< Back to OSY 1.0 thread list

OSY 1.0 Thread Viewer

Thread #: 1090

More Intel chipset rumormongering from Xbit Labs

Imitation Gruel

Fri Oct 19 16:35:19 2001

Intel’s plans concerning the new chipsets for its Pentium 4 processor keep changing. However, there is a single trend, which goes through all these changes: no more Rambus memory. In fact, Intel has already cancelled its Tulloch chipset, which was supposed to support new 4-bank PC1066 RDRAM, having replaced it with Tehama-E, a new i850 chipset revision supporting [color=blue]533MHz Quad Pumped Bus and regular PC800 RDRAM[/color]. As it follows from the last updated roadmap of Intel’s, [color=red]Tehama-E is to become the last Intel chipset supporting RDRAM.[/color]

Starting from early 2002, Intel will start making DDR chipsets for Pentium 4 CPU. At first, there will appear Brookdale-D supporting DDR200/266, and then it will be replaced with its renewed modifications, namely: Brookdale-E supporting 533MHz bus and Brookdale-G with the integrated graphics core.

However, we found out that Intel is also working on a couple of really interesting DDR solutions:

Granite Bay. This chipset should appear in the middle of next year and will be targeted for high-performance systems and uni-processor workstations. [color=green]The key feature of this solution will be dual-channel DDR266 support. Using 2 64bit channels will provide 4.2GB/sec memory bus bandwidth, which will allow to load with data bus of new Pentium 4 processors supporting 533MHz Quad Pumped Bus. As for its performance, Granite Bay promises to prove much faster than i850, because due to channels interleaving the latency of its memory subsystem is expected to be extremely low.[/color] Moreover, Granite Bay will support AGP 8x interface and will be equipped with ICH4 South Bridge supporting Serial ATA and USB 2.0 interfaces.

Springdale. This chipset is expected to arrive in Q2’03 and is aimed at the new Pentium 4 processors made with 0.10micron technology. It will also support dual-channel DDR memory. In this case, however, it is very likely to be about the new DDR II, as the year 2003 is still quite far away. This chipset will use ICH5 supporting wireless LAN and 3GIO. Supposedly, Springdale will be aimed at the mainstream market sector.
[color=purple]So, now we hope no one any more doubts about the memory type of the future.[/color]

You have been informed.

First point: 'Tehama-E' may not be very attractive because having a 533MHz FSB P4 with PC800 RDRAM will introduce additional latency due to the asynchronous clocking. This is not spectacular.

Second point: 'Granite Bay', with dual-channel DDR266, mated to 533MHz FSB P4's, should be an extremely attractive high-end chipset, a worthy platform for Intel's flagship desktop processor.

We'll have to see how the competition responds. I don't think I'll be building a new system before 'Granite Bay', probably not even until later, after the AMD Hammer K8 core has been released and has some respectable platforms.

Information courtesy of Xbit Labs, all emphases mine.

[URL]http://www.xbitlabs.com[/URL]

[color=green]This message has been edited. Twice, even.[/color]

(Edited by Imitation Gruel at 9:54 am on Oct. 19, 2001)

Riso

Fri Oct 19 17:32:33 2001

Actually the "new" bus runs at only 532mhz.

Because 133x4=532.

Harbinger

Fri Oct 19 19:24:23 2001

from Riso posted at 1:32 pm on Oct. 19, 2001

Actually the "new" bus runs at only 532mhz.

Because 133x4=532.

Unless the clock was actually 133.3333 (ad infinitum).

Then the bus speed would be 533.3333 (ad infinitum).

Just an old holdover from my Amiga overclocking days, when you'd change a 4-pin clock crystal to overclock (or underclock).  Used to have clocks of 33.3333, 66.6666, etc.

Riso

Fri Oct 19 19:27:20 2001

That reminds me that Asus used to o/c the A7V (or somethin) about 2-3mhz with the use of a special crystal. Everything showed up normal, but in reality it was 2-3mhz overclocked.
DrPizza

Fri Oct 19 20:24:59 2001

In fact, Intel has already cancelled its Tulloch chipset, which was supposed to support new 4-bank PC1066 RDRAM

Boo!

DuffMan

Fri Oct 19 22:51:33 2001

I'm not sure that running the memory slower than the FSB would neccessarliy cause latency. It does with via chipsets and i815, but I think i850 might have found a way arrround that. Some people who have done a lot of FSB overclocking and used the 3/4 multiplier (its there for PC600) and still gotten good memory scores.

