Kaveri speculation thread.

interesting that there might be as many as 768 GCN 2.0 shaders, that would be an interesting chip.

personally, i think it is speculation and 512 is more likely.
 
I want 7750 level graphics on-chip with low-level i5 CPU performance.

I will probably be dissiapointed...
 
i have long thought they should have opted for a triple channel memory system in FM2, and then used the same socket for high-end chips.

i too will be disappointed.

triple channel DDR3 2133 would provide 64GB/sec, enough to properly fill a high-end 100W APU with 512 shaders.
 
KazeoHin said:
I want 7750 level graphics on-chip with low-level i5 CPU performance.

I will probably be dissiapointed...
Click to expand...

Well it is a 2 module part , unless the application is really great at Open-CL or Compute the performance will not be there.
 
Low level i5 performance will be a stretch, but the 7750-level graphics is optimistic, but it could happen... Right?
 
I think the 512 GCN cores are possible just not highly clocked, well not until the kaveri refresh.
 
I'm assuming that means the standard desktop parts (not APUs) will also be getting pushed back - dang :/
 
I have high hopes for this chip, if AMD gets HUMA correct on the first try and achieves the rumored 30% IPC increase this will be quite nice. Though I suspect I will continue riding AMD's hopeful/dissapointed rollercoaster...
 
C'mon, Kaveri! Be the APU that fixes everything.

I want to see:
- FX-8xxx level performance
- Outstanding IGP performance
- Power draw under load in the same arena as a SB or better
- Way less heat output
 
Pieter3dnow said:
according to http://www.digitimes.com/news/a20130716PD202.html

Kaveri won't be making retail this year.
Click to expand...

This is nonsense. Lisa Su already said at Computex 2013 not long ago that Kaveri was sent to various OEM's and was already sampled, and that Kaveri would be shipping later in 2013. https://www.youtube.com/watch?v=ueHW0bJuZPQ Go to the 26:00 mark. Not to mention that the article/source's claims of GloFo not having enough "spare capacity" is complete bullshit because GloFo in fact has a lot of spare capacity.

This rumor is just as baseless and nonsensical as that rumor from late 2012 that said that Steamroller and Excavator were being canceled. This is just as implausible and ridiculous as those rumors a year ago.

Move along, people.
 
Last edited:
Well we just have to wait. according to http://www.rage3d.com/articles/hardware/amd_worldcast/

from the 29th of April
Chekib Akrout, Senior Vice President & Business Transformation Officer

AMD's first steps are to shrink into a self-sustainable footprint, pursuing safe, viable, ongoing revenue sources and then expand into new and innovative markets. Currently that means consumer gaming, dense cloud servers, GPGPU compute accelerators, and embedded platforms. Of these, three are driven by the ongoing development of Heterogeneous System Architecture, which is building momentum and reaching a final stable first state this year with the Kaveri design, launching early 2H 2013.
Click to expand...

The problem is that things been said and reality sometimes differ. But I'm still hoping they can roll out the FM2+ platform with Kaveri this year.
 
Richland was announced before that was posted, and it was launched late 1H 2013, basically. The latest revision of AMD's desktop roadmaps shows that Kaveri is listed around the late 2H 2013 area, so that article there isn't much. For now, with that roadmap and with Lisa Su stating that Kaveri was already sampled to select partners, Kaveri seems to be on track just fine.

I'm not gonna believe any delay rumors unless AMD themselves says anything.
 
A lot of underestimation in this thread. AMD got your expectations that far down people? A-10 7800K (guessing on model number) Kaveri should give the performance of an FX 83xx.
 
I wouldn't be surprised if Kaveri, even with just two modules, would surpass the 8350. Overall IPC enhanced, the full-module load penalty being gone, the SR cores will be better in just about every way. Then the GPU will blow Trinity/Richland away as well.

More evidence that the rumors of Kaveri being pushed back are bull is how FM2+ MOBO's have been produced and either already are, or will be on retail very soon. No reason why AMD would give the go-ahead to release MOBO's at this time if the product wouldn't be launching until next year.
 
