DDR-RAM or RD-RAM

[Guest]

I was looking into Sisoft's Sandra benchmarking utility one night, and I checked out how PC2100 DDR-RAM went up against RD-RAM, and I was disgusted by the results. I thought PC2100 DDR-RAM could have 2.1GB/sec bandwidth, but in there, I saw that AMD760 paired up with PC2100 DDR-RAM gave only 700MB/sec bandwidth while Intel 850 chipset paired with PC600 RDRAM got 1.05GB/sec bandwidth! Can anyone explain why DDR-RAM got such a sucky number? I had high hopes for DDR-RAM.

 

[rcf84]

well the Athlon only uses so much bandwith. Remember the P4 has a 400mhz FSB compared the 266mhz FSB of the athlons.

Check out my RIG <A HREF="http://www.geocities.com/rcf84/home.html" target="_new">Rcf84's celeron machine</A>

 

[Crashman]

I think your right in your reasoning.

Suicide is painless...........

 

[Guest]

But when I look at Tom's Hardware Reviews, I don't understand how the heck does the Athlon beat the P4 in framerates in games like Unreal Tournament and 3DMark2000. With a lower bandwidth in memory, how the heck does the 266MHZ FSB Athlon beat the 400MHZ FSB Pentium IV when the games framerates are supposed to be quite dependent on your system's RAM bandwidth and graphics card ability? I think the Pentium IV is not that good after all, except for the fact that it has SSE2 instructions which make video-editting really fast compared to the Athlon.

 

[TheSandman]

Due to the p4 design it is an extreamly memory bandwidth dependant procesor where the athlon and the p3 ar no where near as dependant this makes the effect of that bandwidth gap insugnificant where if the p4 did not have that huge memory flow it would be cripled.

SANDMAN
LEAN MEAN KILLIN MACHINE

 

[Guest]

so are you suggesting that if the bandwidth of the P4 was even lower thatn 1.05GB/s, it'd practically be worthless?
And if future chipsets are able to take better advantage of the DDR-RAM, would it make the Athlon even faster?

 

[mpjesse]

Well, it's not really an issue with the chipsets. It's an Issue with the Athlon- like someone said earlier. It only has one path between the memory and chipset- and it runs at 133mhz (266DDR). The P4 has 4 paths STRAIGHT to the RDRAM and they run at 100mhz a piece- that's roughly 2-3 times the bandwidth. Now, from my understanding, the Sledgehammer has a similar design to that of the P4 (in terms of memory bus). That's when you'll see a major jump in bandwidth with DDR.

Ultimately RDRAM will fail. It has HORRIBLY high latencys and totally depends on the bandwidth of the processor. The technology is not that impressive. Case and point- RDRAM with the P3 is useless- there's NO gain in performance. DDR SDRAM is already showing a 10 percent increase in performance with the Athlon with NO changes to the Athlon's memory bus.

So, I guess your answer is yes. If the P4 didn't have enourmous bandwidth capabilities and SSE2 it'd be worthless. As for the Athlon taking FULL advantage of DDR-SDRAM- it probably never will. UNLESS they totally redesign the memory bus. That's when the Clawhammer/Sledgehammer steps in.

-MP Jesse

"Signatures Still Suck"

 

[TheSandman]

Yes that about sums it up.

SANDMAN
LEAN MEAN KILLIN MACHINE

 

[Guest]

HI,
you are generally right about some of the things you said,
BUT the matency rumor about rambus is grossly misreported by many who do not understand the overall picture about the P4 and rambus relationship..

a few interesting facts that the media does not report and tom missed as well..

the clock latency on rambus 800 is CL 2,
DDR is 2.5-3 depending..

also, as you said so well, the rambus has 4 paths,
and is clock doubled so this helps compensate for latency..

also, RAMBUS on the chipset 850 which has eveerything to do with memory performance, chipsets usually do,,
is a dual channel design which means they precharge
and can operate concurrently, so as to make rambus much faster..

lastly the P4 has a very deep pipeline and efficient cache
over the athlon that takes rambus design into account and actually compensates for rambus by its ability to reorder and reexecute instruction depending on the situation.

to give you an IDEA, I just ran our test P4 1.5 machine
on MEMTACH and SANDRA benchmarks,
and the DDR with athlon 1.2

in sandra athlon gets 500 INTEGER and 700 FP with PC2100 ddr
P4 gets 2200 or over 4 times faster in integer memory
and 4400 in FP memory with sse 2 enabled
or over 6 TIMES faster

in MEMTACH the results are similiar

RAMBUS scores over 1000 whereas the Athlon machines
are in the 300's

one of the most respected programs that is cross platform
testing Memory is STREAM by VIRGINA TECH EDU.

here are some posted scores including ours

SGI_Octane_300 375.3

Sun_Ultra60-360 355.2

Apple_PwrMac_G4-500 558.1

Asus_Athlon AMD_800 mhz 600.3

Cray_J932 Supercomputer 1413.6

DELL Pentium 4-1400 1437.2

CYBERIMAGE P4 1.5 Ghz 1544.4

as you can see the P4 ends up being the fastest CPU in the world and its rambus blows away SUN and SGI workstations
as well as a Cray supercomputer

there is alot of BS in the media about the P4 and rambus not being fast,
and as I have said before many times here, you have to use a program that
1- takes advantage of the P4 code and compilers,
with an OS like ME and DD8b that are compiled for P4

2- most importantly, use a program that can hammer the system hard enough to show where the P3 and Athlon start to
get saturated and decline or level off in performance where the P4 keeps on going..

