Matose benchmarks Intel Z3770 monster tablet SOC

[Fottemberg]

I know that the Intel 22nm yield is not great. Good, but non great like the 32nm. I don't know the exact percentange, but it's the rumor. However, the price of the Z3770 is very low. I hope that Intel gains market shares.

[HKPolice]

Intel is NOT selling at a loss, they've had to idle fab capacity due to weak demand before: http://www.anandtech.com/show/6378/i...e-fab-capacity so it's not like they're building a brand new fab just for these chips.

Check out any BOM for a high end smartphone and you'll see that their quadcore 28nm ARM SoCs sell for $30-40 max. The margins on those chips are lower if made by TSMC since they have to make a profit too, and yet Qualcomm is still able to make billions every year selling at those prices.

[Russian]

.....they should have tested against temash 4 core soc, instead of the 15w kabini. once again review sites fail to show proper apples to apples.

Just an fyi. There aren't many Temash systems out there and AMD didn't sample any Temash devices to reviewers.

[Fottemberg]

Intel is NOT selling at a loss, they've had to idle fab capacity due to weak demand before: http://www.anandtech.com/show/6378/i...e-fab-capacity so it's not like they're building a brand new fab just for these chips.

Check out any BOM for a high end smartphone and you'll see that their quadcore 28nm ARM SoCs sell for $30-40 max. The margins on those chips are lower if made by TSMC since they have to make a profit too, and yet Qualcomm is still able to make billions every year selling at those prices.

The masks, the lithographic processes, etc. are cheaper for the 28/32nm planar node than for 22nm FinFET ones. Intel makes a special effort with these low prices, without a doubt.

[Monstru]

And some Silvermont arch details if anyone is interested...

[HKPolice]

The masks, the lithographic processes, etc. are cheaper for the 28/32nm planar node than for 22nm FinFET ones. Intel makes a special effort with these low prices, without a doubt.

Yes but that doesn't matter because Intel has their own fabs. If they had to order the chips through TSMC/Samsung like Qualcomm or Apple then the margins on a $37 chip will be lower because the fabs need profit too. For sure a 22nm SoC selling for $37 won't be as profitable as a $150 22nm desktop chip but there is no way Intel is losing any money per SoC because Qualcomm & Apple are still making billions selling chips made from other fabs that cost them more than Intel.

[Fottemberg]

Yes but that doesn't matter because Intel has their own fabs. If they had to order the chips through TSMC/Samsung like Qualcomm or Apple then the margins on a $37 chip will be lower because the fabs need profit too. For sure a 22nm SoC selling for $37 won't be as profitable as a $150 22nm desktop chip but there is no way Intel is losing any money per SoC because Qualcomm & Apple are still making billions selling chips made from other fabs that cost them more than Intel.

The problem is exactly that: Intel has its own FABs. Intel have to pay the new lithography machine, the R&D of the new processes, etc. These are very expansive. So, when Intel sells its own chips, have to gain enough money for these achievements. Qualcomm, AMD, Broadcom, Mediatek, Rockchip, and the other companies are all fabless, so they CAN sell with low margin without risk. The only things that they have to pay are the design chip and the mask, mainly.

[Aberration]

The problem is exactly that: Intel has its own FABs. Intel have to pay the new lithography machine, the R&D of the new processes, etc. These are very expansive. So, when Intel sells its own chips, have to gain enough money for these achievements. Qualcomm, AMD, Broadcom, Mediatek, Rockchip, and the other companies are all fabless, so they CAN sell with low margin without risk. The only things that they have to pay are the design chip and the mask, mainly.

All of those costs are built in to the price a foundry charges.

[qcmadness]

Just notice the comparison with Jaguar cores - almost same perf at half power. AMD is getting a beating across the whole CPU segment - their GPU is holding them in the game (the consoles that is). ATi was a smart acquisition
On a sidenote, I am really excited to see what Merrifield can do - maybe Snapdragon 600 level of perf ?

AMD uses TDP as the power consumption numbers.
Intel uses SDP as the power consumption numbers.

Not a direct comparison.

[kromosto]

yes review handled this intel says 2w review says they measured 2.5w. still cant compare this measured value with amds tdp but it is enough to say better then amd

Sent from my GT-N7000 using Tapatalk 2

[Fottemberg]

All of those costs are built in to the price a foundry charges.

Not exactly. If Z foundry has a bad XX nm process, Y company goes to GloFo, UMC, Samsung, TSMC, etc. Or, if Z foundry has an expansive XX nm process, Y company goes to GloFo, UMC, Samsung, TSMC, etc.

If Intel has a bad/expansive XX nm process, so ... it's a thorny problem for its own products.

[qurious63ss]

Not exactly. If Z foundry has a bad XX nm process, Y company goes to GloFo, UMC, Samsung, TSMC, etc. Or, if Z foundry has an expansive XX nm process, Y company goes to GloFo, UMC, Samsung, TSMC, etc.

