More Consideration

[FatBurger]

You're in Hillsboro? I'm in Portland, as are Matisaro and Jollygrinch.

<font color=orange>Quarter</font color=orange> <font color=blue>Pounder</font color=blue> <font color=orange>Inside</font color=orange>

 

[Raystonn]

Take this scenerio: memory traffic is at it's maximum, and the FSB is running at the same speed as the RAM. And then, a PCI care or AGP card requests to send data to RAM or to the CPU. It'll have to wait until free cycles are available for the FSB, the CPU isn't the limitation here because it can process information significantly faster than the main memory can provide it. With a faster FSB, wouldn't the FSB be able to manage saturated memory traffic as well as be able to relay messages from the southbridge to the CPU, and thus wasting less cycles.

This is exactly what happens. Your FSB is used to transmit data back and forth between the processor and a number of components. These include main memory, AGP, PCI cards, EIDE/RAID controllers, USB ports, etc. Most of the smaller components actually share with the PCI bus. The point here is that in order for the FSB not to bottleneck your system at all, it needs enough bandwidth for all of these things to be used at once.

If you have 3.2GB/s of FSB bandwidth and 3.2GB/s of main memory bandwidth, then what happens in your memory intensive applications when the AGP port's 1066MB/s of bandwidth kicks in? Now your FSB is acting as a bottleneck. You cannot possibly access your full memory bandwidth and AGP badnwidth at the same time. Add in a couple EIDE channels, perhaps in a RAID configuration, and now you have even more demand for FSB bandwidth. Increasing your system's FSB bandwidth is going to increase system performance even if memory bandwidth remains the same. Nothing kills memory latency more than having to wait for a PCI, AGP, or hard drive data transfer.

-Raystonn


= The views stated herein are my personal views, and not necessarily the views of my employer. =

 

[Raystonn]

To supplement my last post, take a look at <A HREF="http://www.aceshardware.com/Spades/read.php?article_id=45000219" target="_new">this</A> article. In particular look at the figures in the table. Take note of the increased bandwidth and lower latency of the PC800 at 533MHz FSB as compared to the PC800 at 400MHz FSB.

-Raystonn


= The views stated herein are my personal views, and not necessarily the views of my employer. =

 

[FatBurger]

I'm not sure I understood that artcle. Is the PC800@533MHz running with a x3 multiplier, instead of x4?

<font color=orange>Quarter</font color=orange> <font color=blue>Pounder</font color=blue> <font color=orange>Inside</font color=orange>

 

[eden]

Hmm Ray looking at those benchs, I see nothing much exciting from adding 133MHZ to FSB and 266MHZ more to RDRAM. It rather shows that even with those increases, the P4 fails to deliver any better performance in the near future. There were a few nice performance additions, but often games only resulted in 2-10 more FPS, and much of it is not noticeable. IMO the problem is not lying only in bandwidth, the P4 is simly asking to get its much needed components back: FPU, L1 Cache and more... It's just not good like that, and if it had been added to it, the RDRAM+FSB increase would have been even more dramatic than just this.

--
For the first time, Hookers are hooked on Phonics!!

 

[Raystonn]

I'm not sure I understood that artcle. Is the PC800@533MHz running with a x3 multiplier, instead of x4?

Yes. It shows the effect of attaining a higher FSB without having to overclock your memory. You still achieve increased system performance, even without an increase in processor clockspeed or memory bandwidth.

-Raystonn


= The views stated herein are my personal views, and not necessarily the views of my employer. =

 

[Raystonn]

Hmm Ray looking at those benchs, I see nothing much exciting from adding 133MHZ to FSB and 266MHZ more to RDRAM.

This is obviously highly application dependant. While the reduced latency will increase system performance across the board, the increased bandwidth only helps if you actually use it. Some of those benchmarks showed a 12% performance increase with the added FSB and RDRAM bandwidth alone. Performance can increase by up to 33% in bandwidth-hungry applications due to the 33% extra memory bandwidth.

It rather shows that even with those increases, the P4 fails to deliver any better performance in the near future.

Are we looking at the same numbers? I would say a 12% performance improvement without any increase in processor clockspeed is pretty nice. Somewhere near a 33% performance improvement would be likely in the more bandwidth-intensive applications such as Quake3A.

-Raystonn


= The views stated herein are my personal views, and not necessarily the views of my employer. =

 

[ath0mps0]

Ray,

I just looked at this article and it proves my point!!!!!!!!!!!!! (just had to put them !!!!s in thar). The increase in performance by staying at PC800 while increasing FSB to 533 is great, but nothing compared to the increase that you see when pairing a 533MHz FSB with PC1066. Now imagine if you could do the same when you overclock your 1.6A to 2.5MHz...but you'd need PC1250 to do it. This is pushing the envelope with RDRAM. Maybe Samsung will accomplish it late this year, maybe next. If (beating that dead horse again...) Intel were to release a dual-channel DDR P4 chipset, you could do this with today's PC2400 or PC2700.

