P4 at 2.2 and memory

bhdavis

Distinguished
Jan 30, 2002
7
0
18,510
I remaine a bit confused over the memory issue with the P4 2.2 chip.

I had thought that RD RAM was supposed to be a dead end. I know it has been used with the P4's from the beginning, but I thought it had been panned as overpriced and a "dead end". Yet, here it is still being available for use on the new Northwood core.

So, from a performance standpoint is it the memory to put in a new system. Or would DDR-SDRAM be better? If I understood what I've read on Tom's Hardware earlier today corredtly, SDRAM is not the answer from the standpoint of performance.

If the RD-RAM is preferred over the DDR-SDRAM, then by a lot, or does it only minimally improve performance?

Thanks,
BH Davis
 

bhdavis

Distinguished
Jan 30, 2002
7
0
18,510
Yes, I know. If you look at the earlier post you'll see a couple of typos. I fixed them and hit the wrong submit button and tried to STOP the submission. I guess I missed.
 

FatBurger

Illustrious
you double posted

And your point would be...?


The i850 and SiS645 chipsets are approximately the same performance. However, if you feel brave, you can solder on difference clock generators with the i850 and move to a 133 bus. Only do that if you get a Northwood, though.

I'd recommend the i850 chipset. It's (hopefully) about to be replaced by Intel in the next 2-3 months with a new RDRAM chipset with PC1066, but we'll see.

I recommend the Abit TH7-II or the Asus P4T-E for a motherboard.

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

FUGGER

Distinguished
Dec 31, 2007
2,490
0
19,780
Right now RDR is a few bucks more than DDR RAM

I have tested just about every combination of chipset, video card, memory that is possible.

SDR should be put out to pasture.

DDR is trying to makup for SDR, SIS645 chipset is strong memory perf but lacks everywhere else. i845 can clock very high but lacks the memory bandwidth when it really needs it. and Via doesnt even have rights to make a P4 chipset in the first place. Avoid Via P4X266A and SiS645 till further notice.

RDR best choice for maximum performace from a P4

The Asus P4T-E and Abit TH7-II are two of the best P4 boards available. Geforce 3 is better for this platform than the Radeon 8500. I dislike the ATI R8500 for poor drivers and non functional advertised features.

I secured some web space recently, Ill be posting cool stuff for you girls soon.
 

bhdavis

Distinguished
Jan 30, 2002
7
0
18,510
So, does one wait for the new chipset with the faster memory, or go ahead and put a system together now? I mean, there's always somnething better and faster just around the corner.

I've read some reviews that liked the Via chipset better than the i845 even though there is the legal issue of rights to make the chipset.

BH Davis
 

FatBurger

Illustrious
Via won all three lawsuits with Intel, therefore they DO have legal rights to create the chipset. At least, the courts say they do, and that's what matters.

The P4X266 is faster than the i845D, but not as stable.

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

IIB

Distinguished
Dec 2, 2001
417
0
18,780
I'm not holding my breath for this PC1024 RDRAM + 533mhz bus.
simply because a duel channel a DDR-SDRAM chipset should easily out perform it...

333 MHz DDR models should offer
2(Duel models) * 333(MHz) * 8(byte) = (~5300MB/s) 5.3GB/s

1024 MHz RDRAM Models should offer:
2 * 1024(MHz) * 2(byte) = (~4000MB/s) 4GB/s
 

FatBurger

Illustrious
Show me a roadmap with a dual-channel PC2700 chipset.

