Slashdot Mirror
VIA Introduces the Nano Processor
Vigile writes "While the VIA Isaiah architecture had been previously discussed, the new x86 processor is officially being released as the VIA Nano. The Nano marks VIA's first 64-bit, superscalar, speculative out-of-order CPU design and is being built on Fujitsu's 65nm process technology. While direct performance comparisons are still missing, the products being released could bring Intel's Atom platform to its knees: clock speeds as high as 1.8 GHz or as low as 1.0 GHz with a maximum power draw of only 5 watts! VIA's recently announced mini-note OpenBook platform is a likely candidate for the Nano the processors but they will likely find their way into mainstream desktop and notebook computers as well." Reader MojoKid contributes a link to HotHardware's story on
the chip now known as the Nano , as well as a January interview with VIA's Centaur design center president, Glenn Henry, who
"went into fairly deep detail on what VIA had in store with Isaiah."
How long before "Nano" gets renamed because of another electronic processing device.
I'm buying a new chip because the old one was out-of-order!
0.065 m to 0.045 m
(the backwards "u" mark for "micro" won't print in the actual comment link but pastes into the texr box w/ no prob. Nerds? There are nerds here? We need micrometers and math symbols!)
mcgrew's razor: Never attribute to stupidity that which can be explained by greedy self-interest
Intel's chip has a power draw of less than 2.5 watts for the highest-clocked chip. I don't see how a power draw that's twice that amount would bring Intel's atom to its knees.
Also, I don't understand this necessity for cheesy bad-action-flick terminology ("Intel's chip brought to it's knees!") when all that has happened is a bit player releasing a product with no performance figures.
They are comparing it to Atom, but the Thermal Envelopes are far above 4W. So essentially, this is faster, but consumes more power.
So, all it takes to beat this is to release faster Atoms.
I hope this creates some competion to Ultra-Mobile Portable Device market though, having 2 alternatives is never bad. Now AMD needs to make its move.
I must admit that a 5-watt, 64-bit processor sounds pretty spiffy, but I'd really like to see how it compares to the low-power 32-bit machines that are available now.
-jcr
The only title of honor that a tyrant can grant is "Enemy of the State."
Who even knows what 'nano' means? I do, and I assume most slashdotters know too. But what about John Sixpack, the actual consumer?
Why not just call the damn thing VIA Tiny... I'm so sick of nano this and nano that. Not a day goes by or there it is, on slashdot's frontpage.
I should rush off to trademark Muon, Quark, Lepton, Meson and Positron. But seriously, the sudden movement at the bottom of the processor market highlights a seismic shift toward ultra portables. The Asus eee was the vanguard, and I suspect we'll see literally dozens of decent machines in this market segment by the end of the year. It remains to be seen whether anyone will actually make money in this segment, though. Asus set the bar low with a $299 machine and consumers are expecting to be bowled over by increasingly capable machines at that price point.
Intel has 65nm Core Solo processors (the U1300-1500) that are spec'd at 5.5 watts TPD, and they tend to be conservative on that. Now I suppose it could end up that the Via chip does more per clock than the Core Solo, but I'd want to see some real world benchmarks before buying in to that. Via has traditionally not been that powerful per clock, and Intel's Core chips are some of the most powerful per clock of anything we've yet seen.
Also reading the article, 5 watts isn't the max, 5 watts is the TDP at 1GHz. Going up to 1.8GHz you go to 25 watts. This is very similar to the Core Solo (5.5 watts for 1-1.33Ghz, 27 watts for 1.66-1.83GHz). So it seems to me this isn't really a competitor to the Atom, more to the Core Solo. However the Core Solo is a pretty impressive chip,, so to be a real competitor this will need to be as well.
Also Intel has a 45nm factory up and running full steam, with parts available retail. Currently it's Core 2 desktop components it's making, but there's no reason that it can't do these Core Solo notebook chips as well. Of course, going to the smaller process would mean even less power usage.
So we'll have to see how this chip does in real world benchmarks once it's available to third parties. However, it isn't some new part that comes in below what Intel is offering, rather it is in the same segment as their Core Solo. That means it faces some reasonably stiff competition on the performance front.