If Intel has a bad/expansive XX nm process, so ... it's a thorny problem for its own products.

Hmm interesting that you say that since it was Glofo's process issue that delayed Amd's Trinity, also Nvidia has had issues with foundry delays and yield issues. Once you commit yourself to someone's process it is not a simple matter of just moving it over to someone else's.

[Iconyu]

Not exactly. If Z foundry has a bad XX nm process, Y company goes to GloFo, UMC, Samsung, TSMC, etc. Or, if Z foundry has an expansive XX nm process, Y company goes to GloFo, UMC, Samsung, TSMC, etc.

If Intel has a bad/expansive XX nm process, so ... it's a thorny problem for its own products.

You design a chip around one persons process, if you pick a fab and their process is bad, you don't really go anywhere else. One fab's XXnm isn't always the same as another fabs XXnm.

Intel are making 12 core Ivybridge Xeons on this process, there is no way the yields are bad enough for these chips to have a problem.

[Fottemberg]

Sure, so Y company chooses the FAB 2 o 3 years earlier. If the node is bad ... sh1t happens.
For this reason, Qualcomm uses TSMC, GloFo and UMC.

[Iconyu]

Sure, so Y company chooses the FAB 2 o 3 years earlier. If the node is bad ... sh1t happens.
For this reason, Qualcomm uses TSMC, GloFo and UMC.

I think it's more to do with none of those companies would prioritise Qualcomm over Apple, AMD or Nvidia. Qualcomm are nice, but they are middle men.

[Fottemberg]

I think it's more to do with none of those companies would prioritise Qualcomm over Apple, AMD or Nvidia. Qualcomm are nice, but they are middle men.

Apple is a top customer for the foundries like TSMC and GloFo. NVIDIA is not a top customer today, as some years ago (the 20nm only low voltage node of TSMC is a proof of that). Today the only top customers are the SoC producers, like Qualcomm, MediaTek, Apple, etc.

[qurious63ss]

Can we get back on topic? Let's talk about performance here and take the money talk to a Wall Street Journal forum....please.

[Russian]

Just wanted to say, I tested this SoC as well and it'll be a great offering for Windows 8 tablets, but Android still needs a lot of work.

[Aberration]

Can we get back on topic? Let's talk about performance here and take the money talk to a Wall Street Journal forum....please.

If we were only to talk about performance, there would be a few benchmarks and no one would have much to talk about. You could easily just move along.... please.

[Aberration]

Not exactly. If Z foundry has a bad XX nm process, Y company goes to GloFo, UMC, Samsung, TSMC, etc. Or, if Z foundry has an expansive XX nm process, Y company goes to GloFo, UMC, Samsung, TSMC, etc.

If Intel has a bad/expansive XX nm process, so ... it's a thorny problem for its own products.

Non-sequitur. You specifically spoke of litho tool cost, R&D cost, mask cost. That will all be factored into what the foundry charges. A foundry may adjust prices to reflect competition, but all those costs are still factored in. This is economics 101.

Intel has yet to have a bad process, so no point in speaking about what-ifs.

[qurious63ss]

If we were only to talk about performance, there would be a few benchmarks and no one would have much to talk about. You could easily just move along.... please.

What are you talking about, there are several reviews out and plenty to talk about performance wise. Just because Fottemburg derailed the thread six post into it and you decide to follow doesn't make it ok. Besides, read some of Fottemburg posts in other threads and tell me he doesn't have an agenda to make intel look bad.

Intel is short of money?

Bad times incoming for Intel?

Without 14nm, ARM will be the winner.

Intel in trouble?

14nm postpone to early 2015?

If Intel miss a CPU today, could be in trouble.

[qurious63ss]

Just wanted to say, I tested this SoC as well and it'll be a great offering for Windows 8 tablets, but Android still needs a lot of work.

Can you elaborate on this? How much of a difference do you think it is percentage wise? Also, is there something specific that needs work or is it just overall?

Thanks.

[Fottemberg]

qurious63ss, you have reported only the latter end of my posts.
I don't have an agenda to make intel look bad, because I only speak about the foundries. If I want to do this, I should speak about the CPUs and SoCs.
Foundry-themes are not so popular, eheh.

[zalbard]

2008: Z560 (Silverthorne) Die Size about 26 mm2 @ 144$
2012: Z2480 (Medfield) Die Size about 65 mm2 @ about 80$
2013: Z3770 (Bay Trail-T) Die Size about 105 mm2 @ 37$

Looks like someone's never heard about the term "manufacturing costs".

Intel makes a special effort with these low prices, without a doubt.

Mark up may be lower. Competition is good for us.

[Fottemberg]

Looks like someone's never heard about the term "manufacturing costs".

Today the manufacturing costs are growing up.

Some links, since you are skeptics:

http://www.semiwiki.com/forum/conten...t-forever.html
http://www.eetimes.com/document.asp?doc_id=1263318
http://www.eejournal.com/archives/ar...19-transistor/

22nm is very expansive, although Intel doesn't say it.