• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

PC1066 NOW AVAILABLE!!!

J

JackBurton

Lifer
I don't know if this has been posted yet, but Kinston has their 256MB PC1066 sticks out using 32ns modules!!! If you think your PC800 40ns memory overclocks well, I'd like to see how these bad boys will do. 😀

As it so happens, I think I'm going to put in an order for an Asus P4T533-C board and 2X256MB PC1066 sticks VERY soon. 😀
 
Originally posted by: JackBurton
I don't know if this has been posted yet, but Kinston has their 256MB PC1066 sticks out using 32ns modules!!! If you think your PC800 40ns memory overclocks well, I'd like to see how bad boys will do. 😀

As it so happens, I think I'm going to put in an order for an Asus P4T533-C board and 2X256MB PC1066 sticks VERY soon. 😀
Click to expand...

Are you sure its not 3.2ns?
 
Its 32ns, RDRAM is very different from DDR... DDR is low like 3.2 and thats only the memory on a GF4 4600... DDR 333 is like 6?
 
$139 isn't too bad for PC1066 since it just hit the market. Lets hope that this price goes down considerably, as PC2100 is more than twice as cheap cheap, and official PC2700 is cheaper as well.
 
I was thinking of that the other day, how come RDRAM is measured in the 30-50 ns range? Either it works differently from SDRAM in some way that the usual Hz = 1/Refresh Rate doesn't hold true or it only actually works at 30 MHz 😉

-Ice
 

Performance
Now for the million-dollar question: How do DDR and Rambus memory compare performance wise? Sorry, I know you don't want to hear this ? that depends. Both technologies have their own ardent supporters and we have seen several different benchmarks to date that provide conflicting results.

On the surface, it seems simple: Data flow at 800MHz is faster than data flow at 266MHz, right? Unfortunately, it isn't that simple. While Rambus modules may have the ability to transfer data faster, it appears to have higher latency (the amount of time you have to wait until data flows) than that of a DDR system. In other words, the first data item transferred in a Rambus transaction takes longer to initiate than the first data item moved in a DDR system. This is due in part to how the systems are constructed.

In a DDR or SDRAM system, each DIMM is connected, individually and in parallel, to the data bus. So whether you have a single DIMM or multiple DIMMs, the amount of time it takes to initiate a data transfer is effectively unchanged.

In a Rambus system, RIMM modules are connected to the bus in a series. The first data item transferred must pass through each RIMM module before it reaches the bus. This makes for a much longer distance for the signal to travel. The result is higher latency. That's not necessarily a problem in an environment where data transactions involve lengthy streams of data, such as gaming. But it can become an issue in environments where many small transactions are initiated regularly, such as a server.

To further explain, here's an example that we can all relate to ? driving your car to the store. You can take the roundabout freeway and drive 20 miles at 70 MPH. Or, you can take a more direct route and drive just 5 miles at 50 MPH. You might go faster on the freeway but you'll get to the store (Memory Controller) faster on the straight-line route.
 
Originally posted by: Bozo Galora
Performance
Now for the million-dollar question: How do DDR and Rambus memory compare performance wise? Sorry, I know you don't want to hear this ? that depends. Both technologies have their own ardent supporters and we have seen several different benchmarks to date that provide conflicting results.

On the surface, it seems simple: Data flow at 800MHz is faster than data flow at 266MHz, right? Unfortunately, it isn't that simple. While Rambus modules may have the ability to transfer data faster, it appears to have higher latency (the amount of time you have to wait until data flows) than that of a DDR system. In other words, the first data item transferred in a Rambus transaction takes longer to initiate than the first data item moved in a DDR system. This is due in part to how the systems are constructed.

In a DDR or SDRAM system, each DIMM is connected, individually and in parallel, to the data bus. So whether you have a single DIMM or multiple DIMMs, the amount of time it takes to initiate a data transfer is effectively unchanged.

In a Rambus system, RIMM modules are connected to the bus in a series. The first data item transferred must pass through each RIMM module before it reaches the bus. This makes for a much longer distance for the signal to travel. The result is higher latency. That's not necessarily a problem in an environment where data transactions involve lengthy streams of data, such as gaming. But it can become an issue in environments where many small transactions are initiated regularly, such as a server.

To further explain, here's an example that we can all relate to ? driving your car to the store. You can take the roundabout freeway and drive 20 miles at 70 MPH. Or, you can take a more direct route and drive just 5 miles at 50 MPH. You might go faster on the freeway but you'll get to the store (Memory Controller) faster on the straight-line route.
Click to expand...

😕 that still doesn't explain why RDRAM frequency isn't given by 4/[cycle time] (since it's a QDR transfer)

-Ice
 
Originally posted by: Bozo Galora
Performance
Now for the million-dollar question: How do DDR and Rambus memory compare performance wise? Sorry, I know you don't want to hear this ? that depends. Both technologies have their own ardent supporters and we have seen several different benchmarks to date that provide conflicting results.

