DDR vs DDR2 Latency, How Cycles Work, and Dual Channel Marketing

I’ve noticed one thing on the Internet, that stands out above almost all others: most people on the Internet have no clue what they are talking about. Case in point, a lot of ricers and gamerz like to say that DDR is lower latency than DDR2 because DDR2 takes more cycles to do things; except they forget one important thing: cycles are not a measurement of time, they are a measurement of iterations.

That said, there is only one case where DDR actually manages to be lower latency than DDR2 (and this doesn’t mean it has higher performance, or effects benchmarks in any measurable way in favor of DDR), and that is with DDR400 memory vs. DDR2-400 memory: latency is theoretically lower, but you pay a penalty for giving up DDR2’s larger prefetch buffer and better power efficiency. Also, no one actually uses DDR2-400 memory, only 667 and 800. DD2-800 compared to DDR400, latency ends up being similar in impact, and the actual performance is at least twice as much as DDR400, probably even more.

Another thing people say is that DDR2 is slower because it takes more cycles to do things. Yet another thought that hasn’t been fully thought out, and in a similar manner to the whole latency problem (infact, they are directly related; faster timings usually decrease latency across the same memory archetecture). As I said earlier, cycles do not measure time; however, cycles combined with cycles per unit of time measure time. DDR2 in most, if not all, situtations simply performs better.

So, to anyone out there that says that DDR2 is a step backwards: You’re an idiot.

Written by
Open Source software architect and technologist. He's just this guy, you know? Follow him him on Google+.
Published in
Transmissions from the Little Blue Marble

Published September 18th, 2006

Comments

75 Responses

As a budding reviewer (and I say budding because after 4-years I’m still learning) I find there is so much marketing fodder in the tech industry it’s amazing anyone can decipher the prima facie specifications from those “instituites” whom establish them? While JEDEC seems to be on the up and up we know all too well the sad story of Intel’s FormFactor and the demon spawn known as Multi-Rail power supplies? Studying the history tells us Intel processors were so power hungry an additional 4-pin baseboard connector was needed at 12V to feed their power hungry CPU’s. The multi-rail debacle which did nothing for performence except produce many thosuands of cheaply built PSU’s pales in comparison to a single Rail PSU which is simply built correctly. But I digress.

I’ve come to the conclusion that DDR2 while it does have it’s benefits seems to be more of a logical progression for Intel NetBurst and now C2D with 333FSB (quad pumped DDR2 = 1333MHz) then AMD processors. If anything AMD should have skipped the AM2 socket processors with on-die memory controllers reamining at 200FSB and simply waited for DDR3 or as Intel has done, pushed it to market.

I imagine auto-programming will be both a blessing and a curse for some, of course that will depend on motherboard makers who choose to program more aggrassive timings into their memory-subsystem. All that stands betwen an “overclocking” motherboard and JEDEC approved stability are increased voltages and speeds.

The day I saw a heat-pipe on a DDR2 module was the day I saw the proverbial Third Teet.

Any comments on what Van Smith of VHJ would term Marchitecture and how it drives technologies which have less to do with performance and everything to do with more money.

Yes, that indicates that its running it as CAS 5. Also, the sticker says 2.0 – 2.1, so they safely run at 2.0, but can be overvoltaged up to 2.1. So, those are probably safe.

Thanks for the 411. I’ve still got ricer blood in me from my turbo Honda days 🙂 The heat spreaders just look too sweet to pass up, but I didn’t buy into those OCZ’s with the heat sinks and built in water ports! The GSkill’s I found were around the same price as I saw most 5-5-5-15’s sans heat spreaders going for. That made me start thinking they weren’t really faster, despite their marketing…

Right on the box of the G.Skill 4-4-3-5 (2GBHK), the sticker says 2.0~2.1v. Ooops. The other kind are 1.8~1.9v, so I suspect my wife will have the more stable machine.

I just got these chips in… So I fired up CPU_Z. It says 5-5-5-15 on its Memory tab, but 4-4-3-5 on the SPD tab. So this means the mobo is running them at 5-5-5-15? This (low end) Gigabyte mobo will let me mod the voltage, like you say most do, but not the timing. The ECS mobo won’t even let me monkey with the voltage. Oh well, what can you expect from such inexpensive mobo’s 🙂

Glad to hear DDR3 will be more automatic (will that include automatic setting when using mix and matched DDR’s?). I remember back in the day having to set jumpers on the mobo’s to set the speed and multiplier settings manually 🙂

Steve: The timings most likely are not correct due to one reason: most motherboards do not allow you to tweak memory beyond the voltage and CAS speed.

First, I’d like to mention the correct settings for both DDR2-667 and DDR2-800:

For all DDR2-667 memory: CAS 5 (aka 5-5-5-15) at 1.8 volts.

For “value” DDR-800 memory: CAS 5 (aka 5-5-5-15) at 1.8 volts.

For “high performance” DDR2-800 memory: CAS 4 (aka 4-4-4-12) at 1.8 to 2.0 volts.

Now, here is where it gets sticky. High performance DDR2-800 exceeds the DDR2 specification due to voltage: DDR2 requires DIMMs to use no more than 1.8 volts.

However, this rule can be bent due to the fact that it is very hard to manufacturer a CAS 4 DDR2-800 DIMM that runs properly at 1.8 volts, and DDR2-800 is meant to run at CAS 4.

Thus, CAS 4 DIMMS run at 2.0 volts, but DO NOT BUY ONES THAT EXCEED 2.0 VOLTS, this is very far out of spec and many motherboards won’t let you set a voltage that high anyways.

Now, I know you asked about timing and not voltage, but voltage is very important, even more important than timing.

