The Intel Atom Thread

Page 100 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
Aug 27, 2013
86
0
0
The CHT coming in Q4 is known as Cherry Trail VMS = Cherry Trail Value & Mainstream, per a roadmap leaked by VR-Zone. I have heard that this version is 8EU Gen8 LP GPU while the full version is coming in Q1 2015 with 16EUs.

Even 8 Gen 8 EU is a quite a large jump up from Bay Trail. It won't excel on the highest res over 9" tablets but it should suffice for anything smaller, 1080p or even 2560x1600.
 

blackened23

Diamond Member
Jul 26, 2011
8,548
2
0
Yep. The G3220 is a beast for gaming as well if you add a dGPU. Unfortunately anything remotely CPU bound suffers incredibly on the llano or low end trinity parts, which are in the same price range as the G3220. Total price being around 100$ for the Pentium + H81 mobo.

I still can't get over how snappy the G3220 is; I just added on my HTPC setup. Very impressed. I think the 7850k has a case for itself if it were cheaper, but a total 7850k platform costs 200$ more than a G3220 (190$ for the 7850k, 100$+ for an OCable FM3 motherboard), and not all G3220 owners will want to game. But if you do, add a 100-150$ dGPU and the G3220 is still cheaper and faster still.

ANYWAY. This really underscores the fact that the embedded Bay Trail chips are way too close in price to Haswell Pentiums. Needs to be cheaper, this is certain. With that said, Bay Trail on tablets is a very compelling product - those are impressive for the price. The desktop embedded parts? I think they should be 20-30$ cheaper so that the product stack in comparison to the 3220 makes sense.
 

mikk

Diamond Member
May 15, 2012
4,108
2,099
136
G3220 also supports Quicksync with latest driver, a really capable and cheap office and internet CPU. Power consumption is 15-20W higher than Kabini under load but it's not a big deal on desktop for most users. Celeron G1820 looks quite good as well for the price, I could buy this for 31-32€. Bay Trail desktop pricing is way too high in this context.
 

Sweepr

Diamond Member
May 12, 2006
5,148
1,142
131
Nice find mikk. Celeron G1820 is one impressive little chip. Beats Kabini in most MT stuff comfortably and by a large margin in 1-2 threaded apps, faster iGPU (finally Haswell GT1 tests!), low power (a mere 8-15W more than Athlon 5350 under idle/load, quite a bit more than Bay Trail-D though). Easy choice for most desktop users.
 

blackened23

Diamond Member
Jul 26, 2011
8,548
2
0
I can't seem to find any meaningful Celeron reviews anywhere. Anyone have a link to one? Interested in seeing how it compares to the G3220 or possibly anything else for daily driving. I considered one for my HTPC setup briefly, but the web seems to be almost remiss of reviews for it; for some reason all of the LGA reviews are focused mostly on the high end stuff. I did manage to find several G3220 reviews which were all extremely positive, and that matches my reaction to the G3220 as well (I really like my HTPC ;) ) The 1820 was also only 10$ cheaper at the time , which, when combined with the fact there is a lot of info on the net about the 3220 and not so much for the 1820 - shifted my HTPC purchase towards the Pentium.

I am curious about this 1820 though. Anyone have review links with benches? Preferably in comparison to the 3220 and any applicable AMD offerings. Can't seem to find any meaningful reviews.
 
Last edited:

Abwx

Lifer
Apr 2, 2011
10,842
3,295
136
low power (a mere 8-15W more than Athlon 5350 under idle/load,

Add 10W to those "mere 8-15W" to get numbers closer to reality.

Kabini doesnt need a 550W PSU , it could have been a far more
interesting review had they used the same PSU as the 90W they
granted Bay Trail, isnt it..?

Their apparently rigged testbed :

http://pclab.pl/art57195-4.html

Amazing to see that thoses unprofessional sites can still get quoted,
wich is possible, why not , but then their methodologies are to be exposed
for what they are, that is , hardly reliable for anything else than pointing
the sites amateurism.
 

mikk

Diamond Member
May 15, 2012
4,108
2,099
136
I can't seem to find any meaningful Celeron reviews anywhere. Anyone have a link to one? Interested in seeing how it compares to the G3220 or possibly anything else for daily driving. I considered one for my HTPC setup briefly, but the web seems to be almost remiss of reviews for it; for some reason all of the LGA reviews are focused mostly on the high end stuff. I did manage to find several G3220 reviews which were all extremely positive, and that matches my reaction to the G3220 as well (I really like my HTPC ;) ) The 1820 was also only 10$ cheaper at the time , which, when combined with the fact there is a lot of info on the net about the 3220 and not so much for the 1820 - shifted my HTPC purchase towards the Pentium.

