First Look : Rambus RDRAM RIMM4200
- Thread starter burek
- Start date
Ok, if my math is correct, the current i850 with dual channel PC800 RDRAM has the same bandwidth as a single channel RIMM3200 (or RIMM4200 if the FSB is at 133MHz), right? So the ONLY advantage is that you don't have to upgrade in pairs? WTF? I want dual channel RIMM4200!!!!
<< Ok, if my math is correct, the current i850 with dual channel PC800 RDRAM has the same bandwidth as a single channel RIMM3200 (or RIMM4200 if the FSB is at 133MHz), right? So the ONLY advantage is that you don't have to upgrade in pairs? WTF? I want dual channel RIMM4200!!!! >>
A dual-channel RIMM4200 design would necessitate an upgrade to a 266MHz FSB on Intel's part.
"I want dual channel RIMM4200!!!!"
You have it, kind of. RIMM4200 is RDRAM1066 which would give 2.1gb/s of bandwith with a single 32-bit channel to the memory (granted it's two 16-bit ones.) The new RIMM4200 module is the very fist Rambus product that is exciting to me. Although it might be too late, Intel is killing RDRAM market share.
You have it, kind of. RIMM4200 is RDRAM1066 which would give 2.1gb/s of bandwith with a single 32-bit channel to the memory (granted it's two 16-bit ones.) The new RIMM4200 module is the very fist Rambus product that is exciting to me. Although it might be too late, Intel is killing RDRAM market share.
<< You have it, kind of. RIMM4200 is RDRAM1066 which would give 2.1gb/s of bandwith with a single 32-bit channel to the memory (granted it's two 16-bit ones.) The new RIMM4200 module is the very fist Rambus product that is exciting to me. Although it might be too late, Intel is killing RDRAM market share. >>
No, PC1066 RDRAM is still the same memory as PC800 RDRAM but just operating at a 533MHz FSB (133MHz x 4). RIMM4200 is using 32bits instead of the usual 16bits we are using now on PC800 (or PC1066). With the i850 chipset, it uses dual channel, so with PC800 you will have 3.2GB/s of bandwidth (1.6GB/s X 2) and with PC1066 you will have 4.2GB/s worth of bandwidth (2.1GB/s x 2). Now with RIMM4200, you are transferring 32bits at a time so you double your bandwidth but it only uses a single channel. So you are stuck back at 4.2GB/s. WTH? Someone tell me what I'm missing. What the hell is the point of this? Are they just taking a lateral step?
<< dual-channel RIMM4200 design would necessitate an upgrade to a 266MHz FSB on Intel's part. >>
Not necessarily. Remember the i820? That operated on a 100MHz FSB using single channel PC800 RDRAM. The i850 still operates on a 100MHz FSB but it is using dual channel RDRAM. I think all it would take is a new chipset.
Like I said, if they came out with dual channel RIMM4200 (8.4GB/s bandwidth), that would be AWESOME!!
"Are they just taking a lateral step?"
You're kidding right? One single module provides 4.2gb/s of main system memory bandwith! All it needs is one RIMM and a 32bit path to the memory controller. The only thing in mainstream desktops that can come close is the dual-channel nForce at 4.2gb/s, but on that motherboard you have to route two 64bit channels on the motherboard to the memory controller. To get that kind of performance out of DDR in a single channel solution (and even then, it would require a path twice as big) it would need to be DDR533.
You're kidding right? One single module provides 4.2gb/s of main system memory bandwith! All it needs is one RIMM and a 32bit path to the memory controller. The only thing in mainstream desktops that can come close is the dual-channel nForce at 4.2gb/s, but on that motherboard you have to route two 64bit channels on the motherboard to the memory controller. To get that kind of performance out of DDR in a single channel solution (and even then, it would require a path twice as big) it would need to be DDR533.