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

IIB

Distinguished
Dec 2, 2001
417
0
18,780
cant find any roadmaps... however more then once i've heard that VIA is going to make duel channel DDR-SDRAM chip-sets.

this one is from XbitLabs:
About QBM to be Supported by System Chipsets [8:52 pm]

We used to mention that Kentron failed to win the chipset makers’ hearts and persuade them to implement QBM technology support in the existing chipsets. The company’s officials sound still optimistic, however. They hope to find common language with the manufacturers this year. According to Robert Goodman, Kentron CEO, the first products with QBM support are to become available by the end of the year. He has also confirmed the info about QMB DIMM modules to start shipping in the Q’1 of 2002. We wonder whom they’ll be shipped to. : -) <b>Neither VIA nor SIS yet plan to produce chipsets with QBM support, but they are going to make chipsets supporting regular dual-channel access o PC2700 DDR SDRAM.
Then why using the QBM technology at all? Dual-channel PC2700 DDR SDRAM will have enough bandwidth (5.4GB/sec) for Quad Pumped Bus 533MHz (4.2GB/sec).</b>
By the way, at the presently held Platform Conference Mr. Goodman said that Kentron positioned QBM SDRAM as a competitor to RDRAM, but not DDR SDRAM. Of course they do QBM modules are manufactured on the basis of standard DDR SDRAM DIMMs, and RDRAM DIMMs are an absolutely different story, which seems to have a happy end this year.
and i also heard that duel channel DDR-SDRAM chipset for P4 are comming q'2-3...

i know those duel channal DDR-SDRAM chipsets can be made.
i have the Hammer architecture white paper is states (pg4 under "integrated DDR DRAM memory controler"):
"The Hammer microarchitecture incorporates a Dual Channel DDR-DRAM controller with a 128-bit interface"

oh F*uck... here is a problem... the Pentium 4 bus is a 64bit bus being that evry DDR-SDRAM model is 64 bit maybe it will not be able to work Dual channel with DDR-SDRAM... or any 128bit memory controler.
Hmmm... dunnu about this maybe VIA or SIS can make it work...

DAMN! i want my Hammer now! :)
 

Raystonn

Distinguished
Apr 12, 2001
2,273
0
19,780
You should note that:

A) Any memory bandwidth offered to the processor that is over the bandwidth available on that processor's FSB will go to waste.

B) There is no dual-channel DDR SDRAM chipset on any roadmap.

C) Bandwidth is measured per pin, not per 'channel'. Channels are an arbitrary term. Pins are what take up space on the motherboard. i850 uses 32-pins total for RDRAM channels. A 'dual-channel DDR SDRAM' chipset would use 128-pins total for DDR SDRAM channels. This is less bandwidth per pin. For the same amount of real-estate on the motherboard you could instead use 128-pins for RDRAM and get much more memory bandwidth.

D) PC1066 RDRAM will have less latency than your DDR SDRAM solution. (PC1066 RDRAM has exactly 3/4 the latency of PC800 RDRAM. DDR SDRAM does not scale in such a linear fashion.)

-Raystonn


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

FUGGER

Distinguished
Dec 31, 2007
2,490
0
19,780
Via created the P4X266 chipset without the tech specs. I t was created thru reverse engineering, quite a accomplishment I must admit but it was done without Intels blessing.

Via chipset failed my compatability checks, Many anomilies during testing, and poor overclocking.

Via was stronger than the SiS645 in everything but memory bandwidth.

Intel needs competition in this arena to keep prices down.
 

AMD_Man

Splendid
Jul 3, 2001
7,376
2
25,780
Fugger, in any case, Intel prcessor + anything other than an Intel chipset is stupid because you can get Intel Chipsets with RDRAM, DDR and SDRAM. The options are all there from Intel so why lower your standards to the competition. Intel made the P4 and so, obviously, they'd have the best chipsets.

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

FatBurger

Illustrious
Raystonn: <A HREF="http://www.aceshardware.com/read.jsp?id=45000279" target="_new">This</A> chart supports your number of 25% less latency, by the way. Measured in clock cycles, not real time. However, it does not seem to support your statements that SDRAM gets higher latency when clock speed increases.

FUGGER: I didn't realize that it was done through reverse engineering. Via bought a company with a P4 chipset license (can't remember who), so I assumed they gained documentation that way.