I thought a thought, but the thought I thought wasn't the thought I thought I had thought.

 

[Guest]

Ahh, you said 33%.

I've been waiting for months for someone to say a number, any number.

It seems more believable coming from someone who could have seen it in action on more than one instance.

Have a nice day.

PS don't mind me I'm just mumbling here.

 

[Raystonn]

I just looked at this article and it proves my point

What point is that?

The increase in performance by staying at PC800 while increasing FSB to 533 is great, but nothing compared to the increase that you see when pairing a 533MHz FSB with PC1066.

Well sure. That much would be obvious. I will take performance wherever I can get it. Keep the 4X RDRAM multiplier until you hit the overclocking limit. Then flip to 3X to see if this is your bottleneck and continue pushing up the processor's clockspeed. One you reach the maximum clockspeed you should then compare benchmarks at the maximum 4X speed and the maximum 3X speed to see which offers better performance for the applications you will be using.

I would just like to point out here that the bandwidth offered by RDRAM using the 3X multiplier is still greater than that offered by DDR-SDRAM with the 4:3 Memory:FSB ratio.

-Raystonn


= The views stated herein are my personal views, and not necessarily the views of my employer. =

 

[AMD_Man]

Well sure. That much would be obvious. I will take performance wherever I can get it. Keep the 4X RDRAM multiplier until you hit the overclocking limit. Then flip to 3X to see if this is your bottleneck and continue pushing up the processor's clockspeed. One you reach the maximum clockspeed you should then compare benchmarks at the maximum 4X speed and the maximum 3X speed to see which offers better performance for the applications you will be using.

Yes, as I said before, you're <b>not</b> going to <i>lose</i> performance from your CPU by lower the RAM back to PC800, you're just not going to gain the performance of PC1066.

AMD technology + Intel technology = Intel/AMD Pentathlon IV; the <b>ULTIMATE</b> PC processor

 

[eden]

You too find the performance jump that better?
When looking at the benchs, and comparing how an old Athlon 1.33 can pummel down a 1.6A with all the new RAM and FSB still is nowhere near the competition, I feel like it ain't the bandwidth that will save the P4. We're just putting more cream on the banana instead of putting cherries and other ingredients. I find this just sad, Intel is not talking about anything about adding new components to the CPU itself, but only focusing on RAM and FSB. Well yeah that's fine, 133MHZ is a big jump of FSB, but when that much is added plus 266MHZ of RDRAM, you'd figure more would come out. Imagine the prices now...
Ray you can enjoy this now, but in a month, I bet you no one will be looking this much at the P4 when the TBRED is out. This 0.13m is all the AthlonXP needed to push itself further and get back on track. Its overclocking will be almost as good as the P4, but less. However the performance will be tremendous compred to a 900MHZ overclock by the 1.6A, and its price will continue to be more justifiable. So don't comment on this, just wait a month and see why I am holding out on any comments on the 1.6A because I know it won't stay for long in the best list.

--
For the first time, Hookers are hooked on Phonics!!

 

[Raystonn]

comparing how an old Athlon 1.33 can pummel down a 1.6A with all the new RAM and FSB still

This is not even close to true.

Ray you can enjoy this now, but in a month

The future is unknown. It always is. Attempting to talk about it an any kind of objective fashion is pointless and meaingless. This thread is really about someone who is going to make a purchase and wants to know about current options.

However, if you really want to discuss the future, throw in 3GHz Pentium 4 processors with PC1200 RDRAM. Debating the future is really pointless.

This 0.13m is all the AthlonXP needed to push itself further and get back on track. Its overclocking will be almost as good as the P4

I disagree on both of these counts. The Athlon would need a core redesign to push up in clockspeed as quickly as the Pentium 4. I am sure the move to a small die size will allow higher clockspeeds. But I highly doubt they will be anywhere near the jumps being made by the Pentium 4. Only time will tell for sure. (To those who are bound to complain about clockspeed being meaningless, remember this: If you want to complain about that then I never want to hear you complain about the Pentium 4 wasting clocks. You cannot waste something that is meaningless.)

-Raystonn


= The views stated herein are my personal views, and not necessarily the views of my employer. =

 

[eden]

http://www.aceshardware.com/Spades/read.php?article_id=45000221
Look at 3d Studio Max. Where do you see the performance in that? Fact is the P4 at any given speed, will remain with "holes" in performance. If it had been given its components back, it would be pummeling the AXP at its huge speeds, but NOOOOO.... Why? I dunno, you work there, you tell me why... just why do they only focus on one thing like clock speed, rather than exploiting technology in various levels like AMD does to IPC and its subs.

