Richland, Kaveri, Kabini & Temash; AMD’s 2013 APU Lineup Examined

[Martini1991]

Any device I'd get would solely rely on the IGP, I'd much prefer a 7660G than the cutdown rather downclocked hd4000 I'm currently using.

 

[d_brennen]

Check out youtube for A10 4600m or 7660G gameplay vids, theres a few (some by me ), including a guy called LoneSyndal. Not too shabby for an iGPU

 

[CAT-THE-FIFTH]

According to AMD,Richland is Piledriver with process tweaks and improved power management between the CPU and IGP.

 

[CAT-THE-FIFTH]

Great article:

http://www.pcper.com/reviews/Shows-and-Expos/AMD-CES-2013-Temash-Kabini-and-Kaveri-side-Sea-Islands

Kabini has a 20% IPC improvement over the previous CPUs.

So it seems the IGP in Richland has a new arrangement which is more efficient and the core has a slight IPC improvement too. It appears Trinity supports PCI-E 3.0 but this was not activated in the chips, and the reason why the CPUs were moved to FM2 was for future PCI-E support on the platform.

Kaveri is being introduced to desktop first and AMD is trying to release as soon as possible and hopefully before the end of the year.

 

[Martini1991]

It appears Trinity supports PCI-E 3.0 but this was not activated in the chips, and the reason why the CPUs were moved to FM2 was for future PCI-E support on the platform.

Isn't that down to chipset? I don't know if Trinity has the PCI-E controller in the CPU, but AM3+ Chips don't.

And that would be illogical to put make Trinity with PCI-E 3.0 as opposed your high end platform.

 

[CAT-THE-FIFTH]

Isn't that down to chipset? I don't know if Trinity has the PCI-E controller in the CPU, but AM3+ Chips don't.

And that would be illogical to put make Trinity with PCI-E 3.0 as opposed your high end platform.

Trinity has the PCI-E controller in the CPU. I wonder if AMD will also move its non-APUs to FM2 pin arrangement also,as the article also mentions FM2+ too.

Edit!!

AM3 has 941 pins and FM2 has 904 pins,so I expect it might be doable.

 

[Martini1991]

Trinity has the PCI-E controller in the CPU. I wonder if AMD will also move its non-APUs to FM2 pin arrangement also,as the article also mentions FM2+ too.

Edit!!

AM3 has 941 pins and FM2 has 904 pins,so I expect it might be doable.

AMD are illogical.
"Herp derp, lets give the platform that focuses on IGP, PCI-E 3.0" Assuming of course that it's not misinformation, as it doesn't make any sense to disable it, or include a PCI-E 2.0 controller (I'm assuming it's taking up die space?)
It says FM2 supports PCI-E 3.0, as opposed to saying Trinity does (Which makes more sense) by that it could mean they've got the switches in that MSI made such a fuss about and the SR based APU's will bring in PCI-E 3.0? But I'm speculating.

I'm sure I'd read that FM2 CPU's had the PCI-E controller before as well, /facepalm to me

If FM2 replaced AM3+ then I could feel that being a MASSIVE kick in the knackers for AM3+ owners.

EDIT : Although Kabini/Temash looks VERY nice for a Windows tablet, I loved my C-50 with 6250, but this could be a lot better, might have to sell my Vivobook, good thing I got an SSD for it, it'll maintain its price.

 

[CAT-THE-FIFTH]

FM2 and FM2+ will be the primary platform for AMD for quite some time. It was essentially confirmed that Trinity has the ability to run PCI-E 3.0 speeds, but the parts were never certified due to a mix of time, budget, and personnel constraints.

That is from the article.

Considering that AMD bought Seamicro who do high density microservers,it makes me think whether at some point you will see modified versions of their mainstream APUs in servers,on a new platform. There are already Opteron versions for workstations. PCI-E 3.0 is more important there for the compute cards and the CPUs are essentially used to drive the graphics cards, which do a decent amount of the FP operations. AFAIK,Nvidia is trying to do make server CPUs too which use ARM cores,and these can be used to drive their GPUs instead of an AMD or Intel one.

The problem is that after the SB850 in 2010,AMD has essentially not really bothered really updating their AM socket arrangement. All emphasis has gone onto their low power CPUs and the FM socket arrangement.

OTH,AMD did make some noise about replacing HT with technology acquired from Seamicro:

http://www.eetimes.com/electronics-news/4394786/AMD-s-Jaguar-packs-four-cores-in-one-for-mobile

The name sounds too much like Freedom Fries though!!