Why would this worry Intel? Not very many comparative benchmarks, but the IPC of the Nano and a Celeron-M appear to be similar (extrapolating from the bottom graph in TFA). That means a 1GHz Nano (TDP: 5W) would have similar performance to a 1.8GHz Silverthorne Atom (TDP: 2.5W). The 1.8GHz Nano has a TDP of a whopping 25W - that's Core 2 territory. Intel won't be very worried, especially since their parts are built on 45nm, so they get far more chips per wafer.
Chernobyl 'not a wildlife haven' - BBC News
but considering that all of my experiences with Via's products have been problematic at best, I will give this product the same shunning I have given their motherboard products. At least until I see a couple of years of good real world reports... Frankly I am surprised that the company lives
Well, the C7 on the Pico-ITX board apparently draws a lowly 1 watt a 1Ghz. That means that twice the address space costs 5x the power and the same clock speed.
Someone feel free to correct me if my interpretation is flawed, but I'm not really seeing this as worth it.
It appears Via has a decent product, but nothing that will cause Intel to break the crease in their designer jeans.
Invenio via vel creo
Together with its integrated encryption module this would make one bad ass VPN/firewall-box. Many low power chips in similar thermal envelope (5W or less) can manage at most disappointing 5-10 Mbit/s (like Soekris boxes), but this beast... pure speculation here, but could it reach 100+Mbit/s doing ipsec or pptp? Does Openbsd (or something else mature and secure) support Nano's encryption acceleration?
Why not a 2W x86-64 processor like the Atom? ARM may be an inherently more efficient architecture, but Intel have an awesome 45nm process and are getting pretty good at dealing with those clunky old x86 instructions efficiently.
Chernobyl 'not a wildlife haven' - BBC News
Are you serious? First of all, doesn't this sound like a certain Apple product name? Additionally, what kind of name is Nano anyway? Intel with it's noble gas Xeon and then their Centrino have a better choice of names. What marketing exec decided, "Oh we can't think of any good name, so lets just name it after the SI prefix for the chip manufacturing process." What's next, the Pico? Femto? Maybe VIA should stick with its serial-style choice of random numbers and letters. At least the C7 shows some creativity.
I believe that the nano is quite a lot faster due to out-of-order processing and more cache. Do not mix TDP with average power consumption (and btw, don't mix TDP from different companies either, they measure it differently), the C7-M ULV 1 GHz CPU has a TDP of 3.5W, which is not too far from the 5 watts of the Nano U2400.
TFS says that this new CPU is also their first superscalar, speculative out-of-order design. If they've made an effective implementation of that, they should get significantly more performance per clock out of this CPU compared to the C7.
The Nano has twice the integer performance and 3-4 times the floating point performance of the C7 per clock though. A 500MHz Nano would probably compete very well with a 1GHz C7. It might also have more aggressive idle modes (Nano gets 100mW, don't know about the C7).
I know both OpenBSD and NetBSD have had support for the earlier Via hardware crypto devices; if the new ones are sufficiently similar, the support should follow very shortly.
Pretty much every company I've worked for perfected "speculative out-of-order design" years ago.
1997 called. It wants its obligations back.
i forget
Is the TDP based on
a) 100% of usage of the CPU in applications
or
b) if every clock switched at full rate
a is less than b and Intel use (a). AMD use (b) and VIA may well be using (b) too.
After the last-but-one hackathon, pf performance in OpenBSD increased by about 100% on Soekris boards. You can also get very good VPN performance by sticking a crypto accelerator in a miniPCI slot. As the other poster pointed out, OpenSSL has supported the older Via CPUs crypto instructions for quite a while, so this shouldn't be a problem. If this implements a superset of the old ISA then it will already be partially supported.
I am TheRaven on Soylent News
The history of computers is littered with nifty idea that were only 1.5x 2x or even 5x better price-performance than the establishment. Theres too much invested in existing vendor relationships, hardware and especially software otherwise.
Intel's first Atom-based Mini-ITX board will retail for under $80 in early June: http://tinyurl.com/4pljgf It's almost what you're after...
It's still 10-100 times more power hungry than the average ARM you find in an MP3 player or mobile phone. Both chips are totally unsuitable for usage in low power small mobile devices. Intel is quite deluded if they think they have a competitor.
