Sign in with
Sign up | Sign in
Your question

RDRAM PC1066 lower latency than PC800

Last response: in CPUs
Share
January 8, 2002 6:41:25 PM

Again AcesHardware did a quality job.

Matisaro, take a look at the memory latencies, you asked Raystonn for proof:
<A HREF="http://www.aceshardware.com/read.jsp?id=45000279" target="_new">http://www.aceshardware.com/read.jsp?id=45000279&lt;/A>

/Copenhagen - P4 Willamette 1700MHz@2109 MHz, Vcore 1.75V@2.20V on Abit TH7II-RAID.
January 8, 2002 6:50:32 PM

RDRAM latency is still shown as higher than DDR (Raystonn was wrong about the crossing point, maybe?), but you're right, RDRAM latency is getting lower, and it's definitely at a tolerable level.
Also of note is that DDR went down in latency as clock speed increased. This is also in contradiction to what Raystonn was saying. Interesting, but I'd like to see this info from more than just the one site.

Very good review, by the way. One of the best I've seen.

<font color=orange>Quarter</font color=orange> <font color=blue>Pounder</font color=blue> <font color=orange>Inside</font color=orange>
January 8, 2002 6:57:08 PM

<blockquote><font size=1>Svar på:</font><hr><p>Very good review, by the way. One of the best I've seen.<p><hr></blockquote><p>I totally agree.

/Copenhagen - P4 Willamette 1700MHz@2109 MHz, Vcore 1.75V@2.20V on Abit TH7II-RAID.
Related resources
January 8, 2002 7:53:09 PM

Fats,

Has anyone ever found out the truth to the SDRAM(Parallel) vs. RDRAM(Serial) latencies argument?

I still have not found any documentation stating increases or decreases. I have seen the new DDR II specs from <A HREF="http://www.jedec.org/DOWNLOAD/pub21/HotDDR/DDR_II_evolu..." target="_new">JEDEC,</A> but I am still not able to get those documentations.

Anyone have a link?

<b>"The events of my life are quite inconsequential.." - Dr. Evil</b> :lol: 
January 8, 2002 8:17:40 PM

The website above is the best I've seen so far. I'll be glad when someone finds something more in-depth.

<font color=orange>Quarter</font color=orange> <font color=blue>Pounder</font color=blue> <font color=orange>Inside</font color=orange>
January 8, 2002 8:36:23 PM

Quote:
RDRAM latency is still shown as higher than DDR (Raystonn was wrong about the crossing point, maybe?), but you're right, RDRAM latency is getting lower, and it's definitely at a tolerable level.

More benchmarks like this need to be performed. I said earlier that the latency levels of a PC1066 system would approximately the same as a PC800 system's latency multiplied by 0.75. This proves to be the case. However, in most other benchmarks comparing PC800 with DDR SDRAM, when you multiply the PC800 score by 0.75, the RDRAM system has less latency than the DDR SDRAM system. So something is odd in this benchmark.

Quote:
Also of note is that DDR went down in latency as clock speed increased. This is also in contradiction to what Raystonn was saying. Interesting, but I'd like to see this info from more than just the one site.

It will go down a bit, but nowhere near as much as RDRAM. When you increase the clockspeed of DDR SDRAM, more cycles of latency are introduced. It becomes less efficient. It does have more latency, but each cycle of latency takes less time because of the increased clockspeed. This results in less latency measured in real-time, but more latency measured in cycles. The RDRAM will always have the same latency measured in cycles, and thus much less latency measured in real-time as it increases in clockspeed.

-Raystonn


= The views stated herein are my personal views, and not necessarily the views of my employer. =
January 8, 2002 8:48:16 PM

Quote:
This results in less latency measured in real-time, but more latency measured in cycles.


Ah, that's the part I'm missing. Thanks.

<font color=orange>Quarter</font color=orange> <font color=blue>Pounder</font color=blue> <font color=orange>Inside</font color=orange>
January 9, 2002 3:36:44 AM

Raystonn ".... <b>This proves to be the case.</b> However, in most other benchmarks comparing PC800 with DDR SDRAM, when you multiply the PC800 score by 0.75, the RDRAM system has less latency than the DDR SDRAM system. <i>So something is odd in this benchmark.</i>"

so what proves it, if this benchmark says otherwise.