I am curious about this 1820 though. Anyone have review links with benches? Can't seem to find any meaningful ones.


In the above link there is a Haswell Celeron included. Or here: http://pclab.pl/art56815-3.html
 

Shivansps

Diamond Member
Sep 11, 2013
3,828
1,505
136
Add 10W to those "mere 8-15W" to get numbers closer to reality.

Kabini doesnt need a 550W PSU , it could have been a far more
interesting review had they used the same PSU as the 90W they
granted Bay Trail, isnt it..?

Their apparently rigged testbed :

http://pclab.pl/art57195-4.html

Amazing to see that thoses unprofessional sites can still get quoted,
wich is possible, why not , but then their methodologies are to be exposed
for what they are, that is , hardly reliable for anything else than pointing
the sites amateurism.

The Antec is using a brick, those bricks are not so good at power efficient either, anyway, the 5350 power is just 8-14W below a G1820 that is using the same PSU on that test, you cant really belive that a J1900 will use more power than that to get close to a G1820... also the thing is pasive, just stop complaining.

Rigged was the word i was thinking about while i was reading others reviews comparing it to every cpu but the ones that must be compared to.
 
Last edited:

cbn

Lifer
Mar 27, 2009
12,968
221
106
GIGABYTE GA-J1900N-D3V (Celeron J1900 quad core) available at Newegg:

(This is the third Bay Trail Mini-ITX that I know of to make it to retail)

http://www.newegg.com/Product/Produc...82E16813128698

13-128-698-01.jpg


$91.99 with $6.98 shipping.


Not so sure on this pricing. I can grab a G3220 Haswell Pentium and an H81 motherboard for around 110 bucks. SATA6G, USB3, realtek audio, gigabit LAN, whole 9, with an asus H81 mobo. Or I could lowball the mobo and get an ECS or something (not my preference, but i'm just saying) and get both at the 95-100$ mark. The G3220 is a surprisingly awesome processor for the price. Very snappy and performs great in daily tasks. IMO way better than the AMD low end stuff in that price range because frankly, Llano is a freakin dog in terms of CPU grunt. Anyway, for the price the G3220 + H81 platform certainly sounds more compelling than............this. Is this pricing just inflated or what? Because it's new? Seems like 70$ should be the go to mark for this platform, for both a mobo + CPU. Unless i'm REALLY missing something.

But as far as mobile products, Bay Trail hit the mark. I've been pretty impressed by the various 3770 BT tablets floating around. These mITX boards. I feel the price could be a tad lower for what they offer. Not because BT is bad, but because the Haswell Pentiums are really great performing for dual cores and are cheap as a whole motherboard + CPU platform.

In fact, I just built a G3220 Haswell HTPC system and was completely shocked at how snappy it was. Completely surprised and impressed. I'd consider that the competition for these BT desktop boards if anything, since they're squarely aimed at the HTPC niche. On the AMD you have Llano in that price range, or very low end richlands. Llano is a dog. Richland? Eh. Kaveri is on another planet as the 7850k costs near 200$ and the motherboards for it cost about 100$. If you want an overclocking mobo anyway. So i'll just say that the G3220 + H81 is the price performance leader in the 100$ area. And with that being the case...the above board + CPU should really be closer to 70$ or something.

I agree the current pricing is rather high. (especially considering there are only two SATA ports, both 3 Gbps, and four PCI-E 2.0 lanes native to the J1900 SOC).

I think I would rather have an Avoton C2550 consumer board. According to Intel Ark the C2550 SOC is only slightly more expensive than the J1900 ($86 vs $82) while offering higher clockspeeds (2.4 Ghz base clock vs 2.0 Ghz base clock), more SATA ports (6 vs 2) and more PCI-E 2.0 lanes (16 vs. 4). This not mention the AES-NI, ECC and integrated Intel LAN (4 x1 GbE or 4 x 2.5 GbE) on Avoton. About the only downsides to Avoton C2550 vs. Celeron J1900 (that I can think of) is that Avoton C2550 has a higher TDP (14 watts vs 10 watts) and lacks usb 3.0 and an iGPU.
 
Last edited:

Abwx

Lifer
Apr 2, 2011
10,842
3,295
136
The Antec is using a brick, those bricks are not so good at power efficient either, anyway, the 5350 power is just 8-14W below a G1820 that is using the same PSU on that test, you cant really belive that a J1900 will use more power than that to get close to a G1820... also the thing is pasive, just stop complaining.


I made measurements and an average 90W brick has about 3W losses when not loaded , i often saw 2W , on the other front ATX PSUs are never under 10-15W when iddle , that s quite a lot and largely explain both the difference with hardware.info s numbers and the obvious dishonnesty of PClab that willfully set the testbed such that the AMD system start with a 10W penalty at least when "reviewing" a 25W part, this has nothing to do with complaining , unless exposing rigged methodologies wich will contradict your sayings is seen as complains...
 