So,perhaps they are probably looking at a big overhaul maybe in 2014 or 2015 which unifies the platform technology including server?? AFAIK,HT is starting to become a limitation in certain cases.

 

[humbug]

Great article:

http://www.pcper.com/reviews/Shows-and-Expos/AMD-CES-2013-Temash-Kabini-and-Kaveri-side-Sea-Islands

Kabini has a 20% IPC improvement over the previous CPUs.

So it seems the IGP in Richland has a new arrangement which is more efficient and the core has a slight IPC improvement too. It appears Trinity supports PCI-E 3.0 but this was not activated in the chips, and the reason why the CPUs were moved to FM2 was for future PCI-E support on the platform.

Kaveri is being introduced to desktop first and AMD is trying to release as soon as possible and hopefully before the end of the year.

That would bring performance on par with Sandy Bridge, which would be nice.

But Matini is right, i would not be happy having bought a top end AM3+ little more than 6 months ago. Which would also but worthless if AMD switch sockets.

 

['Wes]

amd seems to be most of the time a dirt word here, and no i wouldn't write them off either

Not at all.

 

[CAT-THE-FIFTH]

That would bring performance on par with Sandy Bridge, which would be nice.

But Matini is right, i would not be happy having bought a top end AM3+ little more than 6 months ago. Which would also but worthless if AMD switch sockets.

Kabini is the replacement for Brazos. So it has a 20% IPC improvement and twice the cores.

 

[humbug]

Kabini is the replacement for Brazos. So it has a 20% IPC improvement and twice the cores.

Wait a minute, i'm getting confused with AMD's vast naming array....

Is Brazos the low power Mobile chip? if so is that also based on Bobcat, which if i remember rightly had (at the time) a higher IPC than Intel's equivalent.

And Kabini is the new Desktop APU? No wait thats Kaveri... my head hurts.

 

[Martini1991]

Brazos is the low power stuff, the C-50 etc, they at the time had better core for core performance than Intels stuff.
But as far as I'm aware now, AMD no longer have the higher core for core performance (Released in products) and consume a fair bit more power.

Although, don't think 20% IPC would quite match a sandy constantly core for core (Although, if the FX8350 had 20% better IPC than it did now, it'd be pretty stellar, but not for every situation, but it'd be worth the compromise)

 

[d_brennen]

Bobcat was the codename for the cores. The platform was called Brazos
Jaguar is the follow up core and should be a good deal faster, that platform is called Kabini

Piledriver is a tweaked Bulldozer core, as found in Trinity APUs and the new FX series, Temash for ultra low power/tablets and the like

Could not be more contrived but CPU companies love code names

Edit: Part of the reason for naming the core and the package is AMD's modularity thing. Since Bobcat and now Jaguar, the core can be relatively easily transferred to a different design or even fabbed in a different fab. We haven't really seen anything use this design trait (aside from a few embedded widgets no one hears of) but you will see Jaguar cores elsewhere

 

[humbug]

I'm trying to work out whats what on the Desktop.

Piledriver as far as i can tell is the last of this unsuccessful Bulldozer architecture.

There are two Desktop APU's coming this year, one at 32nm and the next at 28nm, the later is the one thats more interesting. i'm assuming that is Steamrooler (that the last of Bulldozer)
Or is it something else, something scaled up from the Mobile Chips, Bobcat cores perhaps.

Excavator *silly name cringe* is dead, AMD have already said they have killed that off and are working on a new architecture, i'm wondering if that, whatever it is, fits in here somehow.

 

[d_brennen]

Kabini (Jaguar) will be "leveled up" to A4/A6, while Richland will occupy A8 and A10. That alone speaks volumes of what AMD think of the little Jaguar imo. Richland 32mn, Jaguar 28nm. Richland has Piledriver cores, like Trinity but tweaked. Steamroller is a way off, which is why I said we have yet to see the proper follow up to Trinity - end of this year I think they will be showing them off, available early 2014. Steamroller will be 28nm barring major **** ups and will come in both A8/A10 (maybe just A10, maybe a new number - A12?) and FX flavours. AMD is said to be dropping non APUs, so the FX line may not happen, releasing as Opteron for server/workstations in perhaps 8/10/12/16 core variants

Take that with a pinch of salt, that's just the way I've read it

 

[humbug]