Remember, it's the trolls' jobs to state psuedo-fact cum fictional statments.
Please back this up with proof of the proving.
thanx

I spilled coffee all over my wife's nighty... ...serves me right for wearing it?!?
January 9, 2002 5:34:50 AM

This benchmark clearly shows that PC1066 RDRAM has approximately 3/4 the latency of PC800 RDRAM. Just look at the figures. Now take that 3/4 figure and go look at other benchmarks, just about anywhere else, that compare PC800 RDRAM latency with DDR SDRAM. Do the math, multiplying the PC800 score by 0.75, and you will see that in just about every case the PC1066 score will be lower (better) than the DDR SDRAM score.

For some reason this particular benchmark has latency figures for PC800 RDRAM and DDR SDRAM that differ from the average benchmark. They are much further apart than normal. This may have something to do with the way he configured the RDRAM system versus the DDR SDRAM system. But however he chose to configure it, that would not change the fact that PC1066 latency is still about 3/4 of PC800 latency. Use that algorithm and check other benchmarks until we get more actual PC1066 benchmarks.

-Raystonn


= The views stated herein are my personal views, and not necessarily the views of my employer. =
January 9, 2002 11:15:00 AM

I never said rdram latency did not go down as clock speed increased, I said it was still above ddr latency. And I was right, thanks for the proof.

"The Cash Left In My Pocket,The BEST Benchmark"
No Overclock+stock hsf=GOOD!
January 9, 2002 11:17:22 AM

Quote:
It will go down a bit, but nowhere near as much as RDRAM. When you increase the clockspeed of DDR SDRAM, more cycles of latency are introduced. It becomes less efficient. It does have more latency, but each cycle of latency takes less time because of the increased clockspeed. This results in less latency measured in real-time, but more latency measured in cycles. The RDRAM will always have the same latency measured in cycles, and thus much less latency measured in real-time as it increases in clockspeed.



And as any ee will tell you, the only latency which really matters in a system is REAL TIME LATENCY. Wasted cycle latency is meaningless.

"The Cash Left In My Pocket,The BEST Benchmark"
No Overclock+stock hsf=GOOD!
January 9, 2002 3:20:42 PM

Quote:
And as any ee will tell you, the only latency which really matters in a system is REAL TIME LATENCY. Wasted cycle latency is meaningless.


But this does show that RDRAM (real time) latency will get lower than DDR's in the near future.

<font color=orange>Quarter</font color=orange> <font color=blue>Pounder</font color=blue> <font color=orange>Inside</font color=orange>
January 9, 2002 4:28:30 PM

Simply take a look at other benchmarks. The difference in latency between the DDR SDRAM and the RDRAM is higher than normal in this benchmark. In most other benchmarks, a cut down to 3/4 the latency on the RDRAM system will make it the lowest latency solution. Remember, one benchmark does not a theory prove.

-Raystonn


= The views stated herein are my personal views, and not necessarily the views of my employer. =
January 10, 2002 11:39:40 AM

So, dig up more benchmarks, your the one making the claim rayy, that pc1066 has lower latency than ddr, and the only link brought forth so far disagrees with you. Burden of proof is on you my friend.

"The Cash Left In My Pocket,The BEST Benchmark"
No Overclock+stock hsf=GOOD!
January 10, 2002 5:24:05 PM

I am not really interested in proving it to you. If you want DDR SDRAM then go use it. I will stick with what I know has lower latency and we will both be happy. For the moment there are no other benchmarks using PC1066 RDRAM.

-Raystonn


= The views stated herein are my personal views, and not necessarily the views of my employer. =
January 11, 2002 4:20:31 AM

You, rayy, are making the claim, if you cant prove it, dont talk about it.


I am not asking you to jump through a goddamn hoop, I merely wished links to benchmarks to prove your intel biased claims that pc1066 has lower latency than ddr sdram.

If you cant do it, and if you get snotty when asked to do it, then shut up about it.


I dont give a crap about what you know, I give a crap about what you can prove, you ask me to link for proof about some of my claims, and yet when I ask the exact same from you, you berate me and act as if my request is meaningless. Hypocrasy, plain and simple.

"The Cash Left In My Pocket,The BEST Benchmark"
No Overclock+stock hsf=GOOD!
January 11, 2002 7:47:32 AM

As I have said over and over, there are no other benchmarks showing the latency of PC1066 RDRAM yet. Either wait or perform the tests yourself. Or you could simply pull up benchmarks of PC800 latency versus DDR SDRAM and multiply the PC800 scores by 0.75. That will work just as well. There are plenty of those benchmarks around. Go pick one at random.