Any idea how long the FM2+ platform will stick around? I plan on building a new rig around winter time but I'd like to know if the mobo I get will last for a few cpu upgrades
 
TMCM said:
Any idea how long the FM2+ platform will stick around? I plan on building a new rig around winter time but I'd like to know if the mobo I get will last for a few cpu upgrades
Click to expand...

Going by what AMD has been doing with their FM socket thus far, I would not expect any kinda of generational upgrade like what happened with the AM socket. Expect at most a tick tock type of upgrade, like going from Trinity -> Richland. Though it probably would not be worthwhile upgrade unfortunately.

To be honest AMD is best off not going this route. Using the same socket over and over only holds back the potential progress of the chipset and the APU. Such as DDR4 being released in the coming year or two. That is going to bring a lot more bandwidth, which AMD needs to fuel their GPU on the APU. That further adds to seeing the FM2+ socket dying even quicker.
 
TMCM said:
Any idea how long the FM2+ platform will stick around? I plan on building a new rig around winter time but I'd like to know if the mobo I get will last for a few cpu upgrades
Click to expand...

It will most likely last through the Kaveri refresh/shrink but other than that, no one really knows. At least through 2014 and likely some of the way through 2015.

NaroonGTX said:
Richland was announced before that was posted, and it was launched late 1H 2013, basically. The latest revision of AMD's desktop roadmaps shows that Kaveri is listed around the late 2H 2013 area, so that article there isn't much. For now, with that roadmap and with Lisa Su stating that Kaveri was already sampled to select partners, Kaveri seems to be on track just fine.

I'm not gonna believe any delay rumors unless AMD themselves says anything.
Click to expand...

Correct, desktop is still on track to release this year. Mobile partners will get production silicon in a month or two, which is probably what is being misconstrued with all these new articles about Kaveri delays, so that they can have their designs ready to ship late Q1/early Q2.
 
Probably won't see a new socket until Excavator. What little info I could find on that when I did research a few months ago, were supposed quips from AMD where they expressed desire to have Excavator on a unified socket (meaning APU's and FX would fit into it). It would kind of make sense, especially since even their GPU's are gonna get HSA features later. Who knows though, like all things we'll just have to wait and see.

Kaveri will last a decent amount of time: This year (what little is left of it, lol), next year and later with the refresh, then later there would be the 4th gen and last Bulldozer-based APU (for now, no idea what AMD has planned beyond Bulldozer ver1-4). Who knows if Excavator would even require a new socket... Actually, yeah, would probably be FM3 with DDR4 support.

But all I know is that I'm going Kaveri, and I wouldn't mind getting a new MOBO again for Excavator's Carizo later on since it'll bring even more improvements to HSA.
 
Does the FM2+ sockets require a different heatsink than AM3 sockets? I have the Macho HR02 heatsink that is freaking awesome and I would like to use that when I upgrade
 
If the last few years in the AMD forums are any indication of a pattern it will be something like this:
  1. Hyped.
  2. Late.
  3. Hot.
  4. Almost but not quite as big an improvement as the (revised) claims AMD said it would be.
  5. Far less of an improvement as the AMD fanboys swore it would be.
  6. Be disappointing (at least slightly to moderately depending on what hype you believed).
  7. Anyone who points out the disappointment will be labeled an Intel fanboy.
  8. Compared to $100 more expensive Intel chips (in "best AMD vs. best Intel" reviews) and be found lacking.
  9. Said review sites will be labeled Intel fanboys.
  10. That will include [H].
  11. Declared the performance/$ champion.
  12. Anyone who points out that the $ difference isn't that high especially when you consider that many of these chips find their way into $2000-3000 systems to where the extra cost becomes less than 5% of the overall system cost will be labeled Intel fanboys.
  13. Be declared to have much higher OC headroom when someone uses dilithium crystals mined from the rings of Saturn and chilled to 5 degrees below absolute zero manages to do a CPU-Z validation with the clock speed 5MHz higher than previous record attempts for 5 seconds before the dilithium crystals destabilizes and irradiates a large area of the state or country said record was attained in.
  14. Anyone who points out that it's not 24/7 stable or remotely reasonable to maintain and that real world OC results aren't all that different from Intel chips, and that the Intel chips start out at a lower clock speed to begin with so that for day-to-day use that the Intel chips actually OC better will be labeled an Intel fanboy.
  15. That will include me.
  16. Won't OC as well as expected in general use to begin with.
  17. Will be re-released in a year as a 20% faster clocked part with no other noticeable differences for 3x the cost and about 90% less OC headroom and way way more power usage.
  18. Nobody will put together that 17 might have something to do with the 16.
  19. The legacy of Hector Ruinz will be blamed by many for the majority of these failures/shortcomings
  20. That will include me.
That about covers my predictions for the next few AMD releases, at least until AMD does its own NetBurst -> Core transition
 