On the surface, it seems simple: Data flow at 800MHz is faster than data flow at 266MHz, right? Unfortunately, it isn't that simple. While Rambus modules may have the ability to transfer data faster, it appears to have higher latency (the amount of time you have to wait until data flows) than that of a DDR system. In other words, the first data item transferred in a Rambus transaction takes longer to initiate than the first data item moved in a DDR system. This is due in part to how the systems are constructed.

In a DDR or SDRAM system, each DIMM is connected, individually and in parallel, to the data bus. So whether you have a single DIMM or multiple DIMMs, the amount of time it takes to initiate a data transfer is effectively unchanged.

In a Rambus system, RIMM modules are connected to the bus in a series. The first data item transferred must pass through each RIMM module before it reaches the bus. This makes for a much longer distance for the signal to travel. The result is higher latency. That's not necessarily a problem in an environment where data transactions involve lengthy streams of data, such as gaming. But it can become an issue in environments where many small transactions are initiated regularly, such as a server.

To further explain, here's an example that we can all relate to ? driving your car to the store. You can take the roundabout freeway and drive 20 miles at 70 MPH. Or, you can take a more direct route and drive just 5 miles at 50 MPH. You might go faster on the freeway but you'll get to the store (Memory Controller) faster on the straight-line route.
Click to expand...


But we all know which gives better performance... DDR333 vs PC1066. DDR333 and PC800 are pretty even

 
O.K., O.K. Mr. Ice:

Go here

and download this PDF

go to page 46 and at the top you will see clock times
(note that differen ns timings straddle more than one timing spec) callout
so we have:
600MHZ = 1000/600 = 1.66ns times 2 (like ddr) =3.33ns
800MHZ = 1000/800 = 1.25ns times 2 (like ddr) =2.5ns
1066MHz = 1000/1066 = .93808 times 2 (like ddr)= 1.875ns

so now we have the ns familiar timings you worry about so much...... right?

so where does that 40ns come from? Simple, they are serial devices, unlike DDR which is parallel, so.............

3.33 times 16 = 53.28ns total
2.5 times 16 = 40ns
1.875 times 16 = 30ns
guess what 2.0ns is (DOH!!!)

now notice on page one of PDF how all timings line up


but thats just my opinion, and I could be wrong
😉
 
Originally posted by: Bozo Galora
Performance
Now for the million-dollar question: How do DDR and Rambus memory compare performance wise? Sorry, I know you don't want to hear this ? that depends. Both technologies have their own ardent supporters and we have seen several different benchmarks to date that provide conflicting results.

On the surface, it seems simple: Data flow at 800MHz is faster than data flow at 266MHz, right? Unfortunately, it isn't that simple. While Rambus modules may have the ability to transfer data faster, it appears to have higher latency (the amount of time you have to wait until data flows) than that of a DDR system. In other words, the first data item transferred in a Rambus transaction takes longer to initiate than the first data item moved in a DDR system. This is due in part to how the systems are constructed.

In a DDR or SDRAM system, each DIMM is connected, individually and in parallel, to the data bus. So whether you have a single DIMM or multiple DIMMs, the amount of time it takes to initiate a data transfer is effectively unchanged.

In a Rambus system, RIMM modules are connected to the bus in a series. The first data item transferred must pass through each RIMM module before it reaches the bus. This makes for a much longer distance for the signal to travel. The result is higher latency. That's not necessarily a problem in an environment where data transactions involve lengthy streams of data, such as gaming. But it can become an issue in environments where many small transactions are initiated regularly, such as a server.

To further explain, here's an example that we can all relate to ? driving your car to the store. You can take the roundabout freeway and drive 20 miles at 70 MPH. Or, you can take a more direct route and drive just 5 miles at 50 MPH. You might go faster on the freeway but you'll get to the store (Memory Controller) faster on the straight-line route.
Click to expand...


You forgot the biggest difference.... bus width... thats why at 800Mhz RDRAM is the same MB/sec as 200Mhz DDR. RDRAM is 32bit width bus while DDR is at 128bit. DDR transfers 4x the amount of data per clock cycle and but the RDRAM is 4x faster (PC1600 DDR vs PC800 RDRAM) So when you say RDRAM transfers data faster your actually mistaken.
 
hey douchelanic:

"So when you say RDRAM transfers data faster your actually mistaken."

MUAHAHAHAHAHAH!!!!!!

that little paragraph I posted did not come from me - I quoted right from Crucial's OWN WEBSITE!!!
Send THEM an email telling them of your findings on this matter.
Better yet, send them your resume', you are obviously far advanced in your knowledge of RAM architecture, you may have a job opportunity here if you handle it right. 😀

So sad.............
It seems that today's "MTV" generation has lost the ability to concentrate for more than a few seconds on anything conceptual. The article did not say RDRAM is faster, but that it should be faster - but isn't, and that RDRAM and DDR try to acomplish the same thing in different ways. Like parallel IDE and Serial ATA.