-Raystonn


= The views stated herein are my personal views, and not necessarily the views of my employer. =
January 11, 2002 4:01:38 PM

Quote:
As I have said over and over, there are no other benchmarks showing the latency of PC1066 RDRAM yet. Either wait or perform the tests yourself. Or you could simply pull up benchmarks of PC800 latency versus DDR SDRAM and multiply the PC800 scores by 0.75. That will work just as well. There are plenty of those benchmarks around. Go pick one at random.



So let me get this straight ray, the ONLY benchmark for pc1066 rdram says that it has higher latency than ddrsdram, but you say something is off with this only test, and it is wrong, hmmm, if you say so.

"The Cash Left In My Pocket,The BEST Benchmark"
No Overclock+stock hsf=GOOD!
January 11, 2002 5:34:43 PM

Quote:
So let me get this straight ray, the ONLY benchmark for pc1066 rdram says that it has higher latency than ddrsdram, but you say something is off with this only test, and it is wrong

Well, go take a look at benchmarks comparing DDR SDRAM and PC800 latencies. You will notice the gap between their latency figures are not as high (as a percentage) as in this benchmark that also shows PC1066. Something about this benchmarker's hardware likely is different from all the rest.

-Raystonn


= The views stated herein are my personal views, and not necessarily the views of my employer. =
January 11, 2002 5:34:45 PM

(Removed an erroneous double-post. How do these get here when you do not double-click. Hmmm.)
January 11, 2002 7:17:30 PM

Quote:
Well, go take a look at benchmarks comparing DDR SDRAM and PC800 latencies. You will notice the gap between their latency figures are not as high (as a percentage) as in this benchmark that also shows PC1066. Something about this benchmarker's hardware likely is different from all the rest.

-Raystonn



There is something changed, that benchmark is using kt266a, I have not been able to find latency tests on anything faster than an amd760. Regardless, the fact still remains, ddrsdram has lower latency than pc1066.

"The Cash Left In My Pocket,The BEST Benchmark"
No Overclock+stock hsf=GOOD!
January 11, 2002 8:24:29 PM

I read somewhere that the ASUS P4T-E was able to run 150MHz FSB corresponding to PC1200. ASUS should be using another FSB design (not following the reference design) and also they are using 533MHz rated TI DRCGS. Surely the latency of PC1200 would have lower latency than DDR SDRAM, if the scalability holds true.


/Copenhagen - P4 Willamette 1700MHz@2109 MHz, Vcore 1.75V@2.20V on Abit TH7II-RAID.
January 12, 2002 1:47:17 PM

Another thing which bears mentioning, is all of these latency tests are being done on dual channel rdram boards. And when ddr becomes dual channel its latency will decrease as well.

"The Cash Left In My Pocket,The BEST Benchmark"
No Overclock+stock hsf=GOOD!
January 12, 2002 2:54:35 PM

Quote:

Another thing which bears mentioning, is all of these latency tests are being done on dual channel rdram boards. And when ddr becomes dual channel its latency will decrease as well.

I disagree there Matisaro. A Dual-channel configuration only improves bandwidth not latency.

AMD technology + Intel technology = Intel/AMD Pentathlon IV; the <b>ULTIMATE</b> PC processor
January 12, 2002 3:10:33 PM

You can disagree but youre wrong heh.


Having dual channel ram doubles the likelyhood that the requested memory lies within an open page, removing the latency associated with closing a page and opening another one.

There are more reasons why dual channel systems are lower latency, but that is the main one.

(ps: bandwith is not why p4 has dual channel rambus, it was for imrpoved latency, iirc the origional p3 rambus chipset was single channel and its latency was TERRIBLE, extremely so. The dual channel alleviated much of this.(and yes, both used pc800).

"The Cash Left In My Pocket,The BEST Benchmark"
No Overclock+stock hsf=GOOD!
January 12, 2002 3:12:14 PM

"Only Intel's recent high-end RDRAM chipsets i840, i850 and i860 are using two memory banks (in this case called dual channel Rambus). Intel had to do this to ensure reasonable memory latency, because the latency of a single Rambus channel is so high that system performance suffers significantly"


<A HREF="http://www6.tomshardware.com/mainboard/01q2/0106041/nfo..." target="_new">http://www6.tomshardware.com/mainboard/01q2/0106041/nfo...;/A>





"The Cash Left In My Pocket,The BEST Benchmark"
No Overclock+stock hsf=GOOD!
January 12, 2002 3:15:38 PM

Furthermore, from the same link above.