Someone would have compare the latency of PC800 running on a 400mhz bus with a 533mhz bus with the 3/4 multiplier used.

AllYorBaseRBelong2Us

Sat Oct 20 01:23:40 2001

I'm thinking dropping of 1066 RDRAM is most likely due to Rambus not being able to make compatable chips, at least reason would dictate this.

Imitation Gruel

Sat Oct 20 02:28:44 2001

AYB: That's incorrect. Rambus defined the PC1066 DRDRAM spec quite some time ago and Samsung has it currently sampling; it'll go into full production within 6 months.
Imitation Gruel

Sat Oct 20 02:29:58 2001

So honestly I don't know why they just don't give the i850 a new memory controller as well so that it keeps the synchronicity.
Imitation Gruel

Sat Oct 20 02:30:34 2001

Seems pretty stupid to me, on the whole.

Ah well. postcount +++

DuffMan

Sat Oct 20 03:32:28 2001

Okay here's a better explanation. The i850 is a explicitly synchronis memory controller. It can run the memory at 3/4 by just dropping one in every 4 signals on the quad pumped bus. This causes no additional latency.

This is different from async chipsets like the Via Apollo and i810/i815.

Imitation Gruel

Sat Oct 20 03:56:31 2001

DuffMan: If that's so, then if someone stuck a pair of PC1066 DRDRAM RIMMs on to an i850 board, would it operate at 1/1 or 3/4?

I'm hoping 1/1, but I fear 3/4.

By the way, thanks for the information.

Imitation Gruel

Sat Oct 20 03:58:06 2001

I fear 3/4 because 4/4 or 1/1 seems like it would give better performance because it would be accepting every signal.
DuffMan

Sat Oct 20 07:49:07 2001

There is no signal to accept, at least not from the memory anyway. It might be slower because the memory is slower, but that's the whole point.

Whats also interesting is that the SiS and Via DDR chipsets for P4 are also very low latency. Maybe it's something specific to the P4.

Imitation Gruel

Sat Oct 20 08:15:50 2001

There is no signal to accept, at least not from the memory anyway. It might be slower because the memory is slower, but that's the whole point.

So, in your estimation, if you plug PC1066 DRDRAM into a 'Tehama-E' motherboard, with a 533MHz FSB Pentium 4, will the memory operate at 800MHz or 1066MHz?

My guess and fear is that it will operate at 800MHz.

Therefore I'll most likely wait for a 'Granite Bay' solution as the first realistic look at a Pentium 4. And I'd give it a few months, actually, so that the Hammer K8 core would be released. If 'Clawhammer' has some decent platforms from the get-go, it'll be up to one of the two, I suspect.

DuffMan

Sat Oct 20 22:40:46 2001

It will operate at 1066 if you want it to operate at 1066. There will probably be a 3/4 option like there is on current i850 boards.

It will be able to run at both 1066 or 800, so it can use both types of ram, while keeping the 533 bus.

Imitation Gruel

Sun Oct 21 06:06:45 2001

It will be able to run at both 1066 or 800, so it can use both types of ram, while keeping the 533 bus.

Most excellent. But I think I might wait for 'Granite Bay' anyway, because the lower latency of DDR-SDRAM should help the P4 out. SiS's 645 chipset, using a single channel of PC2700 DDR-SDRAM manages to score very close to, or utterly destroy, the i850 in most benchmarks; even with lesser and unmatched bandwidth.

So if Intel equips 'Granite Bay' with a powerful memory controller it could be a hell of a chipset.

We'll have to see how it, and the initial "ClawHammers" perform.

DuffMan

Sun Oct 21 06:20:57 2001

If this granite bay thing comes out, it will mean the death of RAMBUS as we know it. In fact, i845 is already pushing it out of OEM systems, despite the loss in performance (like OEM's care about performance).
Imitation Gruel

Sun Oct 21 06:34:49 2001

If this granite bay thing comes out, it will mean the death of RAMBUS as we know it.

While I might miss the RDRAM technology -- I find it interesting -- I will not miss Rambus, Inc.

In fact, i845 is already pushing it out of OEM systems, despite the loss in performance <snip>.

Mhmm. OEM's care about getting cheap boxes out to the customers. i845 enables cheap(er) Pentium 4 boxen with lots of <hick> dem' dar' megahurtses. </hick>