MS OFFICE and simple programs do NOT do this,
most games do NOT, although QUAKE is an exception
as they do not put a clock or performance timer in theor program like others do, so that explains why
Quake runs at 235 FPS compared to Athlon's 150 !!

when you run very complex programs that have the data to saturate most P3's and Athlons, Like Mpeg 4 encoding,
Photoshop calcs, engineering apps, some games like Quake,
Multimedia apps, video , 3D calcs, like MAX, and Lightwave,
DVD,etc , you can clearly see the results of the 3.2 GPS
of the RAMBUS and the 850 chipset working together

the programmer of memtach wrote me recently when I sent him our P4 results and said this ...

"I'd expect the P4 to do well on the streaming tests (fill double / int, sum double / int) etc..
I'd love to work on a P4 review, but Intel has not been very helpful in the past; they have yet to loan me a system. I'm still working on getting a P4
in for testing - it should do quite nicely on multimedia that can utilize SSE2."

believe me I have run several dozen tests in apps and benchmarks, using the proper atmosphere of ME, WIN XE,
and DD8b and the P4 with dual channel rambus is
like 2-3 times faster than a P3 in some instances
as well as an athlon even with DDR

I know its not what people want to hear if they do not have a P4, but we are relatively unbiased as we have to engineer and sell the highest performing workstations and servers to our clients, and if Athlon and DDR was better , they would demand it and we would supply it.. but they do not,
because the P4 and rambus really is faster,
and I do not have a machine ego in owning a P3, or Athlon
to say P4 is not worth it or is slower to rationalize it,
in fact after testing the P4 and seeing the results
I upgraded my dual 933 P3 to a P4 1.5 immediately,
at some expense but it was worth it..

though you may find this info interesting and keep up the good posts

best
CAMERON





CYBERIMAGE
<A HREF="http://www.4CyberImage.com " target="_new">http://www.4CyberImage.com </A>
Ultra High Performance Computers-

 

[mpjesse]

I'm not disputing the fact that the Rambus/P4 design is extrememly intellegent and quick. My point was that the memory technology itself is nothing new or amazing. What's amazing is the memory bus design and the 850. All RDRAM is is really fast SDRAM- they way it interfaces with the chipset/processor is what makes it so fast. I think Intel deserves all the credit for the enourmous bandwidth capabilities- not Rambus =) So, what I'm really trying to say is that I'm not impressed with RDRAM as memory technology. I am impressed with the 850 chipset and memory bus. But hey, it doesn't matter what I'm impressed with anyways =)

-MP Jesse

"Signatures Still Suck"

 

[Oni]

The point about P4 is the rediculous prices, and the fact that very few things are optimized for it. Also when AMD Thunderbird processors with DDR or SDR ram do as good, or better in some items at MUCH lower cost, and no optimizations needed it truly is the smarter buy. Once everything is recompiled, or newer versions come out then the P4 might be better off. Also the prices are supposed to drop drastically soon which will help intel. If AMD comes out with a processor that actually uses the DDR bandwidth better P4 will have a run for its money. Especially when DDR2 comes out running at 400 MHz front side bus. I really think you can't compare the two chips, they shouldn't be targeted at the same market at the moment. P4 is obviously more for video editing and manipulation (plus quake 3), but Thunderbird is good for everything. Just not as good in video editing and manipulation as P4.
Whats going to be interesting is when the new AMD chip comes out and will hopefully maintain its current performance in most programs, as well as doing better in video editing type programs. I don't mean just video editing, but I mean anything that utilizes large memory bandwidth. Basically I just think you can't compare Thunderbird and Pentium 4, they are for different purposes.

 

[mpjesse]

I agree with everything you said except the "you can't compare the Thunderbird and P4, they're for different purposes". Intel is marketing the P4 as a "high end desktop processor" right now, which makes it suitable for the home PC. Intel doesn't have a dual chipset for the P4 yet. In pure number crunching apps, the Athlon always comes out on top- which (to me) puts it in the high end sector.

My other point is that companies are starting to develop super Athlon clusters- which basically means the Athlon is suited for high performance software like Linux/Unix. I think the processors are one in the same.

-MP Jesse

"Signatures Still Suck"

 

[Guest]

well if u want to see the real results stop using suck a crap program like sandra. it not giving u real world results and only goes by a predefind spcifice setup whitch has about nothing to do witha normal system. the nubmers in sandra aka crap are all synthetic and have no barring on real world numbers. rdram may in thiery be faster but from marks the p-4 rambus combo jsut cant keep up to even sdram systems. ddr im not sure why noone ca find cas2 i know its hard to find but my supplyer has them. and since price is becoming less of a issue the praformance will remain. they can keep tweaking rambus and stuff as much as they want. when dual channel quad pumped ddr comes out running 6.2 gigs of band and the dual athlon givine each cpu 6.2 gigs of band com out (something intel dosnt do there cpus share on bus of 3gigs band) rambus and unless intel changes there ways are going to slip even further in to nomans land.

Computer Shop owner and Head tech.