Corsair announces DDR2 Memory

[Helevitia]

Interesting...I didn't think we would see DDR2 memory until the end of the year. Now they are saying March. I guess all fo the new AMD 939 mobos and Intel's offering will have some sort of support for it? I just got this email a few minutes ago...

San Francisco, CA, February 18, 2004 – Corsair Memory, Inc., a global
leader in ultra-performance DDR memory modules for PC gamers and
enthusiasts, today announced their new XMS2 family of xtreme performance DDR2
memory modules.

The first product in this family is the Twin2X1024-4300, a matched pair
of 512MB 533MHz 240-pin DDR2 modules. This product has been tested
extensively on several different DDR2-based motherboards and is expected to
be 100% compatible with the newest DDR2 chipsets and motherboards.
Every Twin2X1024-4300 matched pair is tested as a pair in a dual channel
DDR2 motherboard at 533MHz to ensure functionality at specified clock
speeds in a dual channel environment. The Twin2X1024-4300 supports JEDEC
standard 4-4-4-12 latencies, and comes equipped with classic black
aluminum heat spreaders.

DDR2, the second generation of Double Data Rate memory technology,
includes new features such as a larger 4-bit prefetch and programmable
on-die termination (ODT) to reduce signal bounce and improve reliability at
high speeds.


btw, this wasn't the whole email, just the part I felt everyone wanted to read.

dave

[Helevitia]

Here is a nice FAQ:

http://www.corsairmemory.com/corsair/products/specs/ddr2_faq.pdf

And this supports the fact that mobos in the march timeframe will have DDR2 support

Q: What motherboards can support DDR2?
A: Check the specs of the motherboard for memory support. The first motherboards that support
DDR2 are not expected to become available until March/April 2004.

[Marshmallowman]

I think it will be a while before AMD supports DDR2, with the athlon 64 the higher latencies are a bit of killer.

There is going to be at least one new release of DDR 1 939 before AMD does an ddr2 CPU, I am just wondering if it will be slot/pin compatible with ddr1, that way you could just get a new AMD CPU to gain DDR2 support in the future

I think corsair is getting ready for intel ddr2 baords , which of cours love bandwidth and are not so picky about latency.

[rylan]

Oh yeah gotta love the insanely high latencies of the joke called DDR2.

[Kooldino]

CAS 4? That's terrible...so what's the advantage? higher bandwidth?

So IIRC, DDR took a sample at the beginning and the end of the clock cycle, correct?

So will DDR 2 sample at the rising and falling edges at the beginning and end of the clock cycle or what?

[Kurt]

CAS 4 for 533MHz is like CAS 2 for 266MHz.

[Helevitia]

Originally posted by Kurt
CAS 4 for 533MHz is like CAS 2 for 266MHz.

Uh, I think it is more like 10, maybe 20% difference.

[rylan]

Kooldino, DDR2 has basically the same clocking for data writes/reads as the normal DDR... sampled on rising and falling edges as you mentioned. DDR2 is really just DDR with a higher clock rate and greatly increased latency to still meet timing requirements of the chips with the faster clock.

[borat]

Im really confused over the whole memory latency and bandwidth issues at the moment. the way i understand it is this - for normal DDR cas 4@200Mhz would be equal to cas 2@100Mhz, this means that so long as you can up the clock by enough Mhzs the latencies start to negate in terms of their actual difference. DDR @ 200Mhz gives 3.2meg/sec bandwidth. Now when we bring DDR2 into the equation we get 533Mhz DDR (which i assume is actually running at 533Mhz and not at 266Mhz?) at cas 4. How does this latency relate to DDR latencies and what bandwidth will you bet getting from a DDR2 533mhz module?

[rylan]

Memory performance of a 200MHz chip at cas4 is not equal to that of one running at cas2@100MHz. The higher clocked memory is going to -transfer- data faster no matter what, as is shown by the higher bandwidth numbers, and especially in block mode transfers. Where the latency hit really shows up is when you have a lot of random accesses to the memory.
CAS (Column Address Strobe) latency is the number of clock cycles that you have to wait after asserting a colum before you can read or write data... along with this is RAS and other delays like RAS-to-CAS and write recovery time etc. Going from CAS2 to CAS4 at the same clock speed doesn't result in a 50% performance hit, its something more like 20% for real performance. For DDR2, when you are accessing blocks of data it takes longer to get to the data, but once that delay time is over the data writes/reads are a lot faster due to the high data clock.
The effects of increased latency are really a system and application dependant thing. AMD systems perform better with low latencies, since the processor and memory controler interface is optimised for low wait states... remember AMD processors process more data clock-for-clock so waiting a couple clocks has more of an impact than on an Intel system, where they rely on high clock speed and do less per clock. For things like gaming, lower latencies will result in higher performance due to the high amount of random access and bank/column switching in the memory array. For apps like streaming video capture, you are doing block mode transfers to the memory, so high bandwidth is preferable.
Its hard to say exactly, but my guess is a DDR2 533MHz module with those high latencies in an AMD system would perform similar to a 433MHz DDR module with agressive timings. Hope that wasn't too confusing

[borat]

not too confusing at all thanks for the summary, wasnt implying that cas2 100Mhz would perform with the same bandwidth as cas4 200Mhz, i was trying to say that in terms of the actual time delay(in (milli?nano?)seconds) of waiting to read/write would be the same for both specs because the 200Mhz memory is clocking twice as many times per second as the 100Mhz chip. from what you say i figure that DDR2 533Mhz would have double the bandwidth of DDR 533Mhz, but latencies would be directly comaprable between the 2.

thanks for the info

[Tjalfe]

.. So basically the Latency is the seek time and bandwidth is the maximum transfer rate if going with a hard drive analogy

[borat]

yeah i get it, only it would be like having seek times of a 10K drive not being directly comaparable to those of a 7k drive

[Kurt]

Originally posted by Helevitia
Uh, I think it is more like 10, maybe 20% difference.

No, it's exactly that. I'm not talking about overall performance, or memory transfer or whatnot.

If it takes you 2 cycles at a frequency of 1Hz to get the column adress strobe, it takes 2 seconds to do so.

If it takes 4 cycles at 2Hz, the total time to get the CAS is still 2 seconds...

[rylan]

Yeah, I think we were confused and thought you meand that 400MHz cas4 performs the same as 200MHz cas2. But in terms of the actual time in nanoseconds, yes cas4@400 = cas2@200. However your memory is sitting there for 4 clocks@400 when you change columns and doing nothing when it could be utilized with a faster latency