HKEPC released some Windows Vista Experience Index results for the Asus R50A (Atom 1.6Ghz, 2GB RAM, 12GB SSD).
CPU: 3.0
RAM: 4.3
Graphics: 2.9
Games: 2.5
HDD: 5.1

http://www.hkepc.com/database/images/2008040720153838265127211.jpg

I don't have any refrence for the CPU because i've only built dual core computers with Vista. But computers with an 630i chipset get like 5.1 on the ram (single channel PC800) and around 3.5 for graphics/games.

I have dreams of a laptop that runs cool-ish and is quiet, fairly compact but still usable, and lasts a long time on a charge.

3 years ago Intel lisenced the PowerVR SGX IP for an undisclosed UMPC project. http://www.theregister.co.uk/2005/05/17/intel_licenses_eurasia/
I think it's easy to assume this is that UMPC project they where talking about.
Still it's hard to estimate the performance of the 3D core since Imagination Tech made many versions of the SGX and Intel didn't disclose the frequency of the core.
After some googling i found out that Intel calls this part the GMA500 and it uses the SGX535 core.
For refrence, the SGX530 @ 200Mhz: 1200Mpix/sec - 13.5Mpoly/sec.
Considering PowerVR uses 'Tile Based Deferred Rendering' and the resolution on these MID's will be low (lower bandwidth requirements), my guess is that it will have higher than Geforce4 performance.

BTW if you check wikipedia for GMA500 and look at the list of devices that use it, you will see that they are all Atom based devices.

Silverthorne has two full 128-bit-wide floating-point ALUs. These ALUs and the shuffle unit also comprise a 128-bit-wide data path for SIMD integer math. The floating-point adder supports 128-bit single-precision adds, but most of the rest of the hardware is 64-bits wide, including the floating-point and integer multipliers.

Sooooo...am I reading that right? SIMD integer (MMX?), but no SIMD floating point support (SSE family)?

I got an Intel Menlow dev box a few weeks ago... Haven't had much time to play with it but seems pretty cool...

http://img212.imageshack.us/img212/590/atommg0.jpg

Great article, sounds like an interesting device.

I'm puzzled by the "Chief among those strengths is Silverthorne's compatibility with Intel's other x86 processors, which Intel equates not just with PC compatibility but Internet compatibility" part. Given that they seem to expect these chips to run Linux (which has been ported to many non-x86 platforms), equating x86 with Internet compatibility is particularly baffling. But then again, this is from the company whose marketing department claimed the Netburst architecture would make your Internet faster, so I suppose I shouldn't be that surprised...

The choice of an Imagination Technologies graphics core makes sense. Intel has been in bed with them for a number of years (they even invested in the company), and sold PowerVR-based GPUs as part of their XScale platform (before they sold XScale off to Marvell).

They are flat-out beautiful, if that can be said about a chip.
Nice article!

I can't help but think this is great logic for Blu-Ray players and the like. Fairly cheap too. Being able to program everything in x86 would be a boon to the developers.

Personally, I see this as the might behind the iTablet thing. I lust for a 8.5" diagonal iPhone-like device, with bigger battery, a bit of local storage, a port or two, and the ability to do 80% of my work on something about the size of an open paperback book.

Just saying.

So let me see if I got this straight... it outperforms an 800MHz ULV part from 4 years ago by 0-30%, glad they didnt compare it to Banias, or maybe even the Pentium Pro. Sure, it's got lower power consumption but the two shrinks (from 90nm->65nm->45nm) would have probably accomplished that (well, maybe not but it is a huge advantage). The die and package size IS great but power consumption is a tad high for the few form factors that could currently be served by this part.

I wonder how the 3.5W ULV Isaiah performs against it, since that will probably be its direct competition. Also, the price for high-end part looks quite prohibitive.

It does look great for UMPCs, though.

I'm impressed by the size of that penny. Haha.

I wonder what Shintai has to say on this.

thanks for the overview - i'm so excited about atom-based eee and *itx systems that i could spit!

It's going to be a good competitors to VIA's offerings for this market, and allow Intel to be more effective in the small devices market. The Atom-based EeePC (hopefully using the 1.6GHz HT-enabled Atom) should be a nice device.

Can you confirm that the Paulsbo chipset only supports 1GB of memory (512MB on the 400MHz FSB version that doesn't include video decode)? Shame it is 130nm as well, it's why Moorestown will have large power savings in 2010 when that portion moves to 45nm.

Is there a chance that the graphics are Imagination Tech (i.e., basically PowerVR). These graphics are used a lot in ARM designs and are apparently scaleable, and Intel did licence it at some point. Alternatively it does seem to match the performance of nVidia's AXP2500 integrated graphics (at least in video decode).

It's not competing where ARM is currently, but it will be a barrier to ARM moving upwards with their Cortex A8 cores.

Where's the overclocking page?

If they stick this in the next EEE PC with a decent amount of flash RAM I'm all in!

Dang a mystery graphics core?! It's impressive that it does dx9. I wonder how it compares against a GMA 3100.

Edit: According to Beyond3d, it's none other than PowerVR behind the 3D graphics core. Yes, the same company behind the poor-performing 3D accelerators of the mid-90s. Though they also did some decent work for Sega's AM2 and Dreamcast systems.

As i read the article all i'm thinking is "how much can you overclock the thing?".

This first version is definetly for sub notbooks not handhelds. But I'm sure someone could come out with some nice blade servers if intel allowed it.

Well I'm sure he'd nod with approval at Atom's clever layout and design.
Nice review TR.