<< You have it, kind of. RIMM4200 is RDRAM1066 which would give 2.1gb/s of bandwith with a single 32-bit channel to the memory (granted it's two 16-bit ones.) The new RIMM4200 module is the very fist Rambus product that is exciting to me. Although it might be too late, Intel is killing RDRAM market share. >>
"No, PC1066 RDRAM is still the same memory as PC800 RDRAM but just operating at a 533MHz FSB (133MHz x 4). RIMM4200 is using 32bits instead of the usual 16bits we are using now on PC800 (or PC1066)."
That's what I just said.
"No, PC1066 RDRAM is still the same memory as PC800 RDRAM but just operating at a 533MHz FSB (133MHz x 4). RIMM4200 is using 32bits instead of the usual 16bits we are using now on PC800 (or PC1066)."
That's what I just said.
<<
<< You have it, kind of. RIMM4200 is RDRAM1066 which would give 2.1gb/s of bandwith with a single 32-bit channel to the memory (granted it's two 16-bit ones.) The new RIMM4200 module is the very fist Rambus product that is exciting to me. Although it might be too late, Intel is killing RDRAM market share. >>
No, PC1066 RDRAM is still the same memory as PC800 RDRAM but just operating at a 533MHz FSB (133MHz x 4). RIMM4200 is using 32bits instead of the usual 16bits we are using now on PC800 (or PC1066). With the i850 chipset, it uses dual channel, so with PC800 you will have 3.2GB/s of bandwidth (1.6GB/s X 2) and with PC1066 you will have 4.2GB/s worth of bandwidth (2.1GB/s x 2). Now with RIMM4200, you are transferring 32bits at a time so you double your bandwidth but it only uses a single channel. So you are stuck back at 4.2GB/s. WTH? Someone tell me what I'm missing. What the hell is the point of this? Are they just taking a lateral step?
<< dual-channel RIMM4200 design would necessitate an upgrade to a 266MHz FSB on Intel's part. >>
Not necessarily. Remember the i820? That operated on a 100MHz FSB using single channel PC800 RDRAM. The i850 still operates on a 100MHz FSB but it is using dual channel RDRAM. I think all it would take is a new chipset.
Like I said, if they came out with dual channel RIMM4200 (8.4GB/s bandwidth), that would be AWESOME!! >>
I mean to actually utilize that bandwidth.
This does look nice, and it shows again how much better RDRAM can scale than DDR because even in a single Channel RDRAM chipset like 820 for example. these RIMM's will get still 4.2GB/ps while with single channels, DDR is stuck at 2.7GB/ps, and in 850, these RIMM's will get 8.4GB/ps, rivaling the fastest DDR that's in video cards, but at a fraction of the cost. *sigh*, but Intel has taken away the cost benefit's of these RIMM's by Canceling Tulloch (Single Channel 850), so this is all good. But, we will never see it's benefits for a long time when 8.4GB/ps is needed, or if Intel decides to bring back Tulloch.
Think about this people! The question isn't why, it's why not? These things cost the same to make as PC800 RIMMs and allow a single channel implemintation for RDRAM, while to get the same bandwidth out of DDR you would have to go dual channel. Does this seem like a flipflop?
Now=DDR can be upgraded with single DIMMs
Rambus must be upgraded with dual RIMMs
Later
DDR must be upgraded with Dual DIMMs
Rambus can be upgraded with Single RIMMs
Anyways, this will allow RAMBUS to saturate a 133MHZ quad pumped bus with only one memory channel (I.E. no more upgrading in pairs)
And this costs nothing more than PC800 implementations.
So again the question is
Why not?
Now=DDR can be upgraded with single DIMMs
Rambus must be upgraded with dual RIMMs
Later
DDR must be upgraded with Dual DIMMs
Rambus can be upgraded with Single RIMMs
Anyways, this will allow RAMBUS to saturate a 133MHZ quad pumped bus with only one memory channel (I.E. no more upgrading in pairs)
And this costs nothing more than PC800 implementations.
So again the question is
Why not?