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

Raystonn

Distinguished
Apr 12, 2001
2,273
0
19,780
Memory latency expressed as a number of memory clock cycles actually remains exactly the same for RDRAM. The clock cycles being used as the unit value there are processor clock cycles. This actually gives you the results in real-time if you convert from processor cycles to nanoseconds. The clockspeed of the Pentium 4 processor is 2.0GHz. Thus, 2 clocks occur every nanosecond. (A nanosecond is a billionth of a second.) The clockspeed of the Athlon is 1.667GHz. Thus, 1.667 clocks occur every nanosecond.

Translating that table to real-time latency, we get:

2.000GHz Pentium 4, Asus i850 motherboard, PC800 RDRAM: 150.5 nanoseconds
2.000GHz Pentium 4, Asus i850 motherboard, PC1066 RDRAM: 116.5 nanoseconds
1.667GHz Athlon (XP 2000+), ASUS KT266A motherboard: 121.8 nanoseconds

Now it begins getting interesting, huh?

-Raystonn


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

lhgpoobaa

Illustrious
Dec 31, 2007
14,462
1
40,780
so any figures for PC2700 latency? either with the athlon (xp333/sis735/kt333a/sis745) or the p4 (sis645)???

The lack of thermal protection on Athlon's is cunning way to stop morons from using AMD. :)
 

FatBurger

Illustrious
So how do you explain the latency differences between Athlon and P4?

There's a thread partially about this <A HREF="http://forumz.tomshardware.com/hardware/modules.php?name=Forums&file=viewtopic&p=33896#33896" target="_new">here</A> if you would like to comment.

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

Raystonn

Distinguished
Apr 12, 2001
2,273
0
19,780
To address the issue on the efficiency of latency on SDRAM versus RDRAM when increasing memory speeds (and thus bandwidth), let us take a look at some figures based on the same chart.

We have seen that when we increase the bandwidth of RDRAM by 33% (PC800 to PC1066), latency tests finish approximately 25% more quickly. This is shown by the PC800 memory latency test taking nearly 33% longer than the PC1066 test. This ratio of 4/3 (PC800/PC1066) translates to 3/4 (PC1066/PC800), telling us that PC1066 has 3/4 the latency of PC800.

Now look at these figures:
2.0GHz Pentium 4, SuperMicro i845 motherboard, PC133 SDRAM: 258 clock cycles, or 129 nanoseconds.
2.0GHz Pentium 4, ASUS i845D motherboard, PC2100 DDR SDRAM: 230 clock cycles, or 115 nanoseconds.

These figures tell us that when we increase the bandwidth of SDRAM by 100% (doubling it), latency tests finish approximately 11% more quickly. This is shown by the PC133 memory latency test taking nearly 12% longer than the DDR266 [PC2100] test. This ratio of 129/115 (PC133/DDR266) translates to 115/129 (DDR266/PC133), telling us that DDR266 [PC2100] has 89% (or 11% less) the latency of PC133.

To help explain the 33% versus 25% thing, I will give an analogy. Point A and Point B lie 45 miles apart. If a vehicle is travelling at 45mph, it will take an hour to travel between the points. If that vehicle gains a 33% speed boost it is now travelling at 60mph. This vehicle is now capable of travelling a full 60 miles in that same hour. However, the two points have not changed distances. It only needs to travel 45 miles. It will be able to go from Point A to Point B now in only 45 minutes instead of an hour. 45 minutes is 25% less time than an hour. Thus, with a 33% boost in speed, it is capable of travelling between any two points in 25% less time.

This is how latency in your system works. When you speed up memory by 33%, you should nominally see 25% less latency. When you speed up memory by 100%, you should nominally see 50% less latency.

To sum it up, when you double the speed of SDRAM, its latency only drops by 11% instead of the expected 50%. It is not scaling in a linear fashion. When you increase the speed of RDRAM by 33%, you get the full 25% reduction in latency (plus or minus a couple points for system anomolies.) This can only be explained by the fact that SDRAM has more memory cycles of latency as you increase its speed, whereas RDRAM retains exactly the same number of memory cycles of latency.