Last edited:

Shivansps

Diamond Member
Sep 11, 2013
3,828
1,505
136
The fact remain that is was tested with the same PSU as the G1820, just 8 to 15W more for a CPU+MB combo that is on the same price range, just look at this:

b4sp.png


AWFUL, really, its not worth it, i can even say it has no place on desktop, like BT, and when compared to a J1900, even if the J1900 used the same power, the J1900 is pasive, that is already a lot for HTPC and home server usage. And you cant really belive that the J1900 gona use more power than Kabini...

Instead of compalining that pclab did not used the same PSU on BT, you should be complaning that, for some "reason", other reviews did not compared it to the correct CPUs.
 

Abwx

Lifer
Apr 2, 2011
10,842
3,295
136
The fact remain that is was tested with the same PSU as the G1820, just 8 to 15W more for a CPU+MB combo that is on the same price range, just look at this:


1920 x 1080 ?...

And what about the G1820 score despite double the TDP and only 30%
higher framerate and anyway no more playable.?.
 

Sweepr

Diamond Member
May 12, 2006
5,148
1,142
131
1920 x 1080 ?...

And what about the G1820 score despite double the TDP and only 30%
higher framerate and anyway no more playable.?.

Same can be said for Kabini vs Bay Trail-D in a lot of recent titles. Except BT offers pretty much the same CPU performance with lower power consumption and can be passive cooled.

I do remember people bragging about Kabini's IGP being marginally faster than Ivy Bridge Celerons a while back though. :p
 

Abwx

Lifer
Apr 2, 2011
10,842
3,295
136
Except BT offers pretty much the same CPU performance with lower power consumption and can be passive cooled.

You keep repeating this as a gospel or some kind of auto suggestion
even you saw hardware.info s numbers that show BT being hardly
more CPU efficient and even less efficient once the GPU is used.

Here the link again :

http://nl.hardware.info/reviews/5330/amd-am1-vs-intel-bay-trail-d-review-goedkope-desktopplatforms

Of course you are free to prefer PClab numbers with their doctored
and purpotedly differentiated testbeds.
 

Shivansps

Diamond Member
Sep 11, 2013
3,828
1,505
136
5350 also has a lot more TDP than a BT-D, the G1820 is a lot better at the same price range, specially on mATX, no point working your way around that, AMD did compared it to a G1620 when first announced, suddently that stopped once the G1820 was released.

And we all knew that the J1900 was worse than a 2Ghz Kabini, there is not even a point in comparing them on performance, but Kabini need for active cooling and a little more power really kills it on many uses that the J1900 have, worse, the I/O is not that good either.
 
Last edited:

blackened23

Diamond Member
Jul 26, 2011
8,548
2
0
Indeed. You'd have to be a little messed up in the head to consider either the desktop Kabini or desktop BT over the G1820 or G3220. All of these products are in the same price range. Except both the celly G1820 and Pentium G3220 stomp the kabini and BT performance wise by such a big margin, it isn't even funny. Even better, both of the latter CPUs are transformable into great gaming setups with a dGPU. Kabini and BT...not so much.

BT is great for mobile, though. I've used a couple of BT tablets and they were great. The excellent mobile use doesn't translate into a good desktop price/performance, that's for sure though.
 
Last edited:

Vesku

Diamond Member
Aug 25, 2005
3,743
28
86
The Antec is using a brick, those bricks are not so good at power efficient either, anyway, the 5350 power is just 8-14W below a G1820 that is using the same PSU on that test, you cant really belive that a J1900 will use more power than that to get close to a G1820... also the thing is pasive, just stop complaining.

Rigged was the word i was thinking about while i was reading others reviews comparing it to every cpu but the ones that must be compared to.

Actually most 'name brand', as in not cheapest components possible, power bricks have been mostly 80+ efficient since before the PC 80 Plus standard showed up for the practical reason they lack a fan to dump excess heat with.

Bay Trail is the clear x86 for sub 10W applications but it is also clear the main thing AMD/Kabini is lacking is not in the design it is in being on GF 28nm instead of Intel 22nm. If the desire is to only have a single fan system in a small form factor Bay Trail is currently the best x86 option and it's a very nice improvement from the Via, Geode, and original Atoms.
 
Last edited:

monstercameron

Diamond Member
Feb 12, 2013
3,818
1
0
i dont know why tdp keeps getting brought up when the power draw is so similar, also what is preventing the 5350 from being passively cooled? the 5350 is a better part than the atom if not for straight performance then for modern instruction set support, avx, aes-ni and a host of others...
 