<< All it needs is one RIMM and a 32bit path to the memory controller. The only thing in mainstream desktops that can come close is the dual-channel nForce at 4.2gb/s... >>
Present PC800 has a very good chance of hitting PC1066 speeds and beyond. So TODAY we have a solution that will provide 4.2GB/s worth of bandwidth, the same as ONE RIMM4200. Like I said, the only thing I see here is that you'll only need one stick to make a bank. Big whoop! Ya that is great step but the performance as of right now, single RIMM4200 vs dual PC1066, has the SAME bandwidth. Ya, in the future when they use a 64bit data path that would be AWESOME! But for right now, I don't see anything to get excited about. We have something TODAY that will give us that bandwidth.
<< So again the question is
Why not? >>
Because there is NO need to upgrade if you already have an i850 motherboard. That's why not.
FishTank, I totally agree! Intel's actions are making my head spin. Not only would Tulloch (Single Channel 850) eliminate the need for RDRAM to be installed in pairs, but as I have said all along, a Single Channel RDRAM solution would be much cheaper than a Dual Channel DDR Setup, and with PC4200 RDRAM, you would get identical bandwidth for both setups for less money! Intel just makes me sick sometimes and this is one of them.
I just thought I'd clear up a few things.
<< Ok, if my math is correct, the current i850 with dual channel PC800 RDRAM has the same bandwidth as a single channel RIMM3200 (or RIMM4200 if the FSB is at 133MHz), right? So the ONLY advantage is that you don't have to upgrade in pairs? WTF? I want dual channel RIMM4200!!!! >>
Here's the memory bandwidth figures for the setups being discussed here:
PC800 on 850e: 3.2GB/ps
PC1066 on 850e: 4.2GB/ps
PC800 on Tulloch: 1.6GB/ps
PC1066 on Tulloch: 2.1GB/ps
PC3200 on 850e: 6.4GB/ps
PC4200 on 850e: 8.4GB/ps
PC3200 on Tulloch: 3.2GB/ps
PC4200 on Tulloch: 4.2GB/ps
Tulloch as I have mentioned is Single Channel 850e which will make both the boards cheaper to produce and easier to make. Tulloch is nothing but good, I dunno why intel canceled it. I will mention JackBurton, that while 8.4GB/ps truely shows the scalability of RDRAM, but it won't be needed until Prescott (It will feature a 800MHz fsb, providing 6.4GB/ps, and by then, PC4800 will be out (32-bit PC1200), and we might see a performance benefit on 850e using PC3200 RIMM's over PC4800 on Tulloch, but Prescott is a Year and a Half off so we got a long ways to go. The bottom line is that, Tulloch with PC4200 will saturate the P4's 533fsb that will remain until 2H'03, but of course Intel canceled it.
<< What the hell is the point of this? Are they just taking a lateral step? >>
Well now, for using only 1 RIMM, but originally, these RIMM's would be the key to Tulloch as I've pointed out that with these RIMM's would provide the same bandwidth as a similar 850e setup, but be cheaper. Any questions?
EDIT:
<< Present PC800 has a very good chance of hitting PC1066 speeds and beyond. So TODAY we have a solution that will provide 4.2GB/s worth of bandwidth, the same as ONE RIMM4200. Like I said, the only thing I see here is that you'll only need one stick to make a bank. Big whoop! Ya that is great step but the performance as of right now, single RIMM4200 vs dual PC1066, has the SAME bandwidth. Ya, in the future when they use a 64bit data path that would be AWESOME! But for right now, I don't see anything to get excited about. We have something TODAY that will give us that bandwidth. >>
I'm not disuputing that, but I am more thinking in terms of the price. Sure nobody would upgrade to Tulloch from 850e, but I'm thinking of the person buiying new later this year, and if Tulloch was not canceled, he would be choosing between Single Channel 32-bit RIMM's(Tulloch), and Dual Channel DDR (Granite Bay). So, for someone who has 850e, then you're absolutely right, there's no reason to upgrade. It's all about the price really.
I just thought I'd clear up a few things.