Kaveri "Athlons" will be as fast as FX-8xx0 series, while being slightly slower than the FX-9xx0 series. Overall, FM2+ replaces AM3+ as a platform.
 
-Dragon- said:
If the last few years in the AMD forums are any indication of a pattern it will be something like this:
  1. Hyped.
  2. Late.
  3. Hot.
  4. Almost but not quite as big an improvement as the (revised) claims AMD said it would be.
  5. Far less of an improvement as the AMD fanboys swore it would be.
  6. Be disappointing (at least slightly to moderately depending on what hype you believed).
  7. Anyone who points out the disappointment will be labeled an Intel fanboy.
  8. Compared to $100 more expensive Intel chips (in "best AMD vs. best Intel" reviews) and be found lacking.
  9. Said review sites will be labeled Intel fanboys.
  10. That will include [H].
  11. Declared the performance/$ champion.
  12. Anyone who points out that the $ difference isn't that high especially when you consider that many of these chips find their way into $2000-3000 systems to where the extra cost becomes less than 5% of the overall system cost will be labeled Intel fanboys.
  13. Be declared to have much higher OC headroom when someone uses dilithium crystals mined from the rings of Saturn and chilled to 5 degrees below absolute zero manages to do a CPU-Z validation with the clock speed 5MHz higher than previous record attempts for 5 seconds before the dilithium crystals destabilizes and irradiates a large area of the state or country said record was attained in.
  14. Anyone who points out that it's not 24/7 stable or remotely reasonable to maintain and that real world OC results aren't all that different from Intel chips, and that the Intel chips start out at a lower clock speed to begin with so that for day-to-day use that the Intel chips actually OC better will be labeled an Intel fanboy.
  15. That will include me.
  16. Won't OC as well as expected in general use to begin with.
  17. Will be re-released in a year as a 20% faster clocked part with no other noticeable differences for 3x the cost and about 90% less OC headroom and way way more power usage.
  18. Nobody will put together that 17 might have something to do with the 16.
  19. The legacy of Hector Ruinz will be blamed by many for the majority of these failures/shortcomings
  20. That will include me.
That about covers my predictions for the next few AMD releases, at least until AMD does its own NetBurst -> Core transition
Click to expand...

And yet, that won't be Kaveri...
 
NaroonGTX said:
Probably won't see a new socket until Excavator. What little info I could find on that when I did research a few months ago, were supposed quips from AMD where they expressed desire to have Excavator on a unified socket (meaning APU's and FX would fit into it). It would kind of make sense, especially since even their GPU's are gonna get HSA features later. Who knows though, like all things we'll just have to wait and see.

Kaveri will last a decent amount of time: This year (what little is left of it, lol), next year and later with the refresh, then later there would be the 4th gen and last Bulldozer-based APU (for now, no idea what AMD has planned beyond Bulldozer ver1-4). Who knows if Excavator would even require a new socket... Actually, yeah, would probably be FM3 with DDR4 support.

But all I know is that I'm going Kaveri, and I wouldn't mind getting a new MOBO again for Excavator's Carizo later on since it'll bring even more improvements to HSA.
Click to expand...