Shivansps

Diamond Member
Sep 11, 2013
3,828
1,505
136
Im starting to think that a $120 or less mITX Avoton C5550 + 820M could be increible nice for the consumer market, OEM could plug-in the 820M on x4, provide a 8x PCI-E and still have 4 pci-e lanes free too add mini-pcie, more satas, and other I/O...

i dont know why tdp keeps getting brought up when the power draw is so similar, also what is preventing the 5350 from being passively cooled? the 5350 is a better part than the atom if not for straight performance then for modern instruction set support, avx, aes-ni and a host of others...

Well, you need to think that BT is something that was designed for passive cooling from the start, BT-D is just the desktop version of it, kabini has been developed for active cooling. Same reason of why a 17W 1017U uses a fan, and the 20W Avoton C2750 is pasive with a no so big cooler.

Socketed Kabini means OEMs will be unable to come up with a workaround(a big built-in cooler), so you will have to go after market cooling solution.

Still, i feel that the Semprons will be able to go passive with stock cooling, but im also sure that J1800 and J1900 will be better than those, so it makes only sence if you need the better igp, AES and AVX.
 
Last edited:

monstercameron

Diamond Member
Feb 12, 2013
3,818
1
0
Im starting to think that a $120 mITX Avoton C5550 + 820M could be increible nice for the consumer market, OEM could plug-in the 820M on x4, provide a 8x PCI-E and still have 4 pci-e lanes free too add mini-pcie, more satas, and other I/O...



Well, you need to think that BT is something that was designed for passive cooling from the start, BT-D is just the desktop version of it, kabini has been developed for active cooling. Same reason of why a 17W 1017U uses a fan.

Socketed Kabini means OEMs will be unable to come up with a workaround(a big built-in cooler), so you will have to go after market cooling solution.

Still, i feel that the Semprons will be able to go passive with stock cooling, but im also sure that J1800 and J1900 will be better than those, so it makes only sence if you need the better igp, AES and AVX.

kabini wasn't designed specifically to need active cooling nor was baytrail for passive, that decision was based on the target IC heat output and its operating environment -the shield uses active cooling while using less power than baytrail or kabini.
That said, they are on par cpu-wise but kabini offers a few modern extensions that baytrail doesnt that can shift the balance to kabini and even though its gpu may not be super powerful it is still opencl 1.2 compliant and can be used to enhance opencl power workloads -that isn't to say that baytrail cant do the same, just not on the same level.
The power of amds gpus -not specifically hsa, or amd only- will start to shine when opencl is prevalent as well as c++amp and other solutions. Even now there are benefits, such as running a video decoder off the gpu shaders, speeding up jpeg decoding and speeding up photoshop/gimp filters.
 

Shivansps

Diamond Member
Sep 11, 2013
3,828
1,505
136
that decision was based on the target IC heat output and its operating environment

You said it yourselft, that means BT had a low target heat output and designed around that, its not the case with Kabini as every SKU has active cooling on it.
The 20W passive 8-Core Avoton C2750 is a clear example of this, the end result is that a BT-D has a real TDP of around 10W depending on SKU, and cooling is implemented according to that, and Kabini AM1 is 25W, especially for 5350, power comsuption is not a direct indicative of tdp, but i do expect that a 10W TDP will be use less power than a 25W tdp one.

AMD had to make a cooler for the 25W 5350, making it full passive it means that they need to make it big = high cost because that same cooler will be used for all AM1, so they went for a small one with a fan, very simple explanation.
 
Last edited:

Khato

Golden Member
Jul 15, 2001
1,199
231
106
You keep repeating this as a gospel or some kind of auto suggestion
even you saw hardware.info s numbers that show BT being hardly
more CPU efficient and even less efficient once the GPU is used.

Here the link again :

http://nl.hardware.info/reviews/5330/amd-am1-vs-intel-bay-trail-d-review-goedkope-desktopplatforms

Of course you are free to prefer PClab numbers with their doctored
and purpotedly differentiated testbeds.

So instead you want us to all look at a review that skews the results in the other direction? I do believe that's the definition of being hypocritical.

In case you can't figure out what I mean, take a look at the idle numbers. 9.6W for the 5350 compared to 16.8W for the J1900. That's the only reason why the 5350 looks even remotely competitive on any of the load tests and it's purely due to their choice in components for the J1900 system. If you assume a level playing field of equal idle power (Baytrail is actually better, but we'll give the 5350 a handicap) then you're looking at delta numbers from that review of 12.1W for the 5350 and 2.8W for the J1900 on 3dmark and then 16.7W for the 5350 and 4.3W for the J1900 on cinebench. Which aren't too far off from what other reviews see for the delta measurements, eg http://www.tomshardware.com/reviews/athlon-5350-am1-platform-review,3801-9.html