The Intel Atom Thread

[cbn]

Finally found one of the socketed Bay Trail D mini-itx locally. Took me a while to get the new bios on it and after all the updates to Windows 8.1 it runs pretty good.

Heres my setup.

PC case - Lian Li pc-q25 mini Q
Mobo - GA-J1800N-D2H REV1.0 http://www.newegg.com/Product/Produc...82E16813128688
OS - Windows 8.1 64bit
SSD - Kingston Digital 120GB SSDNow V300
RAM - Patriot PSD34G13381S (2- 4GB 1333MHz SoDIMM)

I didn't need to get this large of a PSU but it was on sale and had 6 SATA cables. I will be adding 2-3 more hard drives since this is a media server build. Maybe even a Blu-Ray drive for ripping my movies.

Right now with a Kill-a-Watt tester it idles at 15.5 - 16 watts. Another guy on XBMC forums is running a 120W DC-DC powerboard/60W AC adapter and I believe Windows 7 idling at 11.5 with the same tester.

What power supply did you end up going with?

 

[Infraction Jack]

None of the consumer Bay Trail Mini-ITX boards I have seen have either a PCI-E x16 or PCI-E x8. (The two Asrock consumer Bay Trail Micro-ATX boards do have PCI-E x16 however. They are also the only consumer Bay Trail board so far able to take standard voltage DIMMs, the others use DDR3L SO-DIMMs)

http://www.asrock.com/mb/Intel/Q1900M/

If you want an industrial quad core Bay Trail Mini-ITX with PCI-E x8 or x16 I think your options will improve. For example, the I know the Supermicro X10SBA-B has a x8 slot (it is a J1900 quad core Celeron though just like Asrock consumer board.)

How would the J1900 compare to the Athlon 5350/A6 5200 in terms of performance?

 

[cbn]

How would the J1900 compare to the Athlon 5350/A6 5200 in terms of performance?

Here are some Passmark scores:

A6-5200 = 2438 CPU Marks http://www.cpubenchmark.net/cpu.php?cpu=AMD+A6-5200+APU

Celeron J1900 = 2001 CPU Marks http://www.cpubenchmark.net/cpu.php?cpu=Intel+Celeron+J1900+@+1.99GHz

 

[Kneedragger]

What power supply did you end up going with?

Rosewill Capstone 450

 

[Nothingness]

Here are some Passmark scores:

A6-5200 = 2438 CPU Marks http://www.cpubenchmark.net/cpu.php?cpu=AMD+A6-5200+APU

Celeron J1900 = 2001 CPU Marks http://www.cpubenchmark.net/cpu.php?cpu=Intel+Celeron+J1900+@+1.99GHz

And another data point on Geekbench 3: http://browser.primatelabs.com/geekbench3/compare/498904?baseline=490889

Close but it looks like the memory subsystem of the A5200 is not very good.

 

[Vesku]

And another data point on Geekbench 3: http://browser.primatelabs.com/geekbench3/compare/498904?baseline=490889

Close but it looks like the memory subsystem of the A5200 is not very good.

Uh, dual channel vs single channel. Looks like if AMD had done dual channel it would be pretty close to Baytrail's numbers. Interesting that Kabini holds its own 2GHz kabini vs 2-2.4GHz Baytrail Atom. Baytrail has the power advantage by quite a bit thanks in large part to Intel 22nm FinFet, I like that the Asrock Q1900M linked previously is passively cooled.

Is Baytrail successor going to add AES acceleration?

 

[Nothingness]

Silvermont has AES. The funny thing is that Intel fused it off from J1900. How stupid can they go with their dumb segmentation, I wonder

 

[Vesku]

Silvermont has AES. The funny thing is that Intel fused it off from J1900. How stupid can they go with their dumb segmentation, I wonder

Is there a table which shows which Bay Trail (or formerly Bay Trail ;p) SKUs have AES acceleration enabled? Intel ARK does not have that listed. Such segmentation, especially with poor documentation, is a real pain in the butt for computer literate people trying to pick out a product.

 

[Nothingness]

The info is on Ark: All Bay Trail compared.

To sum up only ultra mobile variants (Z37xx) and micro servers ones (Exxx) have AES.

 

[Vesku]

Ah, AES New Instructions is the acceleration. So Intel has decided that home servers with consumer J series motherboards, even Pentium ones, don't get to have decent AES performance. Bit of a shame.

 

[Nothingness]

Yes, ridiculous when the tablet chips have it.

 

[Cali3350]

I hope this isnt offtopic, im really not sure.

Does anyone know if cherry trail (airmont) is a IPC increase over current bay trail, or is it mostly just a die shrink with a better GPU?

 

[Nothingness]

Does anyone know if cherry trail (airmont) is a IPC increase over current bay trail, or is it mostly just a die shrink with a better GPU?

Airmont is a shrink, Goldmont will have a new micro architecture. It's likely Intel will tweak the micro architecture and slightly increase IPC, but don't expect too much.

 

[jdubs03]

Airmont is a shrink, Goldmont will have a new micro architecture. It's likely Intel will tweak the micro architecture and slightly increase IPC, but don't expect too much.

from what i have heard airmont is not just a shrink. both airmont and goldmont are improved architecture.

they need airmont to be a new arch as they'll be competing against a57s and custom armv8.

 

[Infraction Jack]

The Asrock q1900 looks nice, I will have to try and find a very small MATX case now.