Actually, that will not be a fair comparison. Processors such as Atom and Nano are designed for x86 compatibility, an architecture thats almost 30 years old and inefficient as compared to current standards. ARM on the other hand is a comparatively modern RISC architecture. No doubt it finds place in about 75% of 32 bit RISC CPUs in embedded systems.
Clocks aren't switched, transistors are. You can't possibly switch every transistor at its max-rate, due to how the chips are designed and laid out; turning on transistors implicitly turns off other transistors and energy paths; you can only switch so many at a time.
Thusly, testing strategies are built at taking the chips, putting them in a thermal output capture rig, and running applications on them. Intel, AMD and VIA do tests that max out every unit individually, and apply a mathematical algorithm that estimates the maximum power that chip could ever produce, and then round it off for marketing (since numbers like 37.1629W aren't helpful, and numbers like "40W" are).
The difference comes down to the algorithm used to determine the max power, based on the kinds of tests ran. Because there's no industry standard for it, each company does it differently. AMD tends to predict their TDPs much higher, to the absolute brink of the amount of power they could possibly (physically) produce, which changes the estimates so that server manufacturers can get a better read on the kind of cooling they need. Intel tends to predict their TDPs more for the product market they're targeted at (hence, most consumer CPUs have lower TDPs, even though the difference between the consumer part and the server part is usually minimal at best, and absolutely identical at worse). Via seems to be using something more close to AMD's metric, but since they target the embedded space, it's more advantageous for them to round up rather than risk a whole lot of unhappy customers with burnt laps and/or machines failing in the field due to inadequate cooling.
That's it. Stop pulling this bullshit out of your hat every time, and look up how these companies publish this info. It's not even that private, you can find it published on their websites if you Google hard enough.
With literally hundreds of mini processors in 3d cards, why oh why do we not have risc cpu's with 100's of processors...
rather than a small handful of probably over powerful cores...
...Transmeta
TOP DSLR Cameras Reviews of the top DSLRs
5 watts? I could cool this WITH MY FACE and be fine!
Support my political activism on Patreon.
Obviously this is not twice the adress space but 2^32 times the adress space, i.e roughly 4 billion.
I'm sure intel will sell the Atom at a price that no one will be able to refuse until Nano goes away.
Stick Men
The ARM design dates from 1983, so it's now 25 years old. The R is for RISC, so it's not that novel.
Efficiency of an architecture has very little to do with its instruction set. In the case of modern x86 chips they are emulated in microcode anyway.
ARM consumes less essentially because it runs at a few hundred MHz or less, not GHz, and has features for running in embedded environment with not much memory.
"While direct performance comparisons are still missing"... you can get the indirect ones for now.
http://www.fudzilla.com/index.php?option=com_content&task=view&id=6932&Itemid=1
What can I expect, in terms of connection counts and throughput? I'd need 100Mbit/s throughput and max. 20 VPN connections, on a shoestring budget. The protocol doesn't matter, PPTP, ipsec, L2TP - whatever is the fastest (well, other than ssh, or some sort of ssl+proxy, those are not acceptable). I guess a soekris box isn't up to that, even with crypto accelerator card. Via C7 appears to be pretty fast doing AES-128/256 (nearly 1GB/s), but does it really translate to good VPN performance?
Well, in line with the 4-letter convention for naming, the next Intel small cpu will be named the Intel "Shit".
Seems rather appropriate to be able to say that you have "shit" in your computer (although saying Intel is almost as clear I guess).
Have a look.
The Nano and Atom have similar FPU performance, as you would expect from their architecture. But the Nano has the edge on integer performance, with a very efficient out-of-order setup.
Still, the Nano at that clock speed has a TDP of 25w, and the Atom has a TDP of 2.5w. And yes, you can match them up purely on integer performance (1.3 GHz U-series Nano [8W] should equal a 1.8 Ghz Atom [2.5w]), but even then Intel has the TDP edge by a multiple of 3. This does not look good for Via.
Man is the animal that laughs.
And occasionally whores for Karma.
10e9 or 1e9 ?
http://en.wikipedia.org/wiki/Jury_nullification