Or, on the other hand, you may just be looking to liven up your dull day with an argument.

and by the way: you ARE mistaken on spelling "your" - its you're (as in you are)

and furthermore, your simplistic judging of DDR vs RDRAM needs a
little more research on your part. this is an exceedingly complex
subject

 
<<You forgot the biggest difference.... bus width... thats why at 800Mhz RDRAM is the same MB/sec as 200Mhz DDR. RDRAM is 32bit width bus while DDR is at 128bit. DDR transfers 4x the amount of data per clock cycle and but the RDRAM is 4x faster (PC1600 DDR vs PC800 RDRAM) So when you say RDRAM transfers data faster your actually mistaken.>>

It matters what you are comparing. The theoretical transfers of RDRAM are far higher for comparable products. PC600 and PC700 can outpace DDR200 memory. PC800 can outpace DDR333 memory. And PC1066 can outpace DDR400. PC1200 memory is just around the corners, so DDR400 will be competing against it, too.

DDR400 = 2 channels x 64 bits per channel x 200MHz = 3200 B/sec
PC1066 = 2 channels x 2 memory sticks x 16 bits per channel x 533MHz = 4264 B/sec

The average latency of transfers for PC1066 is down to within DDR levels. Besides that RDRAM is better for large sustained transfers of data whereas DDR has to refresh more often. So if someone is going to use a Pentium4 it just makes sense to use PC1066 over DDR400.
 
Originally posted by: Bozo Galora
O.K., O.K. Mr. Ice:

Go here

and download this PDF

go to page 46 and at the top you will see clock times
(note that differen ns timings straddle more than one timing spec) callout
so we have:
600MHZ = 1000/600 = 1.66ns times 2 (like ddr) =3.33ns
800MHZ = 1000/800 = 1.25ns times 2 (like ddr) =2.5ns
1066MHz = 1000/1066 = .93808 times 2 (like ddr)= 1.875ns

so now we have the ns familiar timings you worry about so much...... right?

so where does that 40ns come from? Simple, they are serial devices, unlike DDR which is parallel, so.............

3.33 times 16 = 53.28ns total
2.5 times 16 = 40ns
1.875 times 16 = 30ns
guess what 2.0ns is (DOH!!!)

now notice on page one of PDF how all timings line up


but thats just my opinion, and I could be wrong
😉
Click to expand...

Thanks for the info and explanation Mr. Galora, that clears things up quite a bit 🙂.

-Ice
 
RDRAM is 16bit wide. If you are talking about the dual channel (i850, i850E, i860) then yes, it will be 16bit X 2 transfer, but the RDRAM is still 16bit. The new 32bit RDRAM will be coming soon.
 

it will come bundled with the P4T533 when that mobo comes out in a few weeks.
Note that this has been a done deal since Feb 25:

Rambus Demonstrates RIMM 4200, World's Highest Bandwidth PC Memory Module

Los Altos, CA - February 25, 2002 - Rambus Inc. (Nasdaq: RMBS), the leading provider of high bandwidth chip connection technology, announced today it will demonstrate RIMM? 4200, the world's highest bandwidth- PC memory module, at the Intel Developers Forum in San Francisco, California. The RIMM 4200 module represents the next milestone in the RDRAM® memory roadmap for higher performance computing solutions.

The RIMM 4200 module integrates two RDRAM memory channels to deliver 4.2GB/s of memory bandwidth from a single module. Using standard RDRAM components, RIMM 4200 modules operate with existing RDRAM controllers, and enable single module upgrades of dual channel memory systems. Furthermore, with built-in channel termination, RIMM 4200 modules simplify motherboard design and require less than 5 square inches of system board area. The demonstration system to be shown at Intel Developer's Forum uses a low cost four-layer motherboard and modules that fit in standard memory connector footprints using a 232 pin version of the RIMM connector.

The system demonstration of RIMM 4200 modules, featured in Rambus booth number 404 is the result of joint development efforts by Samsung, ASUSTeK, Kingston Technology, Tyco Electronics, and Rambus.

"As a world leading technology solution provider, ASUS is pleased to be the first company to deliver motherboards supporting RIMM 4200 modules," comments Joe Hsieh, Motherboard Division Director of ASUSTeK. "RIMM 4200 modules provide exceptional performance while enabling the use of smaller form factors at lower costs. The ASUS P4T533 with RIMM 4200 support will be shipping in quantities in Q2 2002."

"The RIMM 4200 uses the 1066 MHz speed bin of current production RDRAM to deliver the highest performance available to date from a single PC memory module," said Tom Quinn, vice president of Marketing, Samsung Semiconductor, Inc. "A single module that supports two channels of RDRAM with 4.2GB/sec of bandwidth makes RDRAM even more attractive from a price-performance standpoint."

"Kingston has already built a number of prototype RIMM 4200 modules and achieved high passing yields in motherboards," stated JK Tsai, Director of Technology, Kingston Technology. "Kingston will be ready to ramp its volume production of RIMM 4200 modules to support OEM, System Builder, Channel and other customers when new motherboards and systems ship."
 
You must log in or register to reply here.

TRENDING THREADS

Back
Top