• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

AMD Launches Three Kaveri APU SKUs in February 2014 – Feature Set For A10 and A8 APUs Detailed

What about 8 core non igpu parts on the FM2+ platform?

I've seen rumors of 8+ cores and bringing back the 'Phenom' monikor, (Just saying). We really have no idea at this point :p. AMD play their cards close to the chest these days.

FM2+ looks to be limited to <100w parts by design. High clocking AMD chips have, of late, been in the 200w+ region :D

Thats the Server road map, nothing to do with us here, you need to find the retail road map. :)

Server parts are huge margins. Desktop (and laptop) parts are derived from these. The fact that there is no sign of SR, an out and out server inspired design, says to me that it's been shelved. Probably due to R&D cost cutting. I reckon AMD are going to skip to 20/16nm Excavator on a new platform, late 2014/early 2015 with DDR4
 
Last edited:
Thats the Server road map, nothing to do with us here, you need to find the retail road map. :)

Piledriver for AM3+ is an offshoot of the server development; if they aren't making server parts from SR, they won't be making them for desktop.

What about 8 core non igpu parts on the FM2+ platform?

I've seen rumors of 8+ cores and bringing back the 'Phenom' monikor, (Just saying). We really have no idea at this point :p. AMD play their cards close to the chest these days.
It's possible, and I hope it's right, but it seems far fetched to me. It would mean developing and taping out a whole separate core for what is becoming a very niche market. AMD sells many more times the APUs than they do CPUs; why go to the effort of manufacturing a niche product when they can pile resources into something which actually sells?

All Trinity/Richland APUs (and I would assume the same will apply for Kaveri) are based on the same tapeout - they just adjust clocks and laser off sections for the lower end parts. Same with Piledriver on AM3+. Much cheaper development.
 
Thats not what i'm saying, just because you can't see the retail chips on the server road map does not mean they don't exist, its the wrong road map, its like looking for Sydney in Amsterdam, it not there.
 
I'm guessing an 8 core Kaveri chip is unlikely, even though the APU's used in the xbone and PS4 are 8 core APU's which use the older Jaguar cores.

An 8 core Kaveri could be a surprising performer, though I'm not sure if it would need L3 cache to help boost performance with 8 cores. Ah well, I guess that would require another socket so very unlikely to happen.
 
Thats not what i'm saying, just because you can't see the retail chips on the server road map does not mean they don't exist, its the wrong road map, its like looking for Sydney in Amsterdam, it not there.

Unless I'm mistaken.
He's saying because there's no server SR chips, there will be no Desktop chips, as lets face it, the desktop chips are just server chips.
 
Thats an assumption, i don't take assumptions as facts. facts are facts. ^^^ to get facts he needs to look at the correct road map.

I'm guessing an 8 core Kaveri chip is unlikely, even though the APU's used in the xbone and PS4 are 8 core APU's which use the older Jaguar cores.

An 8 core Kaveri could be a surprising performer, though I'm not sure if it would need L3 cache to help boost performance with 8 cores. Ah well, I guess that would require another socket so very unlikely to happen.

I don't want an 8 core Kaveri / Steamroller if it needs those 8 cores to keep up with Intel, what would be better is a CPU with less cores that can keep up with Intel, what would be really nice is a CPU with more cores that can beat Intel.
 
Last edited:
Thats not what i'm saying, just because you can't see the retail chips on the server road map does not mean they don't exist, its the wrong road map, its like looking for Sydney in Amsterdam, it not there.

d_ Brennan made the point; there's a lot of money in server chips - if Steamroller was being rolled out for 8-core desktop parts, you can be damned sure they would be used for server grade ones: hence they would be on the server map.

The FM2+ line of 8-cores is possible (excepting FM2+'s 100w limit, and whether the pinout would allow for it), but unlikely for the reasons I've mentioned. Plus, it just doesn't fit in with AMD's HSA push.
 
d_ Brennan made the point; there's a lot of money in server chips - if Steamroller was being rolled out for 8-core desktop parts, you can be damned sure they would be used for server grade ones: hence they would be on the server map.

The FM2+ line of 8-cores is possible (excepting FM2+'s 100w limit, and whether the pinout would allow for it), but unlikely for the reasons I've mentioned. Plus, it just doesn't fit in with AMD's HSA push.

d_ Brennan is amusing that because they don't show up on current server road maps it mean they don't exist in retail.

The sever road map will not show retail chips.

