Via pulls winner out of wraps
Computex 2005 High part of the low-end mobile-market chip
By Charlie Demerjian in Taipei: onsdag 08 juni 2005, 02:29
VIA FINALLY PULLED THE wraps off the C7-M CPU - its new high part of the low-end mobile-market chip - and I think it has a winnner.
This chip is not targeted at a Turion 64 gaming box, nor does Yonah have much to fear here. Companies making things for a thin and light, long battery life mobile notebooks have a new contender though.
This chip appears to be aimed at the Celeron M line. Basically it will be more than adequate for doing work, like me typing this on a plane over the Pacific. It will also be cheap, the die is a mere 30 square mm, or about 1/3rd the side of a PM, and that translates into low cost.
The chip also seems to be a decent performer, especially if you consider that it has some of the nicest features of the previous C3 chips. People often overlook the fact that the last of the C3s had massively powerful security primitives in hardware, they would give an A64 a run for it's money in encryption heavy tasks. This was collectively called Padlock.
They also have MP4 in hardware, and if you have a program that uses it, you can do video tasks with almost no CPU overhead. Together, they add up to doing two of the most compute intensive things on a PC with almost no CPU use. Raw horsepower is not the only game in town here.
So, know that you know where it is aimed, you probably want to know the techy bits. As I said early, the chip called VSC7-M in more formal terms, was code named Ester, and is fabbed by IBM on it's 90nm SOI process. The 30mm die size is amazingly small, and the package size follows this trend also.
If you have ever played with a VIA C3, you know how small they are. C7 is no slouch in this category either, the NanoBGA package is just 21mm on a side. This is really really tiny, you have to see it to realize how clunky a P4 is.
One of the unique ideas behind this chip is two PLLs for fast speed and power switching. The C7 runs two PLLs , marketing name Twin Turbo, one at the high clock speed, one at the lower one. When you need to swap frequencies from a high use, high power mode to a lower one, you can do it in a single clock cycle. In contrast, they said the PM takes about 15K cycles to do the same, an eye-blink to be sure, but agonizingly long in comparison.
The power savings are fairly large, but as usual with this kind of thing, I'll wait for a real word test before I say anything conclusive. The TDP of the chip is set at 10.1W, an extremely low number to be sure. By comparison, PMs are about 25W for Dothan, 31W for Yonah, and 45W for Merom. Turions are 25W and 35W respectively.
The problem with quoting TDP numbers is that for laptops, it is pretty irrelevant. How many times do you run you laptop flat out on batteries? If you design mobile cooling systems, OK, you care, but everyone else has something else they should care about.
What matters is average power usage, and here, C7 is commendably low. The usage stats I was shown had the 10.1W TDP as a hard ceiling, it would not go any more period. For the most part, it did not need to, just the occasional spike up that high, the rest were in the 1W range. Average power usage over the graph was 1.4W.
It could play DVDs at the lowest power mode, 600MHz@.908V, and would ramp up to 1.5GHz@1.1V when needed. The chip supports all ACPI states (C0-C4), and pulls 750MV at its lowest point.
The heat production, it is again quite low. During the DVD playback, CJ removed the fan, then the heatsink, from the CPU. It ran cool to the touch, and the DVD did not skip. When a high load was put on the chip, again without heatsink, it would get very hot very quickly, and throttle at 70C. Now, if you keep your heatsink on, this CPU should live quite happily at low temps no matter what you do.
One other bit that most people don't think much about, if you have a low power chip, you can get away with a smaller heatsink. For laptops, that means less weight and less cost. If you don't need a fan on it, you can again save money and power. Low power CPUs don't just directly affect battery life, they also affect many other components that support the removal of the heat.
Let's wrap this up with some more stats, C7-M was made with about 85 engineers, not exactly a huge team. With this, they made a chip that has SSE1-2-3, AMD64, and the NX bit. It has 64K L1I, 64K L1D and 128K L2 with a 32 way victim buffer. The FSB is currently 800MHz, and 1066 is doable if needed. The chips are 1.5GHz now, with 2.0 soon, and higher if they want to go there. Volume production should start at the end of the month.
If the performance numbers and power use pan out, VIA will have raised the bar a bit. They claim a 15% performance per watt advantage over the PM chips, and that not a low target to aim for. These chips, if priced as low as VIA usually does, have the potential to make radically lower cost laptops, I don't think a $500 Linux based notebook is all that out of reach. µ