AMD CPU speculation... and expert conjecture

[sarinaide]

^ and jaguar will steal this title from a4

If it comes out in desktop trim sure.

 

[JAYDEEJOHN]

http://www.fudzilla.com/home/item/31264-amd-creates-custom-chip-division

Looks like AMD has setup a shop to do more custom solutions like PS4. Apparently, they will offer integrated x86 and ARM based APUs.

That's catering to a niche that doesn't really exist from a manufacturer yet. Creating a custom solution without having to burn copious amounts of R&D cash. Maybe that will take off, maybe not, but the idea has some merit.

See:
ARM
And no interference from Intel, unless they once again overstep their bounds

 

[Cazalan]

http://www.fudzilla.com/home/item/31264-amd-creates-custom-chip-division

Looks like AMD has setup a shop to do more custom solutions like PS4. Apparently, they will offer integrated x86 and ARM based APUs.

That's catering to a niche that doesn't really exist from a manufacturer yet. Creating a custom solution without having to burn copious amounts of R&D cash. Maybe that will take off, maybe not, but the idea has some merit.

Sounds like a costly endeavor, with few companies able to foot the bill for custom work like Sony/Microsoft. The chip costs don't get much cheaper. It's still 3+ mil per mask. The more chip types you bake the more custom chip tests/testers you need to make.

If they could secure government orders that might work. They have near limitless cash stolen from the tax payers.

Or is it just a clever way to spin a new R&D arm to the investors? That I could see being ultra important. Modeling future devices with vast configurations of cores and graphic compute units.

 

[8350rocks]

I could see it working both ways. Say they cut the R&D costs by 50% on a custom solution with funding from an outside source for half. Then that company essentially saved themselves 1/2 the cost, AMD cuts the cost in half to develop it and gains some proprietary control over the internal architecture developed to fill that niche.

 

[anxiousinfusion]

The A4-4000 Richland, in comparison to the A4-5300 Trinity, appears to make slight gains in 3D applications while consuming less power. The A4-4000 and A4-5300 can be observed competing toe-to-toe in Fire Strike at 1080p:

And the A4-4000 consumes roughly 12% less energy during this benchmark:

Hopefully the higher end Richland chips will make gains in 3D performance at the expense of efficiency. It would pain me to see AMD release another 5800K that only consumes less power. Also, source: http://translate.google.com/translate?sl=auto&tl=en&js=n&prev=_t&hl=en&ie=UTF-8&eotf=1&u=http://pc.watch.impress.co.jp/docs/topic/feature/20130513_599039.html

 

[8350rocks]

The high end Richland parts are supposed to be 20%+ faster over trinity...per AMD's press release announcement at launch.

 

[sarinaide]

Its getting closer now, though I don't really see myself getting a richland APU with Kaveri all but 4-5 months away.

 

[noob2222]

Spoiler

It is known that cinebench uses ICC

Spoiler

http://www.agner.org/optimize/blog/read.php?i=49
Tests with AMD bd 8150
MS 17.00 1 SSE2 code path: 724 clock cycles
Intel 12.1.3 1 generic code: 1950 clock cycles
Intel 12.1.3 1 Intel forced path: 720 clock cycles
obviously this test was to show near maximum benefit with code path and won't always be this obvious, but the truth is still evident.
with ICC and intel's checking system for GenuineIntel, amd gets the slowest code path available (yes at times slower than MSVC, but generally near identical). Benchmarking with this compiler is not representative to AMD's performance on an even playing field.
Cinebench may not be intentially "cheating", ICC is doing that for them

.

Maybe Cinebench people are using assembly code specially for AMD procs, and using ICC only because its the best compiler. Since CB is so CPU dependent, this probability is very very high.Maybe they manually removed the CPU check. Maybe they added other shims for AMD procs.
We only know that CB is compiled by ICC. We dont know what and how is the code written by CB developers.

Edit : I was just checking the link you gave. It mentions only some libraries that give separate code for Non-intel procs. Those libraries are high performance, and obviously tuned for the internal architecture of Intel procerssors.
Or do you think that the internal arch of AMD is the same as Intel, just because both support x86 ? Or does AMD hand out the details of its architectures to general public and competitors ? Or the marketing PR architecture slides we see at review time is sufficient to make a high performing, tuned library ?

that's an aweful lot of unknowns to try and defend intel, however, look at the timeline.

intel told to fix the compiler - dec 2009
cinebench 11.5 - feb 2010
intel offers somewhat better compiler with a footnote about not optimizing for AMD - june 2010.

so ... did intel tell maxon about the fraudulent compiler or just let them fly first knowing every review site would be using it?

 

[cowboy44mag]

Its getting closer now, though I don't really see myself getting a richland APU with Kaveri all but 4-5 months away.

^+1 I'm happy that Kaveri is going to be releasing soon, however from a purely business prospective AMD hurt themselves by releasing Richland this late/so close to Kaveri. Richland may be a nice upgrade from Trinity, however I think the vast majority of people are going to wait the additional few months for Kaveri. IMO Richland is going to go down as a failure (sales wise) while Kaveri will be a larger success not due to Richland being a poor APU, but because of poor timing from AMD.

