AGP/PCI-e argument

[Concillian]

Originally posted by: QueBert

small portion? yes, but even if it was 5% that is A LOT of potential money. Every electronics store in my city has tons of AGP cards on the shelf, they sell. Selling = money for ATI & Nvidia.

It entirely depends on the costs to make the cards in the first place.

I work for a company that does fairly large scale manufacturing. There are times when we are asked to make parts for the repair center for warantee returns and such for older products. These parts require very small numbers compared to normal manufacturing. Our cost per unit approximately triples in order to make this small run due to conversion times and other logistics of dealing with such small runs of product.

Our normal operational mode has us running maybe 4 or 5 products a week with each seeing a considerable amount of uninterrupted time on very expensive equipment where cost per hour usage is a signficant factor. If we have to break in and convert for a short run of a different product the loss of efficiency starts increasing costs dramatically, as well as significantly reducing overall cpacity in terms of units per day or week.

I assume that the scenario is similar for other companies that make computer hardware. I also think that your 5% number is probably a very generous percentage for the number of people who both have an AGP slot and want a better card than the x850XT / 6800Ultra that are available in reasonable quantities. I make these judgements based on what I see from my own company, but neither of us have actual numbers for video cards.

The thing is, there are pretty good cards out there for AGP, just not the very top tier.

I know people here want them, but I also think these people have a complete lack of understanding of what really needs to happen to make things like this viable from a manufacturing and marketing standpoint.

 

[Concillian]

Originally posted by: CP5670

I don't understand why so many people are actually opposed to having AGP versions of current cards. It's not like anyone is forcing you to buy them if you don't want to; they just give some of us more upgrade choices. As for the added costs for the manufacturer, it could to true to some extent, but I don't see anyone saying that AMD for example should stop Opteron production and only make Athlon 64s since none of us are going to buy those and they just add to the company's costs, despite that being an equally valid argument.

Have you seen how much Opterons cost compared to A64s? They are the same chip but they cost significantly more. AMD makes Opterons because they can charge significantly more for the same thing.

The market won't tolerate this kind of thing in video cards, except for the few 'CAD cards' that exist. Ever noticed that those things always cost significantly more than the roughly equivalent desktop type card? Part of it is that they CAN charge that much because it's businesses buying them, but part of it is also the manufacturing cost increases because of the small volumes compared to their desktop cards.

 

[Gamingphreek]

They are the same chip but they cost significantly more.

Ummm not they aren't. They were very similiar to the Clawhammers but there were a few differences.

#1. The Opterons Memory Controller had to support ECC (Error Correcting Code) and Buffered Memory. It also had to be tweaked for high stability.

#2. The core itself was tweaked for stability, as well as server apps.

#3. The Opterons include more Hypertransport interconnects.

#4. The Opterons support dual, quad, and 8-way configurations, the 64's have the connects for the another processor but do not support these other features.

Im sure there are also some other redundant features which increase stability even more. I am also sure there are some other features that i do not know of.

-Kevin

 

[TGS]

Can we drop the line of how great it is for higher end functions, such as large disk arrays.

Even smaller servers are designed with discrete busses to address I/O contention on the heavily used busses. IE your typical 8-way machine will have 2-3 (66MHz/64-bit) PCI-X busses, along with a (33Mhz/32-bit) PCI bus.

Those who are already to active saturate the bandwidth on (66MHz/64-bit) PCI-X connections will and have added more busses, or use a multipath architecture to chain a large grouping of slower busses to achieve the neccesary I/O levels.

For the home user, the bump from PCI to PCIE will eventually pay dividends. Just not today. When the average consumer PCI transferred to a PCIE device requires more than 133 MB/s then the move to PCIE can fully justified. Other than that, it's strictly for SLI capatibilty at this point in time.

Originally posted by: dev0lution
PCI-E x1 RAID, SATA, Firewire and combo cards are all starting to come out now (SIIG cards for example). And personally I'm glad my PCI-E slot has the juice to power my X800XL without hijacking a molex from my power supply. How many AGP X800XL users can say the same?

This is a bit silly. The power from the PSU is a finite amount. Regardless if the draw is through the motherboard or a molex connector, the power draw is still being applied against the PSU. Unless the card is bridged, there shouldn't be any performance gains from having power supplied through a PCIE slot, over a molex.