What we are aguing about here is d_ Brennan said they don't exist because he made an assumption to that effect based on the wrong info. This is exactly how Myth gets turned into fact. I'm saying wait for the correct info before drawing any conclusions.

Why is that so wrong?
 
Last edited:
With the change in design philosophy towards HSA, perhaps not this generation or the next, huge core counts are probably going to be obsolete. With Kaveri, you have 2 models with 4 int units and a massive parallel array of SIMDs just sitting there. In tasks where the GPU portion isn't fully utilised, the idea is that these can be put to work on parallel tasks seamlessly, with little developer effort. I hear that in future APU designs, the FPU might be completely decoupled from the CPU portion
 
d_ Brennan is amusing that because they don't show up on current server road maps it mean they don't exist in retail.

The sever road map will not show retail chips.

What we are aguing about here is d_ Brennan said they don't exist because he made an assumption to that effect based on the wrong info. This is exactly how Myth gets turned into fact. I'm saying wait for the correct info before drawing any conclusions.

Why is that so wrong?

I'm only making observations based on 20 years of interest in the field of processor tech, don't mind me :o

It's d_brennen. Why does everyone add an A where there isn't one? Assumption?
 
I reckon AMD are going to skip to 20/16nm Excavator on a new platform, late 2014/early 2015 with DDR4

Carrizo is the successor to Kaveri. Up to 65w parts so that's got to be 20nm and under.

Socket FM2+ still, but you'll need a board upgrade (from the current A88X boards) if you want to use DDR 4 with it.

These go into production in Q4 2014 for a H1 2015 release.
 
I'm a realist

By assuming that AMD are not going down the well trodden path of creating 3 families of CPU (server, desktop and mobile) in parallel, in order to hedge development costs, you are as much a fantasist as anyone here :D

SR is/was quite far down the line development wise, it's also based on PD with additions and tweaks, but for whatever reason AMD have chosen not to include it on their roadmaps for server or indeed desktop for some time. There's also the slip (typo?) of SteamrollerB cores being utilised in Kaveri. Why SR-B? What was wrong with SR-A? Also, Kaveri itself is ~ 4 quarters late to market. Richland was an unexpected revision of Trinity. So many signs point to something going **** up with SR. Anyone want to take a bet that GloFo couldn't deliver on their promises? AMD are tied into contracts with them til 2024 :o

Carrizo is the successor to Kaveri. Up to 65w parts so that's got to be 20nm and under.

Socket FM2+ still, but you'll need a board upgrade (from the current A88X boards) if you want to use DDR 4 with it.

These go into production in Q4 2014 for a H1 2015 release.

For some reason were talking "performance" CPUs here, rather than APUs :confused: (AM rather than FM socket)
 
Last edited:
I'm a realist, i try not to assume facts based on assumption, facts are facts and assumption is just that. i separate the two.

Whats wrong with that?

It's not assumption to take available evidence, weight up expertise and experience, and come to a conclusion - there are no assumptions other than to believe AMD's server roadmap is correct and complete.

The server roadmaps historically correspond very well to their consumer products - see the Kaveri-equivalent server parts, for example - and there is just no logical reason to believe AMD would bring 8-core SR to retail alone.
 
With the change in design philosophy towards HSA, perhaps not this generation or the next, huge core counts are probably going to be obsolete. With Kaveri, you have 2 models with 4 int units and a massive parallel array of SIMDs just sitting there. In tasks where the GPU portion isn't fully utilised, the idea is that these can be put to work on parallel tasks seamlessly, with little developer effort. I hear that in future APU designs, the FPU might be completely decoupled from the CPU portion

If this meant that the FPU functions could be carried out by the graphics side of the APU when you are using a separate dedicated graphics card then surely that could mean huge performance gains when it comes to gaming?
 
If this meant that the FPU functions could be carried out by the graphics side of the APU when you are using a separate dedicated graphics card then surely that could mean huge performance gains when it comes to gaming?

If it's stuff that can be parallelised then yes (e.g. Physx).
 
When I first sniffed an upgrade itch a few months back and trawled these very forums, these questions were remaining unanswered and after some reading I decided not to 'wait' for a SR 8 core as I would be waiting ages!

Of course I would love a SR iteration, but after mulling over xbone, PS4 or a 2015 PC I decided on biting the bullet for a FX83 to tide me over.

As mentioned earlier the server market does help making decisions and I too don't see anything materialising in 2014. A definite shame really as I thought it would be a great seller.
 
Back
Top Bottom