is it vice-versa). AMD's is 32 & 32 and 64 & 64 data and instruction, respecitvely. While I don't know how IBM allocates Apple's L1 Cache for their processors.

I read that somewhere that the Cache doesn't run at full speed but half on Apple processors, is that true? And does it detract or add to performance? I know AMD's and Intels cache run at full speed, but yeah. I'll check back laterz.
 
The G5 (970FX) uses 64k Instruction + 32k data L1 cache, along with 512k L2 per processor. The "old" G5 (970) uses 64k L2 (32k data, 32k Instruction). Those caches are all running at processor speed.

Opteron and Athlon XP use a full 128k L2 and 512k L2 of the Athlon XP or 1MB L2 on the Opteron, but I'm not sure of the speeds.
 
only on the older g4's with external l2 cache like the 7400/7410 and external l3 cache with the 745x ran as less than the cpu speed.
 
Everyone here is talking as if the FSB only communicates to the System RAM. The FSB speed is the speed that it communicates from the CPU to the System controller on a Mac at least. From the system controller the speed is at whatever level the individual components communicate at. The advantage of having a FSB many many times faster then any one component is the lack of bottlenecks. All these components communicate to the system controller at the same time and then all the data is piled into the FSB channel. The faster the FSB the faster that group of data moves through without building up a que.
 
MacBandit said:
Everyone here is talking as if the FSB only communicates to the System RAM. The FSB speed is the speed that it communicates from the CPU to the System controller on a Mac at least. From the system controller the speed is at whatever level the individual components communicate at. The advantage of having a FSB many many times faster then any one component is the lack of bottlenecks. All these components communicate to the system controller at the same time and then all the data is piled into the FSB channel. The faster the FSB the faster that group of data moves through without building up a que.


The reason we are using the ram as the FSB is the FSB is what is used for the CPU to commicated with the rest of the sysstem yes. Also the FSB commmicated with the system ram. I could go in and start explain the diffent parts. Just the FSB part is the one the commicateds with the ram and CPU. Prouble is you push the FSB to high *aka over 200mhz it is going to make the system unstable because the ram can not handle it. PC3200 ram is not designed to run over 400 mhz after the multiplier (2X200mhz=400mhz) to go higher than that a divider needs to be put on the fsb for the ram.

Start looking into over clocking and when the fsb is starting to be pushed past 250mhz the cooling on it starts running into problems. when people Watercool there system some of it goes FSB to keep that from over heating.
 
Timelessblur said:
The reason we are using the ram as the FSB is the FSB is what is used for the CPU to commicated with the rest of the sysstem yes. Also the FSB commmicated with the system ram. I could go in and start explain the diffent parts. Just the FSB part is the one the commicateds with the ram and CPU. Prouble is you push the FSB to high *aka over 200mhz it is going to make the system unstable because the ram can not handle it. PC3200 ram is not designed to run over 400 mhz after the multiplier (2X200mhz=400mhz) to go higher than that a divider needs to be put on the fsb for the ram.

Start looking into over clocking and when the fsb is starting to be pushed past 250mhz the cooling on it starts running into problems. when people Watercool there system some of it goes FSB to keep that from over heating.
That description is accurate for anything that isn't based on the PowerPC G5. The G5 is different. Unlike other computers, the G5's FSB to CPU multiplier is fixed regardless of the CPU speed. This means that, as shown in Apple's PowerMac line, the faster the CPU (1.8 GHz to 2.0 GHz), the faster the system bus (450 MHz to 500 MHz). However, this bus is DDR, so it effectively doubles its speed (900 MHz to 1000 MHz). Also, the RAM speed is not hard-locked to the system bus speed anymore - it communicates with the system bus as fast as it is able to (with 200 MHz DDR RAM in a dual channel configuration, this speed is 4 times the RAM speed - 800 MHz). It is true that the system bus speed and the RAM speed no longer match under this configuration, but as it turns out, this is desirable due to additional latency accessing the RAM with this configuration (no, I don't know where the additional latency comes from).

