Motherboard, and I/O support.
If you have read my analysis of the Instruction Set then you will, no doubt, be aware of the fact that the Crusoe is not your average run of the mill CPU. It has its own complexities and its own simplicities, which means that the motherboard the Crusoe is to be used with will need to be properly designed to deal with them. For obvious reasons I have not been able to get a lot of information on the motherboards that will be supporting the Crusoe, since both Transmeta and the motherboard manufacturers are keeping quiet until a product is available on the market. For most of this section I will assume that the system in question is a laptop computer (as opposed to a touch sensitive multimedia internet devices with no hard disk) as this gives a more relevant comparison to the competition from Intel and Apple.
The memory on the motherboards that will support the Crusoe will be similar to the memory currently in use today. It will (as far as can be deciphered from Transmeta's white paper on the subject) be normal SDRAM. However, there is a difference between normal SDRAM and the SDRAM for which the Crusoe is (primarily) designed to interface with: the Crusoe is designed for DDR SDRAM. The DDR stands for Double Data Rate and is almost exactly what is says: normal RAM that's twice as fast. The Crusoe processor has a specialised controller built into the chip, to deal with interfacing to the DDR memory.
DDR memory is described by Transmeta as being: "a new type of SDRAM memory that is both higher performance and lower power than standard SDRAM" and also; "available in Q3 of 2000". They got it right about the RAM being higher in performance and having lower power consumption, but their claim about its availability is wrong. I think that this claim was made in order to sell the product, which can only be expected. DDR SDRAM is available now; in fact the computer that this report was written on uses this memory on its graphics card. In my opinion DDR memory is a good idea. The way it works is to trigger on both edges of the clock signal in order to increase the clock speed to the memory. This is done because even though the integrated circuits can handle the higher clock speed, the PCB tracks (Printed Circuit Board) cannot. This means that just increasing the clock signal sent to the device is not an economically viable option. It is an interesting solution to the relative slowness of today's large memories. It is even more interesting that Transmeta have put in a piece of hardware to deal with it natively. Although I could not find any information on the subject, I think it is fair to infer that Transmeta are not initially going to go for RAMBUS compatibility as Intel has done. This is interesting as RAMBUS is faster than SDRAM but it is also very expensive. In my opinion Transmeta have chosen wisely, because the price of RAMBUS memory is predicted to remain very high in the near future and their products are not aimed at the top of the range performance computing market.
Something that I have been very interested in is the fact that Transmeta have decided to integrate the Northbridge into their Crusoe processor. The Northbridge is the interface used to bridge between the CPU and Memory and the I/O busses (such as the PCI bus). Usually the Northbridge was part of the chip set on an IBM compatible's motherboard. Integrating this into the silicon of the Crusoe means that memory and devices interface straight to the processor, which increases the bandwidth and the response time to any devices that use the Northbridge. This has obvious performance advantages, but also helps to reduce the power consumption of the system. This is because the Northbridge chip was usually a very power hungry chip on standard motherboards. This is a good technique, which should help to give the Crusoe a performance boost as well as an efficiency boost.
One issue that motherboard manufacturers will have to be aware of is, the successful implementation of a motherboard that will support Transmeta's Code Morphing Software. Transmeta have used their Code Morphing Software to deal with the complex issues associated with memory mapped I/O in a x86 compatible system.
First of all, the motherboards will need the correct Flash ROMs on them to store the Code Morphing Software in. The motherboards are also supposed to have sixteen megabytes of special fast access RAM to store the code translator, and also to store the data structures that are associated with doing the translate from x86 to the native VLIW instructions. One of the Transmeta engineers said that this could be considered as a large L3 cache for sole use by the Code Morphing Software. This is not a major problem for motherboard manufacturers, but it is an intricacy that needs to be pointed out.
Except for the "L3 cache" for the Code Morphing Software the Crusoe requires no special allowances to deal with memory mapped I/O. All of the intricacies of memory mapped I/O are dealt with by the Code Morphing Software. The software knows what parts of the address space belong to DRAM and what belongs to I/O, so it loads and translates more aggressively from DRAM and decides whether or not to cache translates and optimise loads from I/O space depending on the address range. This means that it will uniquely tailor it's parameters to an I/O device on a case by case basis. This sounds, to me, as though it could be a major advance by way of keeping I/O devices working at their limits to get better performance from a system.
Graphics are important to end users and as such they are relevant in this section, since most laptops use on-board graphics cards that are integrated into the motherboards (along with a standardised LCD graphics controller). From questions posed to the Transmeta engineers at a press conference: the gist of it is that the Transmeta has no issues with any graphics cards that have been tested to date. They said that there will be an emphasis on the power efficient graphics cards and that Transmeta will give preference to companies who concentrate on reducing the power consumption of their card. There have also been rumours, mostly unsubstantiated, that NVIDA and Diamond Multimedia will be involved in the graphics side of things.
My last topic of discussion is about the LongRun technology used by the Crusoe processor to conserve power. This is implemented only on the high-end version of the Crusoe, the TM5400. It works by monitoring the precise performance level required by the system and adjusting the Voltage and clock speed as is required. With LongRun the Crusoe can make adjustments while applications are running and so make the most efficient use of power, for a longer battery life. Here is a table of the levels that the Crusoe TM5400 will run at:
This diagram shows the huge range of different performance levels that are accessible by Transmeta's LongRun technology. I can only assume that this has some issues as far as the motherboards are concerned, since they will have to provide some mechanism for allowing voltage and clock adjustments. Again, the motherboard manufacturers are not making waves about it so it can't be much of a problem. Again, I think that this is a great idea. It is supposed to save 30% of a battery's life over a days use !!!
To conclude this section, I think that there are several features of the Crusoe that make will make it appealing to motherboard manufacturers. The power saving techniques do not seem to have any real design issues that need to be worked around in order to get a solid mainboard for the Crusoe. This could be an important factor in the high-end market because in the past AMD's sales for the K7 were crippled by the incompatibility of its motherboards with some popular peripherals. Even although there have been no confirmations from motherboard manufacturers (to date) I think the future for the Crusoe looks very bright, on the motherboard side of things.