 

[Shivansps]

None of the consumer Bay Trail Mini-ITX boards I have seen have either a PCI-E x16 or PCI-E x8. (The two Asrock consumer Bay Trail Micro-ATX boards do have PCI-E x16 however. They are also the only consumer Bay Trail board so far able to take standard voltage DIMMs, the others use DDR3L SO-DIMMs)

http://www.asrock.com/mb/Intel/Q1900M/

If you want an industrial quad core Bay Trail Mini-ITX with PCI-E x8 or x16 I think your options will improve. For example, the I know the Supermicro X10SBA-B has a x8 slot (it is a J1900 quad core Celeron though just like Asrock consumer board.)

X16@x1 BT has not free pci-e lanes, Supermicro one is X8@x2, Supermicro X10SBA-B is better, probably because Asrock is wasting lanes on the X1 slots.

 

[cbn]

X16@x1 BT has not free pci-e lanes, Supermicro one is X8@x2, Supermicro X10SBA-B is better, probably because Asrock is wasting lanes on the X1 slots.

Yes, that is right. The J1900 SOC only supports 4 PCI-E 2.0 lanes.

So the longer slots have to be "short stroked" electrically.

I guess it is better having the full physical slot (for supporting the card) than having a full length card inserted into an open ended PCI-E x1 or open ended PCI-E x2 slot?

 

[Nothingness]

from what i have heard airmont is not just a shrink. both airmont and goldmont are improved architecture.

That's interesting! Do you happen to have a link to an official statement or to a serious site?

 

[Shivansps]

Yes, that is right. The J1900 SOC only supports 4 PCI-E 2.0 lanes.

So the longer slots have to be "short stroked" electrically.

I guess it is better having the full physical slot (for supporting the card) than having a full length card inserted into an open ended PCI-E x1 or open ended PCI-E x2 slot?

Its easy to damage the slot by cutting it if its not factory open, so yeah, but Asrock could have easily remove 1 pci-e x1 and use x2 on the x16. X1 on x16 slot its just pointless.

Still, if there is anything good from all this is confirmation that 2 lanes can be paired to make x2, I hope OEM choose to use it if planning to add a 820M on a BT device.

 

[witeken]

from what i have heard airmont is not just a shrink. both airmont and goldmont are improved architecture.

they need airmont to be a new arch as they'll be competing against a57s and custom armv8.

No, Atom is now also on a tick-tock update cycle as Core. Silvermont was a tock, Airmont is the tick to 14nm, Goldmont will be another Tock. Sure, there might be some architectural improvements, but a tick is mainly die shrink.

Airmont will also have a better GPU (gen8, 16EUs) and a higher clock speed (up to 2.7GHz).

 

[jdubs03]

No, Atom is now also on a tick-tock update cycle as Core. Silvermont was a tock, Airmont is the tick to 14nm, Goldmont will be another Tock. Sure, there might be some architectural improvements, but a tick is mainly die shrink.

Airmont will also have a better GPU (gen8, 16EUs) and a higher clock speed (up to 2.7GHz).

you are probably correct. i did see some info pointing to at least some micro-arch improvements, but i have seen contradictory statements saying airmont is just a shrink too. at this point my caring level is low as we won't leaen anything significantly new til computex.

here is a link for today referring to airmont as a tock:

http://www.anandtech.com/show/7910/apples-cyclone-microarchitecture-detailed

 

[witeken]

I think he means that Intel did the same with 3 ticks in a row (instead of 3 tocks as he's speculating for Apple), going to 32, 22 and 14nm within 3 revisions of Atom. At least, that's what Intel's been advertising:

 

[witeken]

Intel changes the Quark processor of Edison to Atom. It now has 2 Silvermont cores at 500MHz.

http://blogs.intel.com/technology/2...-edison-technology-provide-expanded-benefits/

Edit: it's "slightly larger" though.

 

[Khato]

they need airmont to be a new arch as they'll be competing against a57s and custom armv8.

Why exactly does such require Airmont to improve? What's known about A57 thus far implies it will only have a ~30% performance boost over A15, which would be quite good if it maintained the same power consumption. However as A15 demonstrated ARM isn't necessarily great at efficiency when they're pursuing higher levels of performance, so A57 may well continue that trend.

Unfortunately we don't have any clues as to what to expect from Qualcomm, Apple, or even NVIDIA really. NVIDIA's Denver core certainly has potential, but it could fall pretty much anywhere in power/performance given what we currently know. We can speculate that Apple's next SoC will at least maintain the current level of performance, but how will it improve? If I had to, I'd guess that Qualcomm's next high performance core will likely be the slowest of the bunch simply because they seem to have been aligning roughly to the progress of the ARM cores/staying balanced on performance and power. But we'll have to wait and see.

Anyway, Silvermont is already pretty much on par with Apple's Cyclone when it comes to integer performance. (Makes sense considering that for integer it's half the 'width' but pretty much twice the frequency.) So without any improvements in Airmont it may fall a bit behind in absolute performance... but I'd be quite surprised if it didn't maintain a marked lead in efficiency. Recall that according to Anand's power testing on the iPad Air - http://www.anandtech.com/show/7460/apple-ipad-air-review/3 - the delta power consumption compared to idle for running Kraken (single-threaded CPU benchmark) is roughly 3W. Baytrail offers the same performance (or better in the case of that particular benchmark) for under half the power. Just think how much lower the power consumption will be on 14nm with Airmont...

 

[erunion]

Intel changes the Quark processor of Edison to Atom. It now has 2 Silvermont cores at 500MHz.

http://blogs.intel.com/technology/2...-edison-technology-provide-expanded-benefits/

Edit: it's "slightly larger" though.

That press release implies they pulled quark and are using Medfield now.