Kabini (Jaguar) will be "leveled up" to A4/A6, while Richland will occupy A8 and A10. That alone speaks volumes of what AMD think of the little Jaguar imo. Richland 32mn, Jaguar 28nm. Richland has Piledriver cores, like Trinity but tweaked. Steamroller is a way off, which is why I said we have yet to see the proper follow up to Trinity - end of this year I think they will be showing them off, available early 2014. Steamroller will be 28nm barring major **** ups and will come in both A8/A10 (maybe just A10, maybe a new number - A12?) and FX flavours. AMD is said to be dropping non APUs, so the FX line may not happen, releasing as Opteron for server/workstations in perhaps 8/10/12/16 core variants

Take that with a pinch of salt, that's just the way I've read it

Ok, *pinch of salt in hand* it makes sense, i don't care if they kill off FX CPU's, they killed the good AMD FX name with Bulldozer.....

I don't care about 8 fake cores, tho 6 real ones would be nice... and i don't actually care about what platform its on, iGPU with them or not.

All i want is for the CPU cores to be competitive with Intel, AMD's biggest mistake in the last few years has been Bulldozer and then sticking with it failing to see its faults.

Its make or break time for AMD.

 

[drunkenmaster]

Bulldozer is anything but a bad architecture, you either go more cores and less IPC first, then add the IPC, or more IPC less cores, then add the cores. Neither way is right or wrong, Bulldozer had major shortcomings because fitting the cores in meant things like shared decoder and other faults you get on EVERY first generation of an architecture. Ath 64 was fast yet still way slower than phenom 2 which was based heavily on it, Yonah vs anything after it, architectures start off basically crap and improve as you fix the biggest bottleneck each (real) generation.

AMD is supposedly canning projects then bringing them back, based on who they've brought in recently, Steamroller supposedly came back from the dead but I'm not sure thats true.

The likelyhood of any consoles being based on Kabini is, while good chips, low because they aren't designed for high clock speeds, great IPC they actually do have, but at 2Ghz or so that is going to significantly limit gaming performance. While we have multithreaded games there are still bigger and smaller threads and we've seen that games can suffer badly if a thread is limited to lower speeds.

That means in almost every likelyhood any consoles using an AMD cpu will be based off Kaveri/steamroller, that means the chip had to be almost done and would take effectively very little to put out for PC's, and because its on consoles they will move Steamroller to 20nm and likely as far as 16-14nm at least as console makers want to make more profit on the hardware over time. So even at worst we'd have a 20nm vs of Steamroller.

Because of the IPC improvements in it and because its 8 core while Intel aren't planning 8 cores in mainstream for some time, Steamroller would well last a lot longer than usual in the mainstream segment, particularly with price drops moving it to 20nm.




As for Kabini taking the lower end APU names, I think it says partially how fast it is, but more that AMD is looking to consolidate a range of products under one brand rather than different brands, which makes a lot of sense.

But for AMD selling circa 100mm^2 in laptops worth what, £250-500 rather than 200mm^2 Richlands in the same laptops will mean more profit for AMD, more volume(twice the chips per wafer) and that it allows more leeway to price the laptops lower to make them more competitive. Ultimately most people looking for low end laptops want enough performance rather than as much as possible.

With PC's being where they are at and AMD's relatively weak showing in laptops and problems with profitability its a key segment for them to target really.

Bobcat was supposed to have roughly 85% the IPC of Bulldozer, and its got a 20% bump, but its relative performance should increase beyond that due to what seems to be just massively higher efficiency all around.

I think that is also the argument being said for Richland vs seemingly slightly lower in the range Trinitys, that efficiency wise they are pushing faster chips into the lower wattage. Power really is the key factor, every watt you save in idle gives you several more watts to spend under load while keeping the device viable for use. Again the argument of say a 25W chip that uses 5W at idle having the same battery life as a 35W Richland that uses 2W at idle, and the 35W chip is way faster than a 25W one under load.(the numbers won't be that dramatically different but the point stands).

There is a reason AMD and Intel are doing the huge majority of their work/focus on power saving/efficiency, because with it gives you the ability to spend more power under performance. The more efficiency you gain the more performance you can spend that saved power on. One of ARM's biggest advantages was AMD/Intel's lack of focus on power saving for, well, donkeys years. Its only taken a couple years for Intel to turn Atom from an utter turd into a semi viable device vs ARM, that is a monumental turn around, and AMD had a better chip than Atom and have seemingly done many similar power saving things.

