Via Unveils 1-Watt x86 CPU

DeviceGuru writes "Taiwanese chip and board vendor Via Technologies has introduced a new ultra-low voltage (ULV) processor aimed at industrial, commercial, and ultra-mobile applications. Touted as the world's most power-efficient x86-compatible CPU, the 500MHz 'Eden ULV 500' processor debuted at an Embedded Systems Conference in Taipei this week. Via says its chip draws a minimum of 0.1 Watts, when idle, and a maximum of 1 Watt, making it a great candidate for consumer electronics devices such as UMPCs, PVRs, and such."

Re:How does it compare?

[pslam]

How does this chip compare in performance per watt against ARM, PowerPC and the like?

Pathetically badly. Most modern low power ARM variants are in the range 0.3-0.5mW/MHz. At 500MHz you'd see them chewing up about 150-250mW. Last I checked the Via x86 chips were single issue, so it's not too unfair to compare an ARM11 (or similar) against them. Quite frankly an ARM11 will outperform the Via chip and run lower power.

The idle power figure is a joke. I can't recall the last time I used an ARM chip that idled at 100mW. More like 1-10mW. Still, it's nice to see an x86 chip get into sub-watt territory.

Of course, ARM doesn't run native x86... and that's pretty much the only reason there's such a large market for these Via x86 chips. It's also the reason you never see them in deeply embedded systems where people don't really care so much about what ISA you're running.

Re:holy cow! and their 1.5GHz is only 7.5W

[Anonymous Coward]

Actually 'pretty feeble' is an understatement, these CPUs are in my expierience _very_ slow. Also forget the mpeg-decoder onboard. Chances are their drivers don't even support their very own chip. Also the documentation for VIA-products is horrible. Sometimes even the most basic tech-sheets are wrong and they often advertise stuff that actually isn't implemented yet.

Re:How does it compare?

[arivanov]

AFAIK all of the Via CPUs are designed to be soldered onto a motherboard.

AFAIK Buswise they used to be compatible with 370. The original Eden 533 and 800 6 years back were actually available in 370 form to be used as a CPU upgrade. There was virtually 0 interest to this form factor and Via dropped it in favour of integrated MB + CPU and soldering the CPU onto the board.

I suspect that you can probably have it done in a socket 370 form. I do not see the point though as most motherboards will not be able to provide the correct voltages and most BIOSes will not have any support for it.

Re:laptop anyone

[arivanov]

Not really.

I have used every single Via CPU from the original Eden 533 up to 1.5GHz C7 and IMO the C3-C5 spec Edens are just about useful for a dedicated appliances, small firewalls, small specialised servers and such. They do not have enough grunt for a laptop. The fact that most of them have are shipped bundled with relatively weak video does not really help either. Even the mpeg support on some motherboards cannot really help. Xterm is probably the most you can do with them as far as clients are concerned. Still better than similarly clocked Crusoe though (now that is a drag of all drags).

C7 is a completely different beast. This is probably the best CPU for a corporate laptop out there at the moment. A laptop is worthless without a "link to the mothership". Intel Core and AMD have to use CPU resource to do all of the encryption and decryption. This may amount to 30-40% of your CPU on a 54G wireless lan. Compared to that Via C7 has hardware AES acceleration so you can actually protect your traffic properly while using less than 1% of your CPU. It also has enough grunt to run most common road warrior apps at acceptable speeds. It is a pity it is not available as a laptop choice anywhere outside the far east.

On a somewhat related note

[value_added]

Soekris is now shipping a New and Improved product, the net5501 [soekris.com]. Early reports suggest that this is their first product that's able to route at line speed. I have two on order that I should receive next week.

The release of Vista suggests that we need more and more powerful systems to do our work, but the irony, at least for me, is that I keep buying more of the little guys. Being able to use fanless cases and/or flash drives is a definite selling point, but there's a surprising amount of processing power available in such products and their uses are as limitless as your own imagination. Besides, hacking those ubiquitous blue boxes can never be as satisfying as building your own.

The VIA units I own could be described as underpowered, but having onboard MPEG decoders, for example, can make up for the shortcomings.

Re:laptop anyone

[arivanov]

Why not. If you are running an Intel Centrino or Core laptop you most likely having less most of the time.

Centrino as well as any Core derived notebook under Winhoze uses voltage and frequency scaling. It will ramp up to its spec-ed frequency only when pushed really hard and in some laptops only when on AC power. If you want to actually have reasonable battery performance on Linux you end up doing the same using the cpufreq susbsystem. Example from a Core Duo on which I am typing this post:

model name : Intel(R) Core(TM)2 CPU T5600 @ 1.83GHz
stepping : 6
cpu MHz : 229.167

Note the actual CPU frequency above (this is using ondemand kernel governor). It is more than twice less than 500.

Re:Cool!

[Alioth]

ARM11 is already better for that kind of application - much lower power still, and for embedded stuff, the need for x86 compatibility really doesn't exist.

Re:laptop anyone

[Ed Avis]

Hardware AES? Can OpenSSH use it?

Re:holy cow! and their 1.5GHz is only 7.5W

[Professor_UNIX]

Also forget the mpeg-decoder onboard. Chances are their drivers don't even support their very own chip.

Stop spreading bullshit FUD. The MPEG2 decoder hardware has been supported for years now in open source. My MythTV frontend, a Via EPIA M10000 running at 1 GHz uses the MPEG2 decoding hardware when playing back video saved from my backend's Hauppauge PVR 250 hardware mpeg2 encoders just fine with very little CPU usage. The only problems arise when you try playing DivX or MPEG4 streams.