 

[cowboy44mag]

Spoiler

It is known that cinebench uses ICC

Spoiler

http://www.agner.org/optimize/blog/read.php?i=49
Tests with AMD bd 8150
MS 17.00 1 SSE2 code path: 724 clock cycles
Intel 12.1.3 1 generic code: 1950 clock cycles
Intel 12.1.3 1 Intel forced path: 720 clock cycles
obviously this test was to show near maximum benefit with code path and won't always be this obvious, but the truth is still evident.
with ICC and intel's checking system for GenuineIntel, amd gets the slowest code path available (yes at times slower than MSVC, but generally near identical). Benchmarking with this compiler is not representative to AMD's performance on an even playing field.
Cinebench may not be intentially "cheating", ICC is doing that for them

.

Maybe Cinebench people are using assembly code specially for AMD procs, and using ICC only because its the best compiler. Since CB is so CPU dependent, this probability is very very high.Maybe they manually removed the CPU check. Maybe they added other shims for AMD procs.
We only know that CB is compiled by ICC. We dont know what and how is the code written by CB developers.

Edit : I was just checking the link you gave. It mentions only some libraries that give separate code for Non-intel procs. Those libraries are high performance, and obviously tuned for the internal architecture of Intel procerssors.
Or do you think that the internal arch of AMD is the same as Intel, just because both support x86 ? Or does AMD hand out the details of its architectures to general public and competitors ? Or the marketing PR architecture slides we see at review time is sufficient to make a high performing, tuned library ?

that's an aweful lot of unknowns to try and defend intel, however, look at the timeline.

intel told to fix the compiler - dec 2009
cinebench 11.5 - feb 2010
intel offers somewhat better compiler with a footnote about not optimizing for AMD - june 2010.

so ... did intel tell maxon about the fraudulent compiler or just let them fly first knowing every review site would be using it?

Intel is like one of those huge lobbyist groups that everyone knows about, most realize are pulling the strings, but nothing can be done to them because they have too much power. The lobby groups buy our elected politicians, buy their votes, and have laws they want passed and laws they don't want squashed. Intel does the same thing hoping people won't ask too many questions about how the benchmark tests are compiled. Just like the powerful lobby groups Intel walks away covered in manure, smelling like a rose, and there is nothing other companies can do about it. Money is power and Intel has more than enough to to skit the boundaries of "fair".

 

[sarinaide]

Its getting closer now, though I don't really see myself getting a richland APU with Kaveri all but 4-5 months away.

^+1 I'm happy that Kaveri is going to be releasing soon, however from a purely business prospective AMD hurt themselves by releasing Richland this late/so close to Kaveri. Richland may be a nice upgrade from Trinity, however I think the vast majority of people are going to wait the additional few months for Kaveri. IMO Richland is going to go down as a failure (sales wise) while Kaveri will be a larger success not due to Richland being a poor APU, but because of poor timing from AMD.

Richland may only appear on mass in Mobility form with Kaveri based notebooks out in Q2 2014, this gives AMD time to sell Richland notebooks albeit at the time DT Kaveri will be out in Q4-Q1. Since Richland is essentially a bump in clocks based on the same process as Trinity with minor metal level changes its basically AMD just binning excess silicon under a different name with tweaked clocks and cache latencies.

 

[8350rocks]

Well, the larger gains were in the upgraded GPUs on the die. Which makes Richland a formidable mobile solution.

 

[cowboy44mag]

Well, the larger gains were in the upgraded GPUs on the die. Which makes Richland a formidable mobile solution.

So Trinity to Richland is basically the same type of upgrade Intel made from Ivy Bridge to Haswell. Focusing primarily on GPU, while tweaking clock speeds correct? From the Haswell review I read there is only a 1-5% gain in benchmarks from Ivy Bridge, with the focus on lower power consumption and GPU which would only make sense for mobility devices. It seems like Intel figured AMD wouldn't be able to release Kaveri, or at least not on time so they focused Haswell on facing Richland not Kaveri.

 

[sarinaide]

Richland was only know later by Intel so its hard to say Haswell was even designed to take on Richland, Haswell's top end was designed to take on Trinity's top end, only time will tell how far they got and at what cost, what has been leaked is nothing useful.

 

[8350rocks]

Thought I would share this:

http://hexus.net/gaming/news/android/55317-nvidia-project-shield-launch-date-price-announced/

Nvidia's handheld console "Shield" will launch supposedly this year. It's android based and runs on Tegra4.

 

[jdwii]

Thought I would share this:

http://hexus.net/gaming/news/android/55317-nvidia-project-shield-launch-date-price-announced/

Nvidia's handheld console "Shield" will launch supposedly this year. It's android based and runs on Tegra4.

I still don't understand who would want to use that. Number 1 reason for me is the controller.

 

[8350rocks]

I don't like it either...