AMD's weakness is cash, less engineers they can afford, they can't afford to make a great chip in every segment, they do seem to be pinning their hopes on the 10-25W range rather than spend shedloads of money trying to make <3W chips, though Temash isn't a bad start.

 

[d_brennen]

Holy wall of text. BD was wrong for the desktop from the get go. AMD bet on parallelism, which hasn't materialised, yet...

 

[humbug]

Bulldozer is anything but a bad architecture, you either go more cores and less IPC first, then add the IPC, or more IPC less cores, then add the cores. Neither way is right or wrong, Bulldozer had major shortcomings because fitting the cores in meant things like shared decoder and other faults you get on EVERY first generation of an architecture. Ath 64 was fast yet still way slower than phenom 2 which was based heavily on it, Yonah vs anything after it, architectures start off basically crap and improve as you fix the biggest bottleneck each (real) generation.

AMD is supposedly canning projects then bringing them back, based on who they've brought in recently, Steamroller supposedly came back from the dead but I'm not sure thats true.

The likelyhood of any consoles being based on Kabini is, while good chips, low because they aren't designed for high clock speeds, great IPC they actually do have, but at 2Ghz or so that is going to significantly limit gaming performance. While we have multithreaded games there are still bigger and smaller threads and we've seen that games can suffer badly if a thread is limited to lower speeds.

That means in almost every likelyhood any consoles using an AMD cpu will be based off Kaveri/steamroller, that means the chip had to be almost done and would take effectively very little to put out for PC's, and because its on consoles they will move Steamroller to 20nm and likely as far as 16-14nm at least as console makers want to make more profit on the hardware over time. So even at worst we'd have a 20nm vs of Steamroller.

Because of the IPC improvements in it and because its 8 core while Intel aren't planning 8 cores in mainstream for some time, Steamroller would well last a lot longer than usual in the mainstream segment, particularly with price drops moving it to 20nm.




As for Kabini taking the lower end APU names, I think it says partially how fast it is, but more that AMD is looking to consolidate a range of products under one brand rather than different brands, which makes a lot of sense.

But for AMD selling circa 100mm^2 in laptops worth what, £250-500 rather than 200mm^2 Richlands in the same laptops will mean more profit for AMD, more volume(twice the chips per wafer) and that it allows more leeway to price the laptops lower to make them more competitive. Ultimately most people looking for low end laptops want enough performance rather than as much as possible.

With PC's being where they are at and AMD's relatively weak showing in laptops and problems with profitability its a key segment for them to target really.

Bobcat was supposed to have roughly 85% the IPC of Bulldozer, and its got a 20% bump, but its relative performance should increase beyond that due to what seems to be just massively higher efficiency all around.

I think that is also the argument being said for Richland vs seemingly slightly lower in the range Trinitys, that efficiency wise they are pushing faster chips into the lower wattage. Power really is the key factor, every watt you save in idle gives you several more watts to spend under load while keeping the device viable for use. Again the argument of say a 25W chip that uses 5W at idle having the same battery life as a 35W Richland that uses 2W at idle, and the 35W chip is way faster than a 25W one under load.(the numbers won't be that dramatically different but the point stands).

There is a reason AMD and Intel are doing the huge majority of their work/focus on power saving/efficiency, because with it gives you the ability to spend more power under performance. The more efficiency you gain the more performance you can spend that saved power on. One of ARM's biggest advantages was AMD/Intel's lack of focus on power saving for, well, donkeys years. Its only taken a couple years for Intel to turn Atom from an utter turd into a semi viable device vs ARM, that is a monumental turn around, and AMD had a better chip than Atom and have seemingly done many similar power saving things.

AMD's weakness is cash, less engineers they can afford, they can't afford to make a great chip in every segment, they do seem to be pinning their hopes on the 10-25W range rather than spend shedloads of money trying to make <3W chips, though Temash isn't a bad start.

It seems completely insane to me that a small core low power mobile chip like Bobcat can have 85% of Bulldozers IPC, i don't know what to make of that... (thats how good Bobcat is, or, thats how bad Bulldozer is) probably a bit of both.

Yes, i can see how Bulldozer may have made sense to AMD, pile on the cores and then bring up the IPC on those cores, end result = Multithreading monster!
Yet that end result does not seem achievable, it sucks up far to much juice and it seems that IPC can't be persuaded out of it, AMD may get another 10% IPC, get the power consumption down a bit more, up the clocks a little more... and then its that much better, but i think they now know the architecture has just about found its limits.
Perhaps a real shame then that it did not work as planned.