That actually seems more likely.

Excavator should be the final step in AMD's HSA convergence of the GPU and CPU where we see full implementation of it. Kaveri is the shared memory space between the two.

With that, DDR4 and PCI-E 4.0 are not expected to be in consumer hands until early 2015. Servers will first see it (and Intel Enthusiast Haswell-E) will see DDR4 late 2014. If AMD continues with dropping HyperTransport as they did starting with Trinity and adding a PCI-E controller on-die, it would be smart for AMD to wait for PCI-E 4.0. This way Excavator, if released around 2015 (earliest) to 2016 (latest), it will have the latest memory and PCI-E revision.

But, all of that is going to require an entirely new socket especially if they went with quad-channel (one DIMM per channel) DDR4 memory bus and PCI-E 4.0. Steamroller-based FX processors should be the last processors that'll be on AM3+ sockets. Anything after that should be as you said-- a future version of socket FMx. FM3? Who knows?

For a lot of AMD loyal buyers, it's a long ways to wait. I think that if patience pays off, we may see what AMD truly intends for the future of computing.
 
-Dragon- said:
If the last few years in the AMD forums are any indication of a pattern it will be something like this:
  1. Hyped.
  2. Late.
  3. Hot.
  4. Almost but not quite as big an improvement as the (revised) claims AMD said it would be.
  5. Far less of an improvement as the AMD fanboys swore it would be.
  6. Be disappointing (at least slightly to moderately depending on what hype you believed).
  7. Anyone who points out the disappointment will be labeled an Intel fanboy.
  8. Compared to $100 more expensive Intel chips (in "best AMD vs. best Intel" reviews) and be found lacking.
  9. Said review sites will be labeled Intel fanboys.
  10. That will include [H].
  11. Declared the performance/$ champion.
  12. Anyone who points out that the $ difference isn't that high especially when you consider that many of these chips find their way into $2000-3000 systems to where the extra cost becomes less than 5% of the overall system cost will be labeled Intel fanboys.
  13. Be declared to have much higher OC headroom when someone uses dilithium crystals mined from the rings of Saturn and chilled to 5 degrees below absolute zero manages to do a CPU-Z validation with the clock speed 5MHz higher than previous record attempts for 5 seconds before the dilithium crystals destabilizes and irradiates a large area of the state or country said record was attained in.
  14. Anyone who points out that it's not 24/7 stable or remotely reasonable to maintain and that real world OC results aren't all that different from Intel chips, and that the Intel chips start out at a lower clock speed to begin with so that for day-to-day use that the Intel chips actually OC better will be labeled an Intel fanboy.
  15. That will include me.
  16. Won't OC as well as expected in general use to begin with.
  17. Will be re-released in a year as a 20% faster clocked part with no other noticeable differences for 3x the cost and about 90% less OC headroom and way way more power usage.
  18. Nobody will put together that 17 might have something to do with the 16.
  19. The legacy of Hector Ruinz will be blamed by many for the majority of these failures/shortcomings
  20. That will include me.
That about covers my predictions for the next few AMD releases, at least until AMD does its own NetBurst -> Core transition
Click to expand...

what he said. I still don't see a reason to buy anything but an X6 for gaming. AMD got a whole crapton right with that one.
 
Yeah, we're definitely a few years off from seeing the 100% HSA stuff, but if it really takes off, the world of computing will be revolutionized more than it has in many of the previous years combined. I think AMD did say late last year that Steamroller FX will come to Socket AM3+, so it'll probably be the last CPU for that socket.

@Dragon: You tried too hard there. Just about everything in there is just concerns of trolls across the web who were never interested in AMD chips to begin with. I will say with the utmost confidence that Steamroller/Kaveri will blow Piledriver out of the water. In terms of IPC, even a modest boost up from bdver2 would be nice. I personally don't need i7-3960k performance, so I'm not worried about anything.
 
-Dragon- said:
nonsense removed
Click to expand...