 

[8350rocks]

News from GloFo about the next processes:

http://www.electroiq.com/blogs/insights_from_leading_edge/2013/04/iftle-142-globalfoundries-2-5-3d-at-20nm-intel-haswell-gt3-umc-scp-prototype-details.html

http://www.pcper.com/reviews/Editorial/GLOBALFOUNDRIES-2013-and-Beyond

Both of those are good reads.

20nm TSV 3D stacking should be online by 2015, with preproduction samples out in 2H 2014.

Sounds like Excavator maybe ends up on that process...or the 14nm XM (What ever the HP variant is) Also discusses FD-SOI and gate last.

 

[juanrga]

Thought I would share this:

http://hexus.net/gaming/news/android/55317-nvidia-project-shield-launch-date-price-announced/

Nvidia's handheld console "Shield" will launch supposedly this year. It's android based and runs on Tegra4.

Yeah, that explains why Nvidia fiercely reaction when AMD and Sony announced the PS4. What Nvidia said then was they were not interested in consoles due to low margins and that they left AMD won the Sony contract... and now they are releasing a console. LOL

No that I did doubt of the real reason why Sony chose AMD (AMD has the technology Nvidia doesn't), but this gives more embarrassment to Nvidia reaction.

Nvidia Tegra 3 chips were already used in an Android-based OUYA console and it is generally considered a performance fiasco with 10-17 FPSs. Tegra 4 is much better, but don't wait miracles.

 

[griptwister]

Imagine what an APU or even a FX 8350 could do with this (talking about the cooler)

http://gigabytedaily.blogspot.tw/2013/05/ln2-ram-coolers-released-but-couple-of.html

And about that Nvidia console... Imo, it's gonna be a phenomenal hand held device, but it's not good enough to replace a PC. Even though you can hook it up some how to a 660 or higher GPU.

 

[Cazalan]

Thought I would share this:

http://hexus.net/gaming/news/android/55317-nvidia-project-shield-launch-date-price-announced/

Nvidia's handheld console "Shield" will launch supposedly this year. It's android based and runs on Tegra4.

Tegra 4 is a decent chip but no way that thing will sell at that price. It would have to be close to the Nintendo DS ($150) to get any traction at all. They're toys to keep your kids occupied in the car. Parents won't spend $349 for that.

 

[cowboy44mag]

Thought I would share this:

http://hexus.net/gaming/news/android/55317-nvidia-project-shield-launch-date-price-announced/

Nvidia's handheld console "Shield" will launch supposedly this year. It's android based and runs on Tegra4.

Tegra 4 is a decent chip but no way that thing will sell at that price. It would have to be close to the Nintendo DS ($150) to get any traction at all. They're toys to keep your kids occupied in the car. Parents won't spend $349 for that.

^+1 hand held console systems are fine when your trying to keep your kids amused or have some time to kill when your away from home. The minute you get home and have a choice of gaming systems your going for the "big boy toys". No hand held unit can match the power of even a console system, let alone a gaming PC. The games are worlds apart when it comes to graphics, physics and depth.

The hand held units are great for keeping your kids, or yourself, entertained while away from home but will never replace a real gaming system.

 

[palladin9479]

Thought I would share this:

http://hexus.net/gaming/news/android/55317-nvidia-project-shield-launch-date-price-announced/

Nvidia's handheld console "Shield" will launch supposedly this year. It's android based and runs on Tegra4.

Tegra 4 is a decent chip but no way that thing will sell at that price. It would have to be close to the Nintendo DS ($150) to get any traction at all. They're toys to keep your kids occupied in the car. Parents won't spend $349 for that.

^+1 hand held console systems are fine when your trying to keep your kids amused or have some time to kill when your away from home. The minute you get home and have a choice of gaming systems your going for the "big boy toys". No hand held unit can match the power of even a console system, let alone a gaming PC. The games are worlds apart when it comes to graphics, physics and depth.

The hand held units are great for keeping your kids, or yourself, entertained while away from home but will never replace a real gaming system.

I can't really agree with this, their designed for different things. I'm was absolutely in love with my DS XL and now with my 3DS XL (don't use the 3D part, it's sh!t). Handheld gaming is for exactly that, playing games while being able to move from point A to point B. My PS3/360/Wii setup at home is wired into my home theater, that's not exactly a portable concept. And honestly ... lately most games for those systems have been complete sh!t, just more military shooter clones. On the flip side I've immensely enjoyed playing Mario Card, Super Mario World, the Castlevania's, Zelda: Orcana of Time, and several "virtual console" titles. Graphically they may not be anywhere near the same, but game play wise their heads and shoulders over everything else. Really game producers these days have focused entirely too much on graphics instead of gameplay (Torthlight II vs Diablo III).

 

[sarinaide]

I have a HTC with Tegra cores, I really hope you don't have to travel long distances because that battery ain't gonna hold up. On just gaming alone the battery is good for an hour if that. Handhelds are good enough for boredom passing and travel occupation but they lack titles and the content console and PC/notebooks have, not to mention battery life.

 

[de5_Roy]

AMD’s Andrew Feldman Talks About ARM
http://semiaccurate.com/2013/05/15/amds-andrew-feldman-talks-about-arm/