New Carrizo Benchmarks

[monstercameron]

It's not an irrelevant lecture and it's common knowledge. The GloFo SHP process is SOI.

shp is a bulk process not soi...I dont even think they have/had a 28nm soi node.

[Shahnewaz]

Double post. Ignore.

[Shahnewaz]

If AMD has no relationship in the first place, it's irrelevant, and as it's shown they have success with their own branded tablet others will join on. Also, Mullins was only better for heavily threaded benchmarks and graphics, two things tablets aren't intense on anyway. Next inane point?

Intel has done nothing wrong using contra revenue to break ARM's stranglehold on the tablet market, and that strategy is entering its sunset stage because Intel stopped givingn Atom chips away for free. They're sold at production cost (not covering fab R&D obviously) in the tablet sector, though in phones Intel will have to give chips away and wait on developers' hands and feet until enough of them convert for real competition.

Can you prove that AMD "has no relationship in the first place"? How do you know that? Or are you assuming stuff? Just because their products aren't used in the best of the designs doesn't mean AMD "has not relationship in the first place". Care to give a thought about other factors? Especially when AMD has a better chip?

 

As for your claim Mullins being only better for "heavily threaded benchmarks and graphics", here is a single threaded benchmark:

Can't you just read some reviews before making claims? Inane point? LOL! :lol:

You can be in denial about Intel doing anything wrong, but you already said, they gave away chips for free. As I said before, if you can't compete with a proper chip, you compete with money. It isn't hard to imagine that if Intel had a proper chip, they wouldn't have to do any of this contra-revenue nonsense and make real profits.

 

Please, have a read:

http://seekingalpha.com/article/2832556-intels-mobile-device-strategy-the-emperors-new-clothes

http://seekingalpha.com/article/2877906-intels-contra-revenue-is-it-worth-the-cost

http://finance.yahoo.com/news/why-did-intel-start-recording-010537142.html

 

I'll just throw the important bits here:

How contra revenue led to negative revenue

Let’s just assume a single Bay Trail chip costs $20. This makes 40 million “samples”—the company’s estimates to ship in 2014—worth ~$800 million. Let’s assume the cost of these parts is another $400 million. This makes a total loss of $1.2 billion because Intel is giving away these chips.

Yeah, plain giving it away. Even worse not even covering any cost (and paying OEMs to use them), since the loss is $4.2 billion for JUST 46 million chips.

 

I'll go more in-depth here:

If you watch Intel's developer conferences, or earnings reports, you get information that makes little or no sense, and obfuscates the real reasons for the failures Intel is experiencing in this segment. Stacy Smith, the CFO, has said the cost of using Bay Trail is roughly $15 more than competing products. Intel lost $6 million in this division. So, basically, he is saying that if Intel's processors were ideal for the platform, at over 100mm2, they would sell for less than $15 each? That's where you'd want to throw well over $10 billion at?

Lets do some simple math here:

First, lets look up the price for the high-end ARM processors:

http://press.ihs.com/press-release/design-supply-chain/samsung-galaxy-s4-carries-236-bill-materials-ihs-isuppli-virtual-t

http://www.quora.com/What-is-the-cost-of-a-Qualcomm-Snapdragon-chip-inside-a-modern-smartphone

https://technology.ihs.com/405783/nokia-lumia-900-carries-bill-of-materials-of-209

From those, we'll go with the highest cost price of $30.

 

As Stacy says the cost for using Bay-Trail is $15 more than the competition, that makes each Bay-Trail chip cost $45.

So, 46 million times $45 makes a total expected revenue of $2.07 billion.

Their actual revenue was a mere $202 million from these chips.

202 million makes $4.39 per chip, as Seeking Alpha correctly reports:

An examination of Hermann Eul's (the head of MCG) presentation from the 2013 Investor Meeting reveals that Bay Trail SOCs were intended for tablets from the start. So it's not as though Bay Trail was repurposed and therefore required a subsidy to make it palatable to OEMs. Even a modest subsidy wouldn't account for the $4.2 billion loss. For 2014, MCG received an average of just $4.39 for each of the 46 million tablet processors it shipped.

