nForce Chipset Perspective

Add Your Comments

Overclockers Post:

Crush12 Chipset: $95!! Ouch!! That’s what these guys are saying. Just for comparative purposes, the average mobo chipset runs at $25-40.
Is that a ripoff? Yes and no. It’s not a ripoff in the sense that you get integrated GF2 MX400 video for your $95. It is if you have no intention of ever using such a thing and just want it for the dual-bank DDR.
I think XBit Labs is being awfully optimistic about a Crush12 mobo only costing $50 more than a Crush 11 mobo. At the beginning, at least, we may be looking at a $250 mobo.

I have to confess; I really dislike the word ripoff when discussing products. My personal bias is to reserve ripoff for the likes of the Rambus patents on SDRAM. Not even RDRAM gets that designation; even if “those people” of questionable ethics produce it.

It seems to me that there are only a couple of questions that are really interesting:

  • One: Why might The Crush 12 cost the price they are asking?
  • Two: At that price, how does that affect the system costs?
  • Three: Why are rational people reacting to this emotionally?
  • Four: Why don’t they ship the gForce 3 core if they want to capture the high end intergrated chipset business?

First question:

Why the price?

The costs of a chipset are engineering, Silicon, and packaging. On the plus side, the xBox should pay for over 90% of the design costs. Another plus is the hypertransport link between northbridge and southbridge, which should drop pin count compared to a PCI link.

On the negative, add the 100 or so pins needed to implement a second memory controller on the nForce 420, aka Crush 12. This is probably right around the number of extra pins that the AMD 760 MP needs to support 2 EV6 busses. Another cost is the additional silicon required for the L3 cache and the video subsystem.

One man’s guess, the silicon is not a major factor. Because the nForce and 760 MP have a similar pin count, they are going to be cost comparable. My conclusion is that the $100 price tag for the chipset doesn’t sound that far out of line, given the comparable chipsets.

Granted, volume is also a factor. We will be jumping for joy at a $250 760 MP and looking at $250 as a starting point for the nForce 420. However, so far, things appear rational.

To support this view on the cost difference, the nForce 220 is priced at about $45 while the nForce 440 is priced at about $90. The difference is the memory controller. Video and other features are the same across both chipsets.

Only the number of pins in the Northbridge and the controller silicon to go with it have changed. The motherboard costs are probably going to be more than $45 difference. I would guess that the cost multiplier is between 1 and 2. That would make the motherboards cost $45 to $90 more than the nForce 220 version of the same motherboard.

Second Question:

What does that do to the price?

I am going to state some assumptions here: First, a chipset that isn’t stable isn’t worth buying at any price.

Assume that the chipset is stable for discussion. Also assume that the NIC, Modem and Sound work as advertised. Now assume that the L3 cache and dual controllers deliver the 15% advertised. Assume that the earth is flat. 😉 (I know, the last is most likely.)

An AMD 760 motherboard seems to run $175 or so.
To make a system, add a NIC for $25.
Add a soundcard for $50 to $100.
Add a modem at $20 to $40.
Ignore the $75 to $100 video.

Currently, the difference between the 1.2 GHz and 1.4 GHz Tbird is about $75, and a roughly 15% speed improvement. So I am arbitrarily valuing a 15% speed improvement by the chipset at the same price, $75.

If you do the math, the 760 Motherboard, with add-ons to make it comparable to the nForce 220, would cost $345 to $415. Even without factoring the performance differences and forgetting the modem, the numbers are at about $250.

The nForce 220 looks like a good deal, and they should sell well. I am guessing that a full sized nForce 220 motherboard goes for $175. That would put about a $75 premium between the nForce 220 and nForce 420. Since the published features are the same except the second memory controller, it’s a straight cost benefit analysis.

My guess:

The nForce 420 will show some performance improvement over the nForce 220. However, it is likely to be less than 15%. It should be an interesting call for some of you. On the other hand, those building a system that are relying on the built in video, it’s a no brainier for anything other than the most basic systems.

Third Question:

Why are rational people reacting to this emotionally?

From what I have seen on the boards, many people have concluded that the major “extra” cost in the chipset comes from the video silicon that they are not planning to use. If true, this would constitute a “tax” to have a (to many) useless “feature”. As you see, I do not believe that to be the case.

The key question is: Will the second memory controller give a performance boost worth the cost difference that will exist between the boards? Only time and shipping chipsets will tell.

Fourth Question:

Why don’t they ship the gForce 3 core if they want to capture the high end intergrated chipset business?

Another educated guess: The fastest memory on the market in DIMM form is PC2100. Anything faster requires that the memory be attached to the motherboard at manufacture.

I shudder at the thought.

A motherboard with even 256 MB of 3.5 Ns DDR would be outrageously priced. My WAG would be $1000 or more. To add insult to injury, no memory expansion is possible with this design. So the question really is “What core is the best match for PC2100”?

NVidia’s answer: Gforce 2 clocked about MX levels. With PC3200 on the horizon, that answer could change over time.

Several days after I first wrote this, I found a short note in Anandtech News about the next generation nForce chipsets, which pointed to a link at Xbitlabs. The summary indicates that PC2700 will allow a Gforce3 core (at Gforce 3 MX levels) to be used in the next generation chipsets, which seems to support my assumptions about core graphics selections.

My conclusions are simple:

NVidia looks to be pricing their products higher due to real costs of features that we are interested in. The real question is: Do the features deliver additional performance to justify the additional costs? Only shipping product will answer that.

Just my opinion

Doug Lockwood

Leave a Reply

Your email address will not be published. Required fields are marked *