RDRAM simply scales much better than SDRAM.

-Raystonn


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

Raystonn

Distinguished
Apr 12, 2001
2,273
0
19,780
I have not seen any numbers for PC2700 DDR SDRAM. If you see any, please link to it here and let us know.

Thanks,

-Raystonn


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

FatBurger

Illustrious
Now look at these figures:
2.0GHz Pentium 4, SuperMicro i845 motherboard, PC133 SDRAM: 258 clock cycles, or 129 nanoseconds.
2.0GHz Pentium 4, ASUS i845D motherboard, PC2100 DDR SDRAM: 230 clock cycles, or 115 nanoseconds.

These figures tell us that when we increase the bandwidth of SDRAM by 100% (doubling it), latency tests finish approximately 11% more quickly. This is shown by the PC133 memory latency test taking nearly 12% longer than the DDR266 [PC2100] test. This ratio of 129/115 (PC133/DDR266) translates to 115/129 (DDR266/PC133), telling us that DDR266 [PC2100] has 89% (or 11% less) the latency of PC133.

Is SDRAM-->DDR-SDRAM really a fair comparison, though? It's not doubling in clock speed in the same way that PC800 to PC1066 is a 33% increase.

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

Raystonn

Distinguished
Apr 12, 2001
2,273
0
19,780
It is in fact doubling the clock speed in the SDRAM itself. Data is transferred twice per memory clock. One data transfer must be complete before the next can begin. Thus, data transfers are in fact taking place in half the time between memory and the processor.

-Raystonn


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

lhgpoobaa

Illustrious
Dec 31, 2007
14,462
1
40,780
im lookin but not findin...
its hard enough finding decent benchmarks for PC2700, let alone latency timings et al.

on a good note the leadtek 7350KDA looks very nice for me :) :) :) 166fsb + PC2700. spiffy!

The lack of thermal protection on Athlon's is cunning way to stop morons from using AMD. :)
 

mala

Distinguished
Oct 12, 2001
45
0
18,530
No, no, no. You can't do that.
You have to compare 100MHz SDRAM to 133MHz SDRAM in this case. DDR200 is not 2x as good as PC100, especially not in latency. The interface transfers data 2 times/cycle, but the dram-core is as laggy as always.(also true for RDRAM)

We are talking of increases in frequency here, aren't we?
If you double the frequency of any type of memory and keep the timings constant the latency excl. system latency will be halved.

Double the frequency and constant timings means that the people actually doing the DRAM-cells (used in all RAM) have to make their CAS, RAS et.c. complete in half the time. It doesn't matter if RAMBUS or JEDEC designs the memory module.
 

Raystonn

Distinguished
Apr 12, 2001
2,273
0
19,780
You have to compare 100MHz SDRAM to 133MHz SDRAM
I would be willing to do that if someone could find a single set of numbers for both PC100 and PC133 or for both PC2100 and PC1600. Thusfar I have not seen anything like this.


DDR200 is not 2x as good as PC100
Sure it is. The memory itself is operating at a frequency of 200MHz. The interface (bus) from the memory to the chipset's memory controller is operating at 100MHz, but transmitting twice per clock. If DDR is only marginally better than SDR in terms of latency, then QDR would be only marginally better than DDR in terms of latency. Latency would really be hurting it. Meanwhile RDRAM is scaling up using frequency...


If you double the frequency of any type of memory and keep the timings constant the latency excl. system latency will be halved.
Right. However, they were unable to keep the timings constant. SDRAM simply cannot keep up. Its parallel nature causes more problems and makes it more difficult to ensure no signal conflicts.

Feel free to get us some latency figures for DDR200 and DDR266, or PC100 and PC133. (Both must be from the same set of benchmarks on the same hardware.)

-Raystonn


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