Answering your quesiton, no, G.Skill nor OCZ (both popular ‘ricer’ brands) do not directly lie about timing. What they fail to tell you is that:

a) on either 667 or 800, timings outside of the three I listed are very uncommon and not supported well.

b) as I just said, voltage that exceeds 1.8 on 667 and 2.0 on 800 is dangerous.

c) as I also just said, motherboards rarely let you set timings down to all four numbers, and only let you set the CAS timings.

Unless all four numbers are 3-3-3-9 or less, it is not CAS 3 (which only a few motherboards support). All the memory you listed in your comments are CAS 4.

The real kick in the pants, however, is that running your memory at 4-4-4-12 is not slower. You simply won’t see a speed increase past standard CAS 4 timings.

So, no, G.Skill is not lying to you, but you shouldn’t care. Treat it as CAS 4.

I wouldn’t blame you if you find this all confusing. Memory timing should not be this customizable, and it only confuses hardware and/or damages hardware in the end.

As a side note: DDR3 is going to have mandatory auto-configuration in the specification, so motherboards can automatically run memory as fast as it can go, ie, automatically setting the four timing numbers and voltage. All this confusion will be a thing of the past.

Correction — meant give her the 2GB of 3-4-3-8.

My main point is: while both are 800Mhz and will be running with the same AM2 CPU (albiet her’s is the lower wattage version), it would be some serious benchmarking to notice any real difference between these latencies.

And so, while I’m sure G.Skill is on the up and up, how would anyone really know if a company is telling the truth about the latencies they put on their product stickers?

If a manufacturer posted fake latencies, would anybody notice?

I have two models of G.Skill, F2-6400CL3D-2GBGA @ 3-4-3-8 and F2-6400CL4D-2GBHK @ 4-4-3-5 (both are 800MHz). I have 2x1GB of the 3-4-3-8, and 4x1GB of the 4-4-3-5. I can’t decide which would be better for my gaming computer. My wife does a lot of transcoding of videos, and we both have AM2 4200+ CPU’s (but she has the low wattage version). Guess I’ll go with the 4GB of 3-4-3-8 and give her the 2GB of 4-4-3-8 🙂

Unfortunately I have (relatively) sucky motherboards — an ECS and a Gigabyte. The Gigabyte is ok, both have onboard gigabit LAN. The ECS sucks since although it has onboard gigabit, it only has SATA 150. LOL. I added a SATA 300 Maxtor SATA controller, which I guess gives me 266mps throuh PCI. Both boards let me overclock the CPU, but have limited memory overclocking support.

Yes, without question. Read the manual that comes with your hardware next time.

i have a motherboard 6100sm-m pci-e 16x (runs in 8x mode)if i buy pci-e 16x video card, do you think it will run?

Joe meant to write ddr2-400 and ddr2-800, not 2400 and 2800. Yes, theres a difference.

Joe: DDR400 and DDR2-800 should have the same or better latency (due to improved memory controllers and on-processor caching systems) and twice the bandwidth.

You have been comparing ddr400 vs ddr2400
“theoretically’

how does ddr400 compare to ddr2800?

John: You mean blog. Anyhow, below 30 is considered quiet, below 25 is what quiet computing fanatics consider quiet. Most off the shelf computers are 30-35.

I know this forum us about ram, but you seem like you know your stuff so I have a different question. What DBA should a case fan have to be considered “quiet” ???

Eddie: You quoted a really really old test. S939 is no longer being used, and that may have actually been true back then.

With AM2 processors (and AM3 right around the corner), and Intel’s Core and Core 2 processors, it is way more than 20%.

Theoretically might be, but in real life… the difference seems to be max. 20%

http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=2741&p=3

Theoretically, twice as fast.

So HOW much faster is ddr2 800 over ddr400?

This just proves no one actually reads what I write. You want 3 gigs of DDR2 800, especially since all new platforms only support that.

In fact, late this year or early next year DDR3 is going to be coming out, with 1200 or 1600 being the high end performance.

For those wondering, the same argument that applies to DDR vs DDR2 applies to DDR2 vs DDR3 as well: DDR3 800 will be higher latency than DDR2 800, but no one in their right mind would ever use DDR3 800.

I am looking into buying a new system to handle directX 10 and all the games associated with it. Should I go for 3 gigs of DDR 400 or DDR2 800?

DDR 500 actually doesn’t exist. There is no actual JEDEC specification for DDR at 500, and any memory that says its 500 is just “overclock friendly” 400.

Also, you’re welcome.

Dude… you just hit the spot. For a good while now, I have been mostly relying on the support of some seemingly well-respected advice on the subject of making the jump to DDR2. I have only been cautious of this next step in performance because all I kept hearing is that DDR2’s latency barely makes up for the jump in speed. Again, thank you for your bit of advice on this.

Just today I was making plans for my next rig to build (selling this one soon to a friend) and caught myself even grazing through the high performance DDR 500 instead of going straight to the DDR2 section. Hah.

Anyways thanks again.

Matt, go get some DDR400, some DDR2-400, and some DDR2-800, and do memory bandwidth and latency tests.

DDR2-400 should be higher latency but the same bandwidth as DDR400, and DDR2-800 should be the same latency but twice as much bandwidth as DDR400.

I have an ASUS socket 775 motherboard with two DDR slots and two DDR2 slots… they are indeed on the market

There is no way to benchmark latency in any scientific way. You’d need a socket LGA 775 motherboard (the kind late P4s and Core/Core 2s use) that has both DDR and DDR2 slots, and I’ve never seen one like that on the market.

If you attempt this with, say, two different cpu+motherboard combinations (such as S939 K8 vs AM2 K8), you’ll be gaining performance purely out of better memory controller design and faster front side bus.

Benchmarks please. 🙂

Thank you!

Now it’s time to link this to some of the idiots I keep running into. 😀

Leave a Reply to Kyle