P4 and DDR

"I heard from my friends that there are now P4 with DDR mobos now. Is it true? So how does it perform compared with P4 and RDRAM? I read alot of Raystonn's articles and I'm pretty convinced that RDRAM will perform better under heavy workloads.

Plus another curious question. So if th P4 (northwood) can be scaled up from 2GHZ to 10GHZ in the future using the same pins, I assume, and more software will become recompiled plus SSE2 optimized, and the IPC of the P4 increased with the new Northwood, the P4 will become faster than anything AMD or Cyrix would have to compare. Is this assumption a good one or bad?

But there's something that concerns me. There's a blue heatsink covering the whole piece of RDRAM. Either that means this type of RAM heats up really fast or it's pure cosmetics.

We put text in here and that's our signature?"

"I heard from my friends that there are now P4 with DDR mobos now"

DDR chipsets for the Pentium 4 will not be available until Q1 of 2002.

"RDRAM will perform better under heavy workloads"

It will. SDRAM (DDR and SDR) suffers from the dead wait cycles when switching from reading to writing and back. This happens all the time in normal applications and multitasking OSes. This penalty will not show up in synthetic benchmarks that do separate batches of reading and writing in different tests.

"Is this assumption a good one or bad?"

I would say it's good speculation. But remember that it's just that. Speculation can easily turn into religious wars.

"a blue heatsink covering the whole piece of RDRAM"

This is to distribute heat evenly along the whole memory module. All types of memory eventually heat up under heavy load. RDRAM is the first (I'm sure DDR SDRAM will add this eventually) to protect against heat damage in this way.

-Raystonn

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

The problem I have with comparing RDRAM to DDR SDRAM in terms of performance is there are no equal system tests to perform such comparisons. The main thing right now is DDR SDRAM Athlons system perform better than RDRAM Pentium 4 systems in most real world test, not synthetic benchmarks. The only way we will truly know which is a better memory type is when the Brookdale Pentium 4 DDR chipset comes to market. That way you have the same processor that can be used with both memory types, which is the perfect way of determining which type is better. However, we saw that PC133 SDRAM usually performed better than RDRAM on Pentium III systems, but it isn't quite a fair test because Pentium IIIs maximum bandwith could not utilize RDRAM to its bandwith potential. Brookdale, or a possible RDRAM Athlon chipset (?), is the ultimate test in determining which memory type is truly better.

"We put the <i>fun</i> back into fundamentalist dogma!"

"So if th P4 (northwood) can be scaled up from 2GHZ to 10GHZ in the future using the same pins"

I don't think anyone has said anything about a 10Ghz P4 having the same socket or pinout - just that the core will get there. The socket has already changed after 3 cpu releases (1.3, 1.5, 1.7) so I see very little hope that Intel will now keep the new socket configuration of the next three years. Maybe they will though - 370 is pretty stable.

-* This Space For Rent *-
email for application details

"DDR ram draws less power and dissapates less heat that SDRAM let alone Rdram"

Do you have any numbers to back this up? By the way, DDR is SDRAM. The only different is it transmits twice per clock instead of once. The actual design is the same.

As far as RDRAM versus DDR, RDRAM has built in throttling capabilities to move RAM into a sleep state when it's not being used. This is much the same as is done for CPUs in notebooks. It has merely been expanded to work with RAM as well. This makes for less power consumption when power management features are enabled.

-Raystonn

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

you would think that someone that works for intel would be aware of that.

A little bit of knowledge is a dangerous thing!

Touchy today are we? Let's test your knowledge. Can you tell me the voltage requirements for PC800 RDRAM? I'll save you the trouble. RDRAM runs at 2.5 volts. There is no difference in wattage or heat between RDRAM and DDR SDRAM.

-Raystonn

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

> Plus another curious question. So if th P4 (northwood)
> can be scaled up from 2GHZ to 10GHZ in the future using
> the same pins, I assume, and more software will become
> recompiled plus SSE2 optimized, and the IPC of the P4
> increased with the new Northwood, the P4 will become
> faster than anything AMD or Cyrix would have to compare.
> Is this assumption a good one or bad?

I would say the assumption is poor, because you assume that only Intel is moving forward. I'm certain that by the time "in the future" (as you say) when P4 reaches 10 GHz, AMD processors will be there as well.

Leo

I dont believe you are working for intel. If you were working for intel you would know that rambus ram is using alot more power than sdram.
Thats one of the resons why it was never used in notebooks.

One of the biggest strengths of RDRAM is its clock-speed, but this has also been a source of many of its problems. Running at these higher speeds, both RIMMs and motherboards must be manufactured with much more stringent tolerances and with shorter trace lengths to avoid signal corruption from electromagnetic interference (EMI). This adds a double-whammy to the system price. Additionally these factors complicate designing RDRAM systems as demonstrated by Intel with its i820 fiasco.
High speeds create an even greater problem: heat. A single tiny RDRAM chip can consume several Watts of power making the device very, very hot very, very quickly. In fact, the chip could quickly burn up if this heat is not siphoned away. Due to this, RIMMs are most often manufactured wrapped in a metal heat sink. It is also recommended that a fan blow directly across RIMMs to help dissipate this excess heat.
<A HREF="http://www6.tomshardware.com/mainboard/00q1/000315/rambus-03.html]" target="_new">http://www6.tomshardware.com/mainboard/00q1/000315/rambus-03.html]</A>