<< Ok, if my math is correct, the current i850 with dual channel PC800 RDRAM has the same bandwidth as a single channel RIMM3200 (or RIMM4200 if the FSB is at 133MHz), right? So the ONLY advantage is that you don't have to upgrade in pairs? WTF? I want dual channel RIMM4200!!!! >>
Here's the memory bandwidth figures for the setups being discussed here:
PC800 on 850e: 3.2GB/ps
PC1066 on 850e: 4.2GB/ps
PC800 on Tulloch: 1.6GB/ps
PC1066 on Tulloch: 2.1GB/ps
PC3200 on 850e: 6.4GB/ps
PC4200 on 850e: 8.4GB/ps
PC3200 on Tulloch: 3.2GB/ps
PC4200 on Tulloch: 4.2GB/ps
Tulloch as I have mentioned is Single Channel 850e which will make both the boards cheaper to produce and easier to make. Tulloch is nothing but good, I dunno why intel canceled it. I will mention JackBurton, that while 8.4GB/ps truely shows the scalability of RDRAM, but it won't be needed until Prescott (It will feature a 800MHz fsb, providing 6.4GB/ps, and by then, PC4800 will be out (32-bit PC1200), and we might see a performance benefit on 850e using PC3200 RIMM's over PC4800 on Tulloch, but Prescott is a Year and a Half off so we got a long ways to go. The bottom line is that, Tulloch with PC4200 will saturate the P4's 533fsb that will remain until 2H'03, but of course Intel canceled it.
<< What the hell is the point of this? Are they just taking a lateral step? >>
Well now, for using only 1 RIMM, but originally, these RIMM's would be the key to Tulloch as I've pointed out that with these RIMM's would provide the same bandwidth as a similar 850e setup, but be cheaper. Any questions?
EDIT:
<< Present PC800 has a very good chance of hitting PC1066 speeds and beyond. So TODAY we have a solution that will provide 4.2GB/s worth of bandwidth, the same as ONE RIMM4200. Like I said, the only thing I see here is that you'll only need one stick to make a bank. Big whoop! Ya that is great step but the performance as of right now, single RIMM4200 vs dual PC1066, has the SAME bandwidth. Ya, in the future when they use a 64bit data path that would be AWESOME! But for right now, I don't see anything to get excited about. We have something TODAY that will give us that bandwidth. >>
I'm not disuputing that, but I am more thinking in terms of the price. Sure nobody would upgrade to Tulloch from 850e, but I'm thinking of the person buiying new later this year, and if Tulloch was not canceled, he would be choosing between Single Channel 32-bit RIMM's(Tulloch), and Dual Channel DDR (Granite Bay). So, for someone who has 850e, then you're absolutely right, there's no reason to upgrade. It's all about the price really.
MadRat
Lifer
- Oct 14, 1999
- 11,999
- 307
- 126
The decision to kill RIMM4200 support is just another notch is Intel's "I am too slow to react to the consumers" belt. This new technology would allow you to use a SINGLE stick of memory in your motherboard for 20% premium. As it is I lose alot more value upgrading a pair of RDRAM sticks in an 850 motherboard.
Intel's board must operate something like this:
P = Performance
$ = Price
E = $/P (Efficiency of Price to Performance)
If E = great then dropproductline
If E = horrible then increaseproductline
Intel's board must operate something like this:
P = Performance
$ = Price
E = $/P (Efficiency of Price to Performance)
If E = great then dropproductline
If E = horrible then increaseproductline
<< "I want dual channel RIMM4200!!!!"
You have it, kind of. RIMM4200 is RDRAM1066 which would give 2.1gb/s of bandwith with a single 32-bit channel to the memory (granted it's two 16-bit ones.) The new RIMM4200 module is the very fist Rambus product that is exciting to me. Although it might be too late, Intel is killing RDRAM market share. >>
How is Intel killing it off? In the GamePC article they show a Asus board that will use RIMM4200 so there must be a chipset that supports it.
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes + WCL Discussion Threads
- Started by Tigerick
- Replies: 23K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
-
Facebook
Twitter