Re:I wish mainstream CPUs / GPUs would focus on po

[DrXym]

Ireland sets the rate of annual motor tax based on the size of the vehicle engine. Someone with a 1.6 litre engine pays over a hundred more euros than someone with a 1.3 litre engine. It's probably explains why SUVs are quite scarce in Ireland. Which isn't a bad thing at all.

Re:How does it compare?

[dreamchaser]

Wow, flashback to the 90's. There is really no such thing as RISC or CISC anymore. Even massive general purpose CPU's like the x86 family use cores that are basically RISC by the classical definition, at least at the microop level. Conversely, today's RISC processors have instruction sets that have grown considerably in complexity since the days of true RISC chips.

Your premise is correct that it is an apples to oranges comparison, but not really for the reasons you describe.

Re:laptop anyone

[RAMMS+EIN]

``The next question is, of course, (1) is there any power-efficient hard disk, and (2) is there any way to make sure the disk will not be spinned up more then once a few hours.''

(1) Yes. Take a look at 2.5" drives used in laptops, for example. You could also use flash instead of an actual disk. Having done that myself, I must give you a word of warning: don't do flash+usb on Linux. It will hang because of I/O errors every few days. I believe this is due to there being a hardcoded limit on the number of writes somewhere in the Linux drivers involved in this, but it could also be the flaky hardware (VIA SP8000E, never buy that one).

(2) Yes. Run off a ramdisk, and just write changes to your real disk every few hours. Puppy Linux does something like this.

Another thing you will want to look at are efficient power supplies (particularly, ones that are efficient at the low power draw of your machine).

Finally, this being about VIA, you will have to be careful with enabling CPU frequency scaling. The board I have is known to crash when the frequency is changed too often.

Re:How does it compare?

[pslam]

I am very much aware that comparing power:mhz is often an incredibly bad thing to do, but I am also very much aware of the internal architecture of these two cores.

• Via is single issue like ARM11. (At least, last I checked, but maybe they've changed that)
• ARM1176 (a common variant) has hit-under-miss caches, some SIMD extensions (slow compared to Via, though), and a decent FPU (optional). No write order dependency issues on ARM. Very comparable to Via.
• ARM11 has much lower memory latency than Via. It only takes 3 clocks to drag something out of memory while execution continues, you get write-to-read forwarding and a lot more registers which means less memory hits overall anyway.
• The bus speeds of both are comparable. External memory is implementation dependent, but can also be matched.

So they're actually pretty similar in a lot of ways. I guess I should restrict my usage of "better" to the usual target market: ARM11 is a better core at being a web browser or general multimedia device. It'll do the same job at lower power. In addition, you could run an ARM11 comfortably past 600MHz without a fan or even a heatsink. If you felt like splashing out, you could get a 4 core variant that would cream pretty much anything Via offers. Obviously 4 times the peak power, but still the amazing idle power consumption most ARMs give you.

Again, x86 is still the obvious choice in the market due to a SOFTWARE problem. I say "problem", but what I really mean is the unwillingness of vendors to write portable code and realise that recompiling for multiple platforms stopped being a major headache about 10 years ago.

It's pretty sad that an ISA is still a barrier to this kind of thing. There is a major example in the volume market where this did work out, though: Apple switching to x86. Hell, they even had to switch endian and everything worked. If anyone ever tells you that switching ISA is prohibitive, go point them at Apple, because it worked out damn well for them, and you can always use Apple as some kind of Godwin's Law because nobody will ever argue they're wrong.

Re:laptop anyone

[arivanov]

By what I recall - no, but I have not looked at the OpenSSH latest and greatest (it has been 6+ months since I looked at this).

The reason is that at least as of the versions present in major distros openssh does not for some reason support openssl engines. AES (and RSA in latest Via CPUs) is done using an openssl engine which has to be initialised and loaded. This can be done for OpenVPN, Apache, Pound and nearly any other piece of software using OpenSSL, but not OpenSSH. For some reason Theo's people in their infinite wisdom left that part out (it is trivial). There was a patch, dunno if it has made it into the main tree.

As far as non-OpenSSL software is concerned, the kernel itself can use the hardware AES for filesystems and IPSEC. I have run it for quite a while for both OpenVPN and IPSEC. It can run around a Dual Xeon in circles. I would expect it to have the same killer performance for encryption of filesystems and encrypted backups as well. In fact this is possibly the only CPU on the market at the moment where having all of your data encrypted is a realistic proposition. The rest will choke on it and crawl like a 486.

There is also further improvement from having true on-CPU hardware RNG which all programs in need of good random numbers can use as it is implemented at kernel level.

Probably the highest praise to it is the fact that most of these features have now started showing up on Intel roadmap documents for the future x86 CPUs destined for the embedded market. It is the Athlon history repeating. When someone else is doing something right Intel copies it, claims innovation and launches a marketing salvo trying to lie that "they have been doing it all along".

Re:How does it compare?

[letxa2000]

Why would you require a certain clock speed?

In true embedded applications (which are built with microcontrollers, not processors), it's usually critical for establishing baud rates. A certain clock speed also may be necessary if you use a single clock source to drive multiple parts on the board which can reduce electrical noise caused by multiple clocks flailing aroumd.

Apart from that, there is no reason to prefer a 100MHz part to a 50MHz one that can do twice as much work per clock cycle.

I agree. That's why I agreed that MIPS vs MIPS and watts vs. watts is a better comparison. I simply pointed out that in real embedded work, clock speeds are often a design criteria. You still look for the best MIPS and power consumption you can get, but constraints on the clock speeds often reduce the field of viable options even if another part might have better MIPS or better power consumption but doesn't meet your clock speed requirements.