Er wordt de laatste tijd wat heen-en-weer bekritiseert . Nu zijn Hardware Central en Firingsquad onderwerp van discussie, omdat de recentelijke Rambus artikelen van beide sites een nogal rooskleurig beeld geven van de prestaties en mogelijkheden van Rambus. JC vroeg een aantal industry insiders om gefundeerde kritiek (positief/negatief) op de reviews van Hardware Central en Firingsquad, wat resulteerde in een aantal interessante reacties waarin HWCentral, Firingsquad en Rambus het zand in worden geboord. Ik quote een aantal interessante opmerkingen:
The reasons for higher prices is: Low yields, cost of validation/testing, die size (20% penalty vs. SDRAM), packaging... and supply.1) Yields for DRDRAM are exceeding 90% right now - however, PC800 DRDRAM is below 50%. Recall that PC700 and PC600 were 'added' after the fact when manufacturers could not qualify any PC800 parts (go do a search of the news from a year or two ago). There is a reason Intel relaxed some of the validation constraints a short while ago.
2) Testing of DRDRAM is very time consuming, because it is *complicated*. Remember, we are not talking about a simple serial transmission, we are talking about a packetized communication channel. Equipment is expensive, and test setup takes time - as does the validation.
3) Die size is approx 20% larger than SDRAM, which is why Intel originally was saying there would only be a 20% cost premium over SDRAM - anyone remember this?
4) There are definitely packaging issues, however it seems that some recent packaging improvements have at least partially taken care of this. The worst situation is that the chips did not sit flush on the module, which would likely mean more breakage during handling (before the heat spreaders were attached).
5) Though demand might be exceeding supply, this in itself is a *huge* marketing ploy. If there are 100 'Tickle-me-Elmo' dolls available, and 110 people want them - you have a supply issue. If there are 100,000 people who want them, you have a high-demand. [break] De lage pincount van de Rambus interface zou theoretisch een voordeel moeten zijn voor moederplank fabrikanten, maar volgens deze engineer is Rambus door de hoge kloksnelheid en de bouw van het Rambus protocol juist buitengewoon gevoelig voor timing en signaal problemen: [/break] In reality, the higher bus speeds along with the protocol nature of RAMBUS cause the motherboard design to be MORE limited often requiring additional layers to be added to the board to allow a lower signal to noise ratio. While I personally like to see all my bus lines straight from point A to point B, it is nice to be able to move one around a bit to fit things in if needs be. This luxury is not an option with RAMBUS. It is very sensitive to impedance differences from one line to the next. The net result is that the system is more expensive. [break] De hoge warmteproduktie van Rambus chips verhoogt de produktiekosten omdat er een koelplaat nodig is om de hitte kwijt te raken. Bovendien bemoeilijkt dit het testen van Rambus chips, omdat de werking van een RIMM alleen getest kan worden wanneer de chips al gemonteerd zijn: [/break] Power and Cooling. PC133 maximum heat worst case is claimed but if all memory accesses are within the same bank on both types (worst case that engineers must design cooling for), 4.6 Watts from one RDRAM chip has a higher dissipation per unit area than 11.6 / 8 or 1.45 Watts per PC133 SDRAM chip. This is the real reason for the spreader to spread the heat to at least 4 chips in size to allow passive cooling. Since the spreader adds to chip height, larger spreading between slots is required and the spreader adds cost. This is also why full speed testing to worst case of RDRAM die must wait for packaging and incur higher costs due to wastage. Notice that the cost to a large OEM is not even estimated. This allows them to imply something that probably is not true.
Mocht je er nog niet van overtuigd zijn dat Rambus bepaalde nadelen heeft, dan zou ik dit artikel van JC zeker lezen.