For more information regarding this topic:

Ars Technica article on the PPC 970 part 1
Ars Technica article on the PPC 970 part 2
 
Please remember the original question was just about FSB not the memory aspect only. AMD wins the RAM to CPU FSB (2+GHz) but the speed with which other parts communicate such as processor to processor is slower. Apple does have the fastest FSB at least on paper for all general computing activities (bar RAM transfer).
For those discussing RAM as a side issue also remember that the dual channel aspect of the G5's memory controller means that the effective FSB for memory is essentially 800MHz, or 400MHZ before the 2x DDR multiplier. The faster the RAM they use the faster the transfer will be so they could in theory use PC2-5300 (DDR2-667) which in dual channel operation would give 1.334GHz. The problem here is without the onboard memory controller seen in the AMD processors the FSB will become a bottle neck once again, albeit only in extreme cases.
Ideally once apple release the 3GHz+ G5 this will no longer be an issue and they can upgrade the memory they use. The problem is, as pointed out, AMD can state all they like that the CPU can access RAM at processor clock speeds but the RAM just can't keep up so the technology is essentially wasted especially as I know of no AMD motherboard which will support the fastest DDR2 standards. If you ignore latency issues the G5 setup is the most practical for everyday computing. If IBM add the fabled on chip memory contoller then it is no bad thing but I can't forsee any anormous gains compared to what we have now.
 
psycho bob said:
Please remember the original question was just about FSB not the memory aspect only. AMD wins the RAM to CPU FSB (2+GHz) but the speed with which other parts communicate such as processor to processor is slower. Apple does have the fastest FSB at least on paper for all general computing activities (bar RAM transfer).
For those discussing RAM as a side issue also remember that the dual channel aspect of the G5's memory controller means that the effective FSB for memory is essentially 800MHz, or 400MHZ before the 2x DDR multiplier. The faster the RAM they use the faster the transfer will be so they could in theory use PC2-5300 (DDR2-667) which in dual channel operation would give 1.334GHz. The problem here is without the onboard memory controller seen in the AMD processors the FSB will become a bottle neck once again, albeit only in extreme cases.
Ideally once apple release the 3GHz+ G5 this will no longer be an issue and they can upgrade the memory they use. The problem is, as pointed out, AMD can state all they like that the CPU can access RAM at processor clock speeds but the RAM just can't keep up so the technology is essentially wasted especially as I know of no AMD motherboard which will support the fastest DDR2 standards. If you ignore latency issues the G5 setup is the most practical for everyday computing. If IBM add the fabled on chip memory contoller then it is no bad thing but I can't forsee any anormous gains compared to what we have now.

You are wrong on the ram speed. it is under 250mhz. Take the ram speed and divid it by 2. The fast out there is 566mhz ram which is only 233mhz.(it is 233*2) it is 233mhz before it hits the DDR2 2x multplier.

As for DDR2 mobo supported by AMD a few socket 940 and 939 supported itor will soon/ 754 doesnt not and never will because it does not have the bandwith for it but the funny part is even with out DDR2 it is getting though more data than the one with DDR2 due to the ingetraded memory control on the CPU it self. But also remeber DDR2 is still really new. and the AMD chips are still out running the intel chips that do support along with the Apple chips because of the ingrated modal it self
 
Anyway DDR2, as the rumor mill goes, may be skipped in its entirety for the next generation DDR standard. DDR2, until only recently, finally started reaching latencies that are even marginally close to DDR (cas 3). Cas 3, in the DDR1 world is value-ram territory, where a gig can be had for as low $100-120. Try finding a gig of DDR2 for that kind of money >> it ain't gonna happen.

Cas3 DDR2 doesn't come in until you hit the $230-250 mark. Just for cas3!

On your other point of DDR maxing out at 233, I'm afraid I'm going to have to differ. G.Skill, a Taiwanese Memory company is one of many companies using the legendary Samsung TCCD (Brainpower PCB) DDR chips that are "rated" at up to 250Mhz officially. However, G.Skill has tested their chips to hit as high as 275Mhz, and obviously overclockers reach even higher than that (e.g. 290 or 300Mhz are not uncommon).
http://www.gskill.com/pc4400-2.5-3-3-7-pc3200-tccd-2-2-2-5-dc.html

And for your viewing pleasure, a pic of an overclock by one of the GSkill employees themselves:
http://www.e04hardware.com/fx55/fx55ram9k.jpg

Yes that means DDR600 if you want marketing speak. They do tend to work much better on the 939 platform (the 754 platform can only do it with one stick, up to 240-250, as opposed to 939 where it can hit 300 easily in dual channel).
 
