AMD on track for launch of Kaveri in February 2014

[Vesku]

Actually more code then ever can be shared between PCs and upcoming consoles now that both will be using x86. A lot of the porting issues will be gone or diminished and it is very likely the thread friendliness of PC games will increase. I see very few people on this forum disputing that even if they argue about exactly how much impact that will have on how games perform.

I have no idea how things will progress but AMD will be missing out on an opportunity if the console wins don't result in more widespread adoption of their HSA framework and thus better utilization of the iGP in their APUs.

 

[sniffin]

Actually more code then ever can be shared between PCs and upcoming consoles now that both will be using x86. A lot of the porting issues will be gone or diminished and it is very likely the thread friendliness of PC games will increase. I see very few people on this forum disputing that even if they argue about exactly how much impact that will have on how games perform.

I have no idea how things will progress but AMD will be missing out on an opportunity if the console wins don't result in more widespread adoption of their HSA framework and thus better utilization of the iGP in their APUs.

x86 or not consoles are different. They allow low level access to 8 Jaguar cores. Neither low level hardware access or 8 Jaguar cores are something found on PCs. Console code will always be highly specialized because it is a specialized platform. Specialized code does not translate to an open platform because said specialized hardware isn't actually present. In either case, there is nothing stopping big high frequency cores you'd find in a PC from doing the work of multiple small low frequency cores in a console.

People on this forum not disputing it doesn't actually mean anything. If you want highly technical discussion with actual developers there are better forums than this

 

[Vesku]

AFAIK, console coding doesn't happen much if at all in assembly language. Therefore with the core components having the same ISA there will be far more code reuse opportunities than when the consoles were using non-x86 architectures.

Then there is AMD providing the Mantle PC API which increases the opportunity for code reuse further.

Here are comments from an actual development studio:

http://www.vg247.com/2013/07/15/warframe-from-pc-to-ps4-in-three-months/

”I was every surprised how fast we got up and running – I think Sony was as well. When we went to [E3] people said, ‘Holy smokes this is on PlayStation 4? It looks amazing.’”

 

[NaroonGTX]

I think AMD will do just fine with their FM2+ desktop platform with Kaveri's CPU performance. The removal of the CMT thread scaling penalty will make the quad-cores perform as true quad-cores, i.e. full 4x scaling as opposed to 2.5x~3.0 scaling. The boosts in ST performance translate directly to boosts in MT performance.

Need I remind people how Intel's i5's and i7's (quad's with HT) would still pull past AMD octocores even in games that supported more than four threads? That's obviously because their individual core strength is higher. I would take four stronger AMD cores (Kaveri/SR) over 8 weaker ones that don't even truly scale 800% like they should any day.

When the benchmarks come out and people see the A10-7800k surpassing the FX-8350 in gaming benchmarks, they'll see the light after they pick their jaws up off the floor.

In regards to there possibly being FX variants with the SR x86 cores powering them, regardless of socket, this is dubious at this moment but the truth will be revealed in just a few days when the new roadmaps get released. The desktop roadmap will tell us what the channel will get in 2014 and possibly Q1 2015 -- in which case the future of Socket AM3+ will be spelled out.

Socket AM3+ will get one more chip (AMD did say this, and I don't think it was the FX-9000 series they were talking about) -- either this will simply be a Piledriver refresh which is trickled down from Warsaw with RCM enabled (see: Richland -- RCM was physically on both Trinity and Vishera, but wasn't enabled) or it will be a SR-powered part.

That's just my random (though realistic) speculation though. We'll find all of this out in a few days when the roadmaps come out. Whether or not AM3+ is truly dead or we don't see anything else for AM3+, at least we'd know for sure and people can feel free to go rioting in the streets and burning their MOBO's out of anger or what have you.

However I would like to say that it's not necessary for a hypothetical SR FX chip to be required to be a descendent of an Opteron part. They could make a die and release it on desktop first, but who knows...

 

[schmuckley]

I can tell you this:If they abandon AM3+,I will be boycotting AMD products forevermore.
I've been waiting for a better chip since 2009.I bought an m4a89gtd-pro to be ready for "Bulldozer"
IMO Thubans are best chip they've brought to market yet.

 

[NaroonGTX]

Why would you boycott them for that? They never once promised a Steamroller chip for AM3+ nor did they say AM3+ would receive more than Piledriver. AM3+ was meant for Bulldozer, i.e. Zambezi at the forefront. Piledriver was originally targeted at the Komodo platform: It was gonna be on Socket FM2 (yep, FM2 would receive FX *and* APU parts, hence an early move toward their goal of a unified socket) and the top-end part would've had 5 modules/10 threads max. Somewhere along the way this platform was canceled and they removed a module and sent it over to AM3+ instead. No idea why...

Also Thuban can't beat an 83xx processor. Vishera can reach higher clocks and has basically the same single-threaded perf -- sometimes slightly higher -- so Vishera wins by default.