"NVIDIA's nForce combines the good of both worlds. It is using the low-latency SDRAM (SDR or DDR) and offers two memory banks, which doubles the bandwidth and halves the latency. The nForce420 IGP comes with two 64-bit wide memory controllers, adding up to a 128-bit wide memory interface. Each of the two controllers can work independently from the other to ensure low latency. Each bank can have different memory sizes as well as memory types (!!!), which is a major difference to Intel less intelligent dual channel Rambus memory controllers of i840/i850/i860, which require that two banks are equipped with identical RDRAM RIMMs. "

I want to see some latency tests with the nforce board, if it has half the latency(im sure tom is way paraphrasing that one) then it would further BLOW AWAY, pc1066, and dual channel ddr ram solutions will ensure ddr-sdram has the lowest latency for some time to come.

"The Cash Left In My Pocket,The BEST Benchmark"
No Overclock+stock hsf=GOOD!
January 12, 2002 3:17:15 PM

Quote:
I want to see some latency tests with the nforce board, if it has half the latency(im sure tom is way paraphrasing that one) then it would further BLOW AWAY, pc1066, and dual channel ddr ram solutions will ensure ddr-sdram has the lowest latency for some time to come.



Which btw, was my only point on the subject....and I was right.

"The Cash Left In My Pocket,The BEST Benchmark"
No Overclock+stock hsf=GOOD!
January 12, 2002 3:31:46 PM

I still don't understand how multiple channels can improve latency! It doesn't make sense!

AMD technology + Intel technology = Intel/AMD Pentathlon IV; the <b>ULTIMATE</b> PC processor
January 12, 2002 4:46:35 PM

When you access data from a ram module(not sure on rambus's specifics) you must open a page, this reqquires cycles of latency.Each ram type has a different latency for opening and closing pages also Each ram module can have a certain number of open pages when a data request falls within an already open page, the latency caused by closing the current page and opening a new one is removed from the equasion. Having 2x the ram modules simply doubles the number of allowed open pages, reducing latency in many cases adding up all the instances when data was stored in an open page gains alot in net latency reduction.


There are other reasons, which function similar to the above reasons why dual channel systems have lower latency than single channel systems. Furthermore IIRC the reasons dual rambus systems have lower latency are universal to dual sdram and dual ddrsdram systems, so they both would benifit from them.


Does anyone else who knows about ram access and latency have any other means a dual system would have lower latency than a single system.

"The Cash Left In My Pocket,The BEST Benchmark"
No Overclock+stock hsf=GOOD!
January 12, 2002 4:55:49 PM

<A HREF="http://arstechnica.com/paedia/r/ram_guide/ram_guide.par..." target="_new">http://arstechnica.com/paedia/r/ram_guide/ram_guide.par...;/A>



<A HREF="http://arstechnica.com/paedia/r/ram_guide/ram_guide.par..." target="_new">http://arstechnica.com/paedia/r/ram_guide/ram_guide.par...;/A>


an interesting snippet.


"While the long, thin RAMBUS channel pumps a lot of bandwidth over a small number of traces, it's nonetheless one of RAMBUS' most controversial features. Operating at up to 400MHz, it's very fast, and since it makes for a minimal number of signal traces that have to be etched into the motherboard, it's simpler overall than SDRAM's interleaving of data buses. However, it still carries with it some drawbacks. One problem with the long, fast bus is its effect on cost. Some of the savings in cost that RAMBUS gets from using fewer traces are cancelled out by the fact that the RAMBUS channel is a long series of wires that have to run at a whopping 400MHz. To get the bus speed up that high, the board has to be manufactured to a very high standard of quality in order to reduce noise, stray capacitance, variations in line impedance, and other problems associated with rising bus speeds. In some cases, you may even have to add more layers to the motherboard just to be able to provide a clean enough signal."


^ heh




read the articles on how ram works, very infomative, and i think you will better understand why a dual channel system has dramatically lower latency than a single channel system.


"The Cash Left In My Pocket,The BEST Benchmark"
No Overclock+stock hsf=GOOD!
January 12, 2002 7:06:37 PM

exquizit site!:) 
nice one.thx.