Not wanting to argue with you I just got the details off the Crucial website. The fastest non DDR2 RAM they sell is PC4000 or DDR 500. The fastest DDR2 RAM they sell is 667MHz. I quite agree that with DDR RAM you have the speed to get the true clock speed. You use the example that the fastest is DDR 566MHz but 233x2 is not that. So unless 100MHz just vanished you did your sums wrong.
Using dual channel RAM effectively doubles the bandwidth. So for DDR400 it works out as 200x 2(multiplier) = 400 x2(dual channel) =800MHz.
For single CPU Opteron systems the memory bandwidth is the same as it is for the G5 6.4Gb/s, the only question is one of latency which favours the AMD design. Only when you add more CPU's does the bandwidth increase over that of the G5. I assume ths isbecause while the G5 shares RAM between processors the Opteron has it dedicated to each CPU is dual configuration.
I still don't see how having the integrated memory controller means you can get data faster than the speed of the RAM you use. All you do is remove the latency and free up the FSB into the CPU for other tasks.
 
psycho bob said:
Not wanting to argue with you I just got the details off the Crucial website. The fastest non DDR2 RAM they sell is PC4000 or DDR 500. The fastest DDR2 RAM they sell is 667MHz. I quite agree that with DDR RAM you have the speed to get the true clock speed. You use the example that the fastest is DDR 566MHz but 233x2 is not that. So unless 100MHz just vanished you did your sums wrong.
Using dual channel RAM effectively doubles the bandwidth. So for DDR400 it works out as 200x 2(multiplier) = 400 x2(dual channel) =800MHz.
For single CPU Opteron systems the memory bandwidth is the same as it is for the G5 6.4Gb/s, the only question is one of latency which favours the AMD design. Only when you add more CPU's does the bandwidth increase over that of the G5. I assume ths isbecause while the G5 shares RAM between processors the Opteron has it dedicated to each CPU is dual configuration.
I still don't see how having the integrated memory controller means you can get data faster than the speed of the RAM you use. All you do is remove the latency and free up the FSB into the CPU for other tasks.

other wise it also reduces stress off the FSB to where it can do more with the ram instead of hte CPU. Sorry dont remeber where they showed the bench marks in how the lack of DDR2 or running full DDR the AMD 64 where still faster than the Pent 4 that could run the DDR2. The intgerted control basicly makes the CPU much less relanted on the FSB to handed some stuff. But to translate to the intel numbers from what AMD runs it runs at 1.6ghz. So that should be some food for though
 
um, RAM and FSB is all about latency. While the G5's run on 3-3-3-8 sticks of RAM, Opterons/FX's run can from anything as slow as 3-3-3-8, to something as fast as 2-2-2-5.

If you don't understand latencies, then you won't understand why AMD has an inherent memory bandwidth advantage (at least for now). If G5's could run their memory at cas 2 and Trcd (ras to cas) 2, they'd gain a substantial amt of bandwidth, but w/ the delay of communicating to a northbridge on the motherboard, the benefits of these low latencies are considerably reduced.

An Opteron/FX/A64 (939) chip can "have cake, and eat it too" cuz not only does it have the integrated memory controller, but it has the option of running cas2 and trcd 2 as well. And of course, overclocks to further enhance bandwidth.

Considering I haven't really looked into the G5 firmware, I wonder if it is even possible to change those latencies in the OS. With Windows, it is done rather easily in real-time, so I'm sure if someone develops to software, Macs can at least get cas2 and trd2.
 
Just read an interesting article on tomshardware.com which discusses dual channel RAM and what bandwidth differences it gives. For systems without on board memory controllers or ultra fast DDR buses like the G5 the RAM is held back by the FSB of the processor. So in the case of an Athlon XP running at 200MHz using DDR400 in dual channel configuration would only provide 3.2GB/s of bandwidth rather than the 6.4 we see on the G5 and Opteron systems.
The G5 does not have this problem its FSB can easily soak up the full throughput of dual channel DDR400 and apple could have increased the RAM spec to DDR500 and it would still have been fully utilised.
 
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.