Heat IS an Issue
Despite the new arguments to the contrary, heat IS an issue. RDRAM and its controlling chipsets have elaborate thermal regulating circuitry to insure that no device overheats. According to Paul DeMone <http://www.realworldtech.com/page.cfm?ArticleID=RWT112299000000&PageNum=4>, a respected expert on memory technologies, when pages are open on an RDAM device, the single tiny RDRAM chip can consume nearly 4W! Obviously the little chip could burn up quickly if not properly cooled.
Unlike SDRAM, this heat can be concentrated into a single chip. As a remedy RDRAM has a special heat sink so that the heat from one chip can be pulled away and dissipated across the heat sink as well as the other chips it comes in contact with (as with most heat sinks, each chip has a thermal interface to insure efficient heat transfer, so in actuality the heat sink does not truly touch the RDRAM - that would be ineffective).
If there wasn't a problem with heat, then RIMM's would not come with the pricey heat sink attached to them. This heat sink is also called a heat spreader, but understand that the later is a subset of the former, despite strange arguments to the contrary <http://www.anandtech.com/showdoc.html?i=1239&p=8>.
A Hot Tradeoff: Performance vs. Energy Consumption
Open pages can be viewed as tiny caches on RAM devices. RDRAM is capable of having more open pages than SDRAM, which should help it latency wise, but in actuality a page read in RDRAM has a hit comparable to bank read latency on SDRAM. Additionally, as stated above, an active RDRAM device consumes a lot of power and gets hot quickly.
So there is a tradeoff: more open pages, higher energy consumption, more heat (and possibly less stability), but better performance, versus fewer open pages, less energy consumption and less heat, but also poorer performance. The current RDRAM systems are performing a careful balancing act, and the heat sink is the balancing pole that allows RDRAM to run a little faster across this tightrope to improve performance.
[url]//www6.tomshardware.com/mainboard/00q2/000529/index-02.html[url]

A little bit of knowledge is a dangerous thing!

"High speeds create an even greater problem: heat"

DDR SDRAM has the same problems as RDRAM in this regard. Higher speeds will always cause more power consumption, and therefore more heat. DDR SDRAM is not immune from this.

"Unlike SDRAM, this heat can be concentrated into a single chip"

This is actually the key reason why many people think RDRAM produces more heat than SDRAM. Over the whole memory module, the average heat dissipation is going to be about the same as that of a DDR SDRAM module running at approximately the same speed. The heat spreader keeps heat from building up in small areas of the memory module. Over the whole module however, (i.e. on average), the disappated heat will still be about the same as DDR SDRAM once it reaches the levels of performance of RDRAM.

The main differences between DDR SDRAM and RDRAM once bandwidth equalizes between the two will be latency during normal system use (reads and writes intermingled) and the unique power management features of RDRAM, especially useful in conserving power in notebooks during less demanding applications.

-Raystonn

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

"If RDRAM is truly so great, why did it perform sub-par to a supposed inferior memory type?"

For an answer to that question we must look to the benchmarking software that was used in the comparisons. SDRAM (SDR and DDR) has a well known problem switching from reading to writing and vice versa. This incurs great latency penalties. Were the benchmarks working in a way that avoided this problem altogether? Did they do a batch of reading separate from a batch of writing? This is usually the case with synthetic memory benchmark applications, and hence you get results that do not mimic real-world performance.

-Raystonn

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

The thing is Raystonn, you constantly stress that synthetic benchmarks make the playing field unfair, but you fail to explain the reason why in real world tests, RDRAM still failed. Look at some of Tom's benchmarks in

Quake III
i840 PC800 - 149 fps
440BX PC133 - 160 fps
i815E PC133 - 153 fps

Expendable
i840 PC800 - 101.8 fps
440BX PC133 - 105.7 fps
i815E PC133 - 102.8 fps

Unreal Tournamnet
i840 PC800 - 42.37 fps
440BX PC133 - 44.99 fps
i815E PC133 - 45.00 fps

In most synthetic benchmarks, yes, PC133 does trounce RDRAM. There were a few in the article where i840 did indeed beat PC133, but in most of these cases PC133 did beat i820. I mean, what's the deal with this? Here's the article I got the information from:

<A HREF="http://www.tomshardware.com/mainboard/00q2/000611/index.html" target="_new">http://www.tomshardware.com/mainboard/00q2/000611/index.html</A>

"We put the <i>fun</i> back into fundamentalist dogma!"

I really don't buy your explaination of a crappy chipset. i840 is an extreme performance chipset for high end workstations. Intel, touting the benefits of RDRAM and trying to convince the world that it is the future of memory, would not waste time creating a crappy chipset. Most likely they did the best they could with what they had. It is believed by many in the computer world that Intel intentionally handicapped the i815 in order to make the RDRAM solutions look better. Via's chipset can't really be compared, well, because it's Via. Intel is much better at making chipsets than Via is. I just don't think Intel would make a crappy chipset for the "future" of memory technology.

Considering the huge differences in the technology behind SDRAM and RDRAM, it is extremely unlikely that we will see a chipset that supports both types. DDR and SDR SDRAM are more similar, so that is why ALi's MaciK1 and Intels Brookdale will be able to support both.

"We put the <i>fun</i> back into fundamentalist dogma!"

That's what the i820 was supposed to be; we all know what a mess <i>that</i> turned out to be. Besides which, Intel blatantly and intentionally handicapped the SDRAM performance in order to push RDRAM. No disrespect to you or your job, but I wouldn't put it past Intel to try another such chipset and handicap DDR/SDRAM in some more subtle manner, just to keep pushing RAMBUS.

I think I'd have to see what someone like ServerWorks could do with the P4--see if they could produce both a DDR and an RDRAM chipset--and put the two head-to-head then.

Kelledin

bash-2.04$ kill -9 1
init: Just what do you think you're doing, Dave?