--
For the first time, Hookers are hooked on Phonics!!

 

[eden]

I forgot to mention, the % up by the 0.13 die shrink for AXP will be as "scalable" as the P4. It of course will not mean it reaches P4 speeds, by all means it's just impossible for it, but it will scale the same as P4 like Matisaro explained. This means the AXP Tbred could possibly end up at 3GHZ scalable. If SOI does its job, it may even go further. I dunno of the future too man, but I do know AMD has a lot going on, and that Intel should do something more worth than just pumping up MHZ and FSB...

--
For the first time, Hookers are hooked on Phonics!!

 

[Matisaro]

comparing how an old Athlon 1.33 can pummel down a 1.6A with all the new RAM and FSB still



--------------------------------------------------------------------------------

This is not even close to true.

At stock ray, a 1.33ghz tbird on a kt266a chipset WILL beat a 1.6a p4. You are the one whos not even close to true there.

"The Cash Left In My Pocket,The BEST Benchmark"
No Overclock+stock hsf=GOOD!

 

[eden]

Wow I didn't even notice he said it wasn't true, about the 1.33 vs 1.6A... I thought he was referring to the 2GHZ or so...
Indeed you're the one who is wrong Ray, and you may check tons and tons of benchs, where the 1.33 is the king at any given day, even with 1.6 having the added 512K. Only when OCed does the 1.6 beat it, which is no longer a 1.6A ANYWAY.

--
For the first time, Hookers are hooked on Phonics!!

 

[Matisaro]

I disagree on both of these counts. The Athlon would need a core redesign to push up in clockspeed as quickly as the Pentium 4. I am sure the move to a small die size will allow higher clockspeeds. But I highly doubt they will be anywhere near the jumps being made by the Pentium 4. Only time will tell for sure. (To those who are bound to complain about clockspeed being meaningless, remember this: If you want to complain about that then I never want to hear you complain about the Pentium 4 wasting clocks. You cannot waste something that is meaningless.)

And you are wrong ray, the athlon WILL gain the same % of a clockspeed increase as the NW from the new process. The clockspeed jump from willamette to northwood is ONLY caused by the process shrink, the axps will get the SAME speed jump from the process shrink because it is a purely physical thing.

Now in sheer mhz the p4 will gain more, but %'s they will gain the same, simply because the p4 is designed to have a higher clockspeed.


"The Cash Left In My Pocket,The BEST Benchmark"
No Overclock+stock hsf=GOOD!

 

[Matisaro]

Yes, the % gain is the same, raystonn is wrong, or trying to be misleading, of course the tbred wont scale as fast in pure numbers as the NW, it will however gain more performance per clock, thus reaching an equilibrium.


For those tho thing the 1.6a is the best overclocker ever, the slowest grade tbred WILL do the same, and it will be unlockable to truely reach its heights.


As for tech out right now, the xp1600+ will with a good hsf overclock to 2000+ speeds most of the time, simply buy the chip and an xp333 chipset mobo, and run it at 166fsb, and you get EXTREME value. Perhaps not as much value as say a NW 1.6a@2.4ghz, but a very good value none the less.

"The Cash Left In My Pocket,The BEST Benchmark"
No Overclock+stock hsf=GOOD!

 

[AMD_Man]

At stock ray, a 1.33ghz tbird on a kt266a chipset WILL beat a 1.6a p4. You are the one whos not even close to true there.

Yes, true. Think about it.

A 1.7GHz P4 Willy was about as fast as a 1.33GHz T-Bird on an AMD760 (give or take a few percentage points for either).

Now, the KT266A improves system performance by ~10% overall over the AMD760. The Northwood improves performance by ~10% overall over the Willamette so, in fact, the Athlon 1.33GHz with a KT266A would still match a theoretical 1.7GHz P4 Northy.

AMD technology + Intel technology = Intel/AMD Pentathlon IV; the <b>ULTIMATE</b> PC processor

 

[Matisaro]

Yes, and with deductive logic, amd_man once again is correct.


I would even like to add, that on an amd760 chipset, the 1.33ghz would be at least as fast.(because the 1.33 is nativly as fast as a 1.7ghz willamette) and the nw improvement may or may not overcome the 100mhz lost.



"The Cash Left In My Pocket,The BEST Benchmark"
No Overclock+stock hsf=GOOD!

 

[eden]

What I do question question however is Ray's posts here.
Does he read THG or what?
I mean yea he is all Intel, but doesn't a person that works for processor company would know a bit about the outside?! I mean he realized just recently that the AXP uses DDR! Maybe he isn't reading the THG, as he also contradicts what we said about the 1.33GHZ being against a 1.7GHZ ...
Ray please read the THG articles. Although not the best, they always are simple enough to know what we look for.

--
For the first time, Hookers are hooked on Phonics!!