 

[Vesku]

I'm somewhat with schmuckley on this one. Running a 1090T in my 990FX board since Bulldozer underwhelmingness was confirmed just before launch. Took them another year to edge past the Thuban @3.8GHz with Vishera. Would be a terrible cap to a mediocre run for AMD if they left AM3+ with their hail marry 220W Vishera CPUs. What better way to announce they are turning things around then by releasing some nice Steamroller based AM3+ chips?

 

[NTMBK]

What better way to announce they are turning things around then by releasing some nice Steamroller based AM3+ chips?

How about releasing some nice Steamroller chips on a platform that's not embarrassingly outdated?

 

[Vesku]

How about releasing some nice Steamroller chips on a platform that's not embarrassingly outdated?

Only thing AM3+ is missing is PCIe 3.0, I'd settle for a 3 or 4 module FM2+ chip but if they did that how much would it cost them to package some up for AM3+? Shouldn't be much since the 220W Visheras show they have the materials still on hand.

 

[NaroonGTX]

What better way to announce they are turning things around then by releasing some nice Steamroller based AM3+ chips?

A better way? Leaving AM3+, an old outdated chipset behind and bringing FX over to FM2+, a better chipset in just about every way?

Since HyperTransport doesn't exist on FM2 (or FM1) since the NB was moved from the MOBO to the CPU/APU die itself, it would make sense that if AMD were to move FX over to FM2+, it wouldn't be trickled down from an Opteron die. An FX-8350 is literally the same die as the Opteron part, just with 3 of the 4 HT links laser-cut off. What use would there be in having that cluttering up the die space if it wouldn't be used?

Hypothetical of course, but lends more credence to the theory of them leaving AM3+ behind for full-on FM2+.

Only thing AM3+ is missing is PCIe 3.0, I'd settle for a 3 or 4 module FM2+ chip but if they did that how much would it cost them to package some up for AM3+? Shouldn't be much since the 220W Visheras show they have the materials still on hand.

Well the dies would be totally different for reasons I just discussed up above. The 220W Visheras were the same dies as the normal 8350's, just better binned, i.e. guaranteed to reach 5Ghz or 4.7 or what have you. It's not like they had separate dies created for that or anything. They didn't even have RCM enabled. Same exact silicon design.

 

[ShintaiDK]

AM3+ platform is a problem, not a benefit. And so far it seems like there is no future for it. And thats how it should be, years ago.

 

[NaroonGTX]

I wouldn't say it's a problem, since they're still pushing processors for it (though not nearly as much as they're pushing APU's). It just wouldn't be a big deal if it were EOL'd as far as new releases went.

 

[NTMBK]

Only thing AM3+ is missing is PCIe 3.0, I'd settle for a 3 or 4 module FM2+ chip but if they did that how much would it cost them to package some up for AM3+? Shouldn't be much since the 220W Visheras show they have the materials still on hand.

It's a 2 chip platform, taking up more space on the motherboard and drawing more power than its rivals. It doesn't have PCIe integrated into the CPU, which reduces PCIe latencies and hence can improve graphics performance. FM2+ is a better platform.

And missing PCIe3 is a big deal, especially now that AMD's top graphics card needs excellent PCIe performance to deliver good Crossfire performance.

 

[NaroonGTX]

Weren't AMD the first to come out with PCIe 3.0 GPU's? They've had it for a while and they didn't get a platform that actually supported it until FM2+ (and currently it's not truly there until Kaveri hits).

 

[R0H1T]

Quad channel DDR4 = > 100GB/s.

60W cpu + 180W gpu = ideal ratio for gaming. Can easily be dissipated using a $20 cooling solution.

Total power of 240W. Nowhere near the limit of cooling.

There is no reason we cant have a 60W cpu and a 180W gpu and 4 channel DDR3 in an APU.

Correct me if I'm wrong here but there isn't gonna be a quad or even dual channel schema for DDR4 right? I was under the impression that point-to-point topology doesn't allow for such a thing, I could be wrong though

 

[guskline]

Time marches on. I own an Asus Sabertooth 990FX AM3+ mb with my FX8350 OC'd to 4600 (21 x 239) so I will be the first to say how I wish AMD would release a Steamroller cpu for it but as the posters above point out the FM2+ socket gives AMD many more advantages. In addition, you rarely hear an Intel supporter complain when they change sockets. My Z68 socket supported my 2500k Sandy and now my 3770k Ivy below but not the Haswell. My socket 2011 supports my Sandy-E 3930k as well as the new Ivy E but will not support the Haswell E next year.

I knew the AM3+ supported my 1100T, my FX8150 and now my FX8350. That's 3 generations!!!

Why beat up AMD if they say that's it? Time to move on.

 

[monstercameron]

Time marches on. I own an Asus Sabertooth 990FX AM3+ mb with my FX8350 OC'd to 4600 (21 x 239) so I will be the first to say how I wish AMD would release a Steamroller cpu for it but as the posters above point out the FM2+ socket gives AMD many more advantages. In addition, you rarely hear an Intel supporter complain when they change sockets. My Z68 socket supported my 2500k Sandy and now my 3770k Ivy below but not the Haswell. My socket 2011 supports my Sandy-E 3930k as well as the new Ivy E but will not support the Haswell E next year.