<font color=green>
*******
*K.I.S.S*
*(k)eep (I)t (S)imple (S)tupid*
*******
</font color=green>
January 12, 2002 7:37:24 PM

amazing links, thanks man!

AMD technology + Intel technology = Intel/AMD Pentathlon IV; the <b>ULTIMATE</b> PC processor
January 12, 2002 8:58:13 PM

Ars-technica ROCKS.

I personally rank it the best pc/tech site there is, it updates a tad slow, but its all about the hardcore tech info.


Also, their forums rock, much more technologically in depth.(they have their fuggers and meltdowns though).

"The Cash Left In My Pocket,The BEST Benchmark"
No Overclock+stock hsf=GOOD!
January 13, 2002 4:34:51 AM

"... I want to see some latency tests with the nforce board, ... "

Cachemem result for Athlon 1800+, K7N420 @ 133/266/66, 2x Crucial 256MB CL 2.5 (set to CL2):

<pre>Cache size/Memory speed info tool 2.6MMX - (c) 1999-2001, LRMS - DJGPP compiled
CPUID support detected... 'AuthenticAMD' with FPU TSC MMX
Family=6 Model=6 Step=2 Type=0 Chipset (Vendor/Device ID(Rev)): 10DE/01A4(B2)
CPU clock: 1537.0 MHz
Using 32MB physical memory block (alignment = 32)
Bandwidth - MMX linear access test... Read/Write/Copy (MB/s)

*** THIS TEST RESULT EDITTED OUT DUE TO FORMATTING PROBLEM UNDER DOS BOX ***

Latency - Memory walk tests... ("pointer chasing")
Null size: 4 cycles 1 cycles (overhead 65 cycles)
steps: 4 8 16 32 64 128 256 512 1k 2k 4k (bytes)
Block of 1KB: 4 4 4 4 4 4 4 4 - - - cycles
Block of 2KB: 4 4 4 4 4 4 4 4 4 - - cycles
Block of 4KB: 4 4 4 4 4 4 4 4 4 4 - cycles
Block of 8KB: 4 4 4 4 4 4 4 4 4 4 4 cycles
Block of 16KB: 4 4 4 4 4 4 4 4 4 4 4 cycles
Block of 32KB: 4 4 4 4 4 4 4 4 4 4 4 cycles
Block of 64KB: 4 4 4 4 4 4 4 4 4 4 4 cycles
Block of 128KB: 4 5 9 10 20 20 20 20 20 20 20 cycles
Block of 256KB: 4 5 9 10 20 20 20 20 20 20 20 cycles
Block of 512KB: 10 19 32 67 157 161 161 162 162 163 167 cycles
Block of 1024KB: 10 19 32 68 157 161 161 161 162 163 167 cycles
Block of 2048KB: 10 19 32 62 158 162 162 164 165 170 182 cycles
Block of 4096KB: 10 19 32 64 157 162 162 164 167 172 186 cycles
Block of 8192KB: 10 19 32 63 158 162 163 164 167 173 188 cycles
Block of 16384KB: 10 19 32 63 158 162 163 164 167 174 189 cycles
Block of 32768KB: 10 19 32 64 157 162 163 164 168 174 189 cycles
Done.
This system appears to have 2 cache levels (enabled).
L1 cache (64KB) speed (MB/s): Read=13355.7, Write=11048.3
L2 cache (256KB) speed (MB/s): Read=4503.2, Write=4681.1
Main memory speed (MB/s): Read=1117.8, Write=742.1

Press any key...
</pre><p><P ID="edit"><FONT SIZE=-1><EM>Edited by MadCat on 01/13/02 01:51 PM.</EM></FONT></P>
January 13, 2002 4:53:36 PM

I updated the latency test result for the nForce. I was suppose to run the test under MSDOS. Gives more consistent results.
January 14, 2002 1:51:51 AM

If im not reading that chart wrong, it is showing the nforce latency KICKING the crap out of the pc1066's latency....interesting.

"The Cash Left In My Pocket,The BEST Benchmark"
No Overclock+stock hsf=GOOD!
January 14, 2002 3:30:35 AM

Where are people getting 1066rdram or even getting the boards that support it. It's doubltful overclocked rdram 800 will be excatly the same performance as 1066rdram which uses a 32bit wide bus instead, unless they changed that.
!