It is so nice of you to type all that stuff I am sure that you took a lot of time and effort thinking up all these things.

Can you explain what this has to do with the delay of Kaveri or you proud of typing a lot of stuff just to feel good about yourself ?
 
That article looks like rubbish. Everything in that article was already revealed ages ago elsewhere (well, if you count rumors as reveals, but whatever). Also, I love how they say it'll have PCIe 3.0 -- as if it wasn't expected -- when FM2+ has already been confirmed to have PCIe 3.0.

The codenames have been revealed months ago. And what do they mean engineering samples won't be ready til August? Engineering samples were already spotted like two months ago, which is where most people got these codenames from! And Lisa Su said that select partners were already playing with chips, too, as well as OEM's.

So... if the MOBO's are coming very soon, and all that I just said happened, what reason would these have to be delayed? So much for "confidential documents". What's so special about this site that only they would get supposed super-secret documents from AMD?

I'm calling BS.
 
I find it quite funny that all those rumored changes and improvement in multi threaded part are in fact AMD backpedalling from their Bulldozer vision back into 2 cores being 2 cores.
 
NaroonGTX said:
@Dragon: You tried too hard there. Just about everything in there is just concerns of trolls across the web who were never interested in AMD chips to begin with. I will say with the utmost confidence that <insert upcoming gen AMD chip here> will blow <insert last gen chip that underperformed but we originally thought would be great here> out of the water.
Click to expand...
Fixed so you can copy and paste it each year.
 
Oh look, two more trolls. Dunno why they keep coming here, but I guess that's just what you do when you get bored.

I fail to see how them fixing a design flaw (two integer units in a module sharing a single decoder) is somehow "backpedaling". The architecture is still mostly the same, they're just tweaking various things with each revision. But clearly, that's backpedaling. Makes perfect sense.

As for bigdogchris, your comment is full of absolute fail. Just about every AMD release improved upon the previous release in their entire history of chips, with the only real exception being the transition from Stars to bdver1. Piledriver fixed various things and was better than K10/K12 at that point. Steamroller will blow Piledriver out of the water, both in single-threaded and multi-threaded workloads.

But of course, it'll "underperform" in your eyes because it won't beat an i7-3960x. Gtfo of here with that garbage, lol. Troll harder.
 
Just about every AMD release improved upon the previous release in their entire history of chips, with the only real exception being the transition from Stars to bdver1.
Click to expand...

Phenom I was also a big disappointment. Although Phenom II fixed most of the problems a year or so later. I can not remember any other backtracking.
 
NaroonGTX said:
Oh look, two more trolls. Dunno why they keep coming here, but I guess that's just what you do when you get bored.
Click to expand...

Major butthurt detected.

I fail to see how them fixing a design flaw (two integer units in a module sharing a single decoder) is somehow "backpedaling". The architecture is still mostly the same, they're just tweaking various things with each revision. But clearly, that's backpedaling. Makes perfect sense.
Click to expand...

Adding things into a module increases module size -> therefore it get's closer to previous die area occupied by separate cores.

As for bigdogchris, your comment is full of absolute fail. Just about every AMD release improved upon the previous release in their entire history of chips, with the only real exception being the transition from Stars to bdver1. Piledriver fixed various things and was better than K10/K12 at that point. Steamroller will blow Piledriver out of the water, both in single-threaded and multi-threaded workloads.
Click to expand...

And all of that to finally reach IPC they had with Phenom II 5 years ago :D

But of course, it'll "underperform" in your eyes because it won't beat an i7-3960x. Gtfo of here with that garbage, lol. Troll harder.
Click to expand...

LOL it won't outperform 3 generations old Sandy Bridge 2500K.

I had 3 AMD cpus - Palomino which was very good , Barton which was amazing and Phenom II X6 which was overpriced crap. Maybe if FX8xxx was 100$ cpu I'd see some value in it but even then it wouldn't be able to even reach performance i want from my pc.
 
You must log in or register to reply here.
Back
Top