I knew the AM3+ supported my 1100T, my FX8150 and now my FX8350. That's 3 generations!!!

Why beat up AMD if they say that's it? Time to move on.

some people actually enjoy kickin someone while they are down, and has been very generous with their sockets, AM2 to AM2+ etc. the only time they seemed to mess up was with fm1.

 

[sm625]

Correct me if I'm wrong here but there isn't gonna be a quad or even dual channel schema for DDR4 right? I was under the impression that point-to-point topology doesn't allow for such a thing, I could be wrong though

That's probably true, but it is just semantics. What I meant was have a memory controller that is 128 or 256bit, to provide more bandwidth. They can design the controller any way they want, it doesnt have to be limited by the memory topology.

 

[R0H1T]

That's probably true, but it is just semantics. What I meant was have a memory controller that is 128 or 256bit, to provide more bandwidth. They can design the controller any way they want, it doesnt have to be limited by the memory topology.

My point was that Kaveri will probably not enjoy the amount of bandwidth you were suggesting, it'll be 128bit IIRC from an article/post on fud or S|A, so the early revisions might well be bandwidth constrained plus has it been established that Kaveri will support DDR4 this early ? I mean it may take AMD quite some time to make the transition from DDR3 to DDR4 unless they allow platform support for both of'em right from the get go.

 

[NTMBK]

My point was that Kaveri will probably not enjoy the amount of bandwidth you were suggesting, it'll be 128bit IIRC from an article/post on fud or S|A, so the early revisions might well be bandwidth constrained plus has it been established that Kaveri will support DDR4 this early ? I mean it may take AMD quite some time to make the transition from DDR3 to DDR4 unless they allow platform support for both of'em right from the get go.

Depends how high the DDR3 is clocked. A DDR3 7750 is bottlenecked, but that only has 128bt DDR3-1600 memory. A Kaveri with DDR3-2400 will have 50% additional bandwidth (though obviously some will be used by the CPU), and we've already got support for 2400 in Trinity.

 

[Erenhardt]

My point was that Kaveri will probably not enjoy the amount of bandwidth you were suggesting, it'll be 128bit IIRC from an article/post on fud or S|A, so the early revisions might well be bandwidth constrained plus has it been established that Kaveri will support DDR4 this early ? I mean it may take AMD quite some time to make the transition from DDR3 to DDR4 unless they allow platform support for both of'em right from the get go.

Phenom2 supported DDR2 and DDR3. The same may be true for kaveri (DDR3 and DDR4 support ofc )

 

[Vesku]

Time marches on. I own an Asus Sabertooth 990FX AM3+ mb with my FX8350 OC'd to 4600 (21 x 239) so I will be the first to say how I wish AMD would release a Steamroller cpu for it but as the posters above point out the FM2+ socket gives AMD many more advantages. In addition, you rarely hear an Intel supporter complain when they change sockets. My Z68 socket supported my 2500k Sandy and now my 3770k Ivy below but not the Haswell. My socket 2011 supports my Sandy-E 3930k as well as the new Ivy E but will not support the Haswell E next year.

I knew the AM3+ supported my 1100T, my FX8150 and now my FX8350. That's 3 generations!!!

Why beat up AMD if they say that's it? Time to move on.

Because it was more like 1.25 generations much like Sandybridge to Ivybridge but with much lower base performance. AMD never offered the jump in MT performance they were trumpeting at the time and if they don't offer 3 or 4 module Steamroller they still will not be offering such. If it's on FM2+ without even some legacy packaged AM3+ SKUs so be it but I'm doubtful they will offer anything compelling to the high end.

 

[R0H1T]

Phenom2 supported DDR2 and DDR3. The same may be true for kaveri (DDR3 and DDR4 support ofc )

Actually I'm hoping for that prediction to come true because DDR4 at launch will be uber expensive & AMD will probably support a revision(or two) of Kaveri till a socket change may be forced upon them due to the impending introduction of SATA express &/or PCIe 4.0 for desktops

 

[NTMBK]

Phenom2 supported DDR2 and DDR3. The same may be true for kaveri (DDR3 and DDR4 support ofc )

I'm hoping for DDR3 and GDDR5, still- I mean, plenty of their GPUs already work with both (hi there 7750!), and the two are a lot closer than DDR3 and 4. Using GDDR5 in some situations (like soldered down boards for Steamboxes) would be nice.

 

[Insert_Nickname]

Depends how high the DDR3 is clocked. A DDR3 7750 is bottlenecked, but that only has 128bt DDR3-1600 memory. A Kaveri with DDR3-2400 will have 50% additional bandwidth (though obviously some will be used by the CPU), and we've already got support for 2400 in Trinity.

No we don't. Maximum official for Trinity is 2133MHz for the 6800K only, everything else is 1866MHz (not counting OC obviously). I might add that my own 6800K has serious trouble with 2400MHz, it requires EXTREMELY conservative timings (15-15-15-37 CR2) to even post. Weather that is just mine or its general for 6800K's, I don't know.