And you thought they sold at production costs?

 

What about R&D spending?

You see the Mobile specific investments are less than 25% of the $11 billion R&D spending.

The unique or specific R&D costs for the MCG is likely to be at the range of $800 million to $1 billion. The $800 million figure is based on Stacy's comments in the Q&A following his presentation, where he noted that incremental R&D spend is around $200 million per quarter. The $1 billion figure is derived from the diagram above (i.e it hovers around 9% or so of the total ~11 billion).

Given that broadly, the unique costs (~40% of MCG R&D spend) are at the range of $800-$1 billion - an extrapolation of the diagram above suggests that allocated costs are in the range of $1.2 - $1.5 billion.

Then where did the rest $2.498 go (after subtracting all costs and revenue)? Exactly. Paying OEMs to use Intel chips.

 

If you don't have a proper chip that can realistically make it's way successfully to smartphones and tablets (read: Tegra), no one will be interested, let alone convert. Unless you're literally throwing money at them. Now you see why even when AMD has a better chip, OEMs are forced their hand by Intel? Now you see why relationships don't even matter, when you're getting free stuff and funding from Intel? I repeat, as I said before, Intel can't compete with a proper chip, now they compete with money.

[Opcode]

If AMD has no relationship in the first place, it's irrelevant, and as it's shown they have success with their own branded tablet others will join on. Also, Mullins was only better for heavily threaded benchmarks and graphics, two things tablets aren't intense on anyway. Next inane point?

Wat? Since when wasn't Skyrim or Crysis thread or graphics intensive?

[Dabombinable]

Wat? Since when wasn't Skyrim or Crysis thread or graphics intensive?

Skyrim saw the same FPS on my Core 2 Duo E8500 running with my GTX 970 as my i5 4440 does, its can't really make use of more than 2 cores effectively. Unlike Crysis 3 which was nearly impossible to play due to stuttering (which was solved by a Xeon X5450).

[Opcode]

Skyrim saw the same FPS on my Core 2 Duo E8500 running with my GTX 970 as my i5 4440 does, its can't really make use of more than 2 cores effectively. Unlike Crysis 3 which was nearly impossible to play due to stuttering (which was solved by a Xeon X5450).

Their just le' examples of how threading and graphics performance is important on tablets. These games run just fine around 30 FPS on the Bay Trail Atoms. Although more performance will help drive up graphics quality and frame rate. So I don't see how "both threading and graphics performance is irrelevant for tablets" stands as a valid argument. Cherry Trail will be coming with 14/16 execution units which will drive up to 50% faster gaming performance. The CPU is still the same old Silvermont but will see a slight IPC gain from the die shrink to 14nm. Long story short is there's lots of people who rely on threading and graphics capabilities on tablets. Including those outside the gaming scope that use the devices for graphic design and other tasks.

[marldorthegreat]

Its pretty good considering its a mobile chip, if it was desktop it would smash kaveri

[That Norwegian Guy]

Lately, threads be like

 

[marldorthegreat]

Source

 

Seen these a few days ago but never got around to posting them. It is SiSoft so take these findings with a grain of salt. The thing that caught my eye was the iGPU performance.

 

 

Which is mind boggling impressive. A10-7850k tests for comparison.

 

 

R7 265 w/ 2GB GDDR5

• 623.68 Mpix/s (1024 SP @ 1025 MHz)

According to SiSoft Carrizo has as much compute performance as a discrete R7 265 with half the shaders. From what I have heard Carrizo runs a custom flavor of GCN 1.2. Needless to say it's quite impressive as these numbers in turn would mean that Carrizo's iGPU compute performance is more than double that of the Iris Pro 5200 and A10-7850k.

Which ones carrizo on that graph?

[patrickjp93]

Which ones carrizo on that graph?

It's not in the chart. The carrizo numbers are up top. You can visualize a bar just over twice as long if you'd like.

[marldorthegreat]

It's not in the chart. The carrizo numbers are up top. You can visualize a bar just over twice as long if you'd like.

oh yeah, my bad