AMD considered GDDR5 for Kaveri, and there could be an 8-core variant coming.

[Vitalius]

Post at /.

 

 

MojoKid writes

 

"Of all the rumors that swirled around Kaveri before the APU debuted last week, one of the more interesting bits was that AMD might debut GDDR5 as a desktop option. GDDR5 isn't bonded in sticks for easy motherboard socketing, and motherboard OEMs were unlikely to be interested in paying to solder 4-8GB of RAM directly. Such a move would shift the RMA responsibilities for RAM failures back to the board manufacturer. It seemed unlikely that Sunnyvale would consider such an option but a deep dive into Kaveri's technical documentation shows that AMD did indeed consider a quad-channel GDDR5 interface. Future versions of the Kaveri APU could potentially also implement 2x 64-bit DDR3 channels alongside 2x 32-bit GDDR5 channels, with the latter serving as a framebuffer for graphics operations. The other document making the rounds is AMD's software optimization guide for Family 15h processors. This guide specifically shows an eight-core Kaveri-based variant attached to a multi-socket system. In fact, the guide goes so far as to say that these chips in particular contain five links for connection to I/O and other processors, whereas the older Family 15h chips (Bulldozer and Piledriver) only offer four Hypertransport links."

 

Well. That's a cool idea. Sad it didn't happen (maybe next APU gen). I can't imagine how good an APU would be with it's own personal GDDR5 RAM. Probably like this, but without the crossfire (as I assume the boost is partly because the APU is getting GDDR5 RAM).

I'm... blank on what having an extra Hypertransport link means. I assume higher bandwidth within the motherboard itself? (i.e. more PCI-e bandwidth specifically) I could be wrong there. 

I ... just have no idea what Hyperstransport is. I assume it is what I said (effectively what gives the bandwidth within the motherboard between all the components/interfaces). 

An 8 core Kaveri would be awesome. As long as it isn't sharing core resources (like FX chips do), but I figure it wouldn't as AMD has learned their lesson there. 

Interesting comments section:

 

Anandtech's writeup (which Hothardware seems to be ripping off) has a much better explanation of what's going on and why it matters.

 

"Let me be very clear here: there's no chance that the recently launched Kaveri will be capable of GDDR5 or 4 x 64-bit memory operation (Socket-FM2+ pin-out alone would be an obvious limitation), but it's very possible that there were plans for one (or both) of those things in an AMD APU. Memory bandwidth can be a huge limit to scaling processor graphics performance, especially since the GPU has to share its limited bandwidth to main memory with a handful of CPU cores. Intel's workaround with Haswell was to pair it with 128MB of on-package eDRAM. AMD has typically shied away from more exotic solutions, leaving the launched Kaveri looking pretty normal on the memory bandwidth front."

 

It's also worth noting that the Anandtech article implies that AMD is still on the fence on Kaveri APUs with more memory bandwidth, and that it may be something they do if there's enough interest/feedback about it.

 

 

But the "dirty little secret" that doesn't get brought up enough frankly is a lot of those "single threaded loads" is as rigged as quack.exe was back in the day [theinquirer.net] thanks to every Intel compiler made since 2002 being made to put out crippled code for any chip that Intel doesn't want to push. Oh and for those that use the "Intel just knows their own chips and optimizes for them" excuse that lie has been disproved and the proof was the last gen Pentium 3. You see the last gen P3 was curbstomping the Netburst P4s in early benchmarks, yet when the cripple compiler comes out? Suddenly the very same Netburst chips are winning by 30%!

 

And the bitch is that any of these so called review sites could test for rigging trivially but they won't for fear of losing Intel advertising revenue. To see if a program is rigged all one has to do is run the code on a Via CPU, Via CPUs allow one to softmod the CPUID so if you change the CPUID from "Centaur Hauls" to "Genuine Intel" and suddenly the chip scores 20%-30%+ higher on the test? Then the program has been rigged by ICC, simple as that.

 

All of these sites like Anandtech and Tom's have more than enough money to pick up a Via chip and keep it around for testing but they won't bite the hand that feeds so everyone should consider their tests to be tainted and as worthless as Quake tests were with the rigged drivers. It would be nice if someone would run some real tests so we could see real numbers, wouldn't be hard to do as GCC is free and there are plenty of FOSS programs like Firefox one could compile with GCC to give accurate tests but so far no review site will do this for fear of pissing off Intel. How they didn't get busted for antitrust is beyond me, this is every bit as bad as "Windows isn't done until lotus won't run" but they were allowed to bribe AMD to the tune of 2 billion to drop the lawsuit and with it the investigation.

[Kevinkt]

i want to see the numbers for the 8core apu and 8350

[gastew15]

Sounds pretty awesome, maybe I'd upgrade to something like that if it actually turns into a thing. It's not really at the top of my to do list, since I can't imagine my current CPU being obsolete anytime soon, but still very cool regardless. I'm glad to see AMD is actually trying to do some cool stuff with their APU's. (Not that they weren't before of course)

[Gunjob]

While an 8 core does sound appealing, throwing more cores at the problem won't help with the lacking IPC improvements that AMD is lacking on CPU side. 

[Grave]

Doesn't sound as awesome as it does a missed opportunity to bust through the doors with their miniguns from Predator blazing..

This is just making me sad, not curious about the future. Too many if's and maybe's. I feel like it just gives Intel more time to create space between them and AMD.

[brownninja97]

I would really like a 8 core apu that would be pretty awesome.

[qwertywarrior]

what happened to that memory CUBE thing AMD was investing in

[7850OC]

we can use a 8 cores APU with a low iGPU (like a R3) to replace the FX line on the FM2+

[samuel]

GDDR5 as system memory is too expensive and latencies will be quit a bit high. However...buying an FM3 moderboard with like 2 GB GDDR5 onboard soldiered to it reserved for HSA besides DDR4 slots sounds pretty damn good. I fully expect an 4GHZ+ 8core APU at Kaveri or better IPC.

[kuddlesworth9419]

AMD's answer to everything is to add more cores. Look where that has gotten them.

[Lord_Kitty]

AMD's answer to everything is to add more cores. Look where that has gotten them.

Interestingly, when Intel released the C2Q, AMD claimed it wasn't a real quad core because it was 2 C2D die in one package. Now with the FX line, they act all smug and claim they have the first 8-core CPU when even their patents list them as quad cores.

[TechFan@ic]

Quad channel GDDR5 support was in fact in the works. The Kaveri chips still support quad channel GDDR5 on the silicone level in fact, however there were two companies working with AMD on a special low cost GDDR5 memory for APUs. One of which went bust and was bought by a larger company, the other company was Hynix which did not want to handle all the risk associated with the project by itself.
A replacement for the company that went bust was not found and so the project was cancelled quite late in the development phase and that's why we see the quad channel memory on the Kaveri die but with no support.
 

what happened to that memory CUBE thing AMD was investing in

The memory cube (HMC) developed by Micron is something that Nvidia has invested in, simply explained it's stacked DRAM cells in one package and then connected to the GPU via an off-chip memory controller.

AMD invested in a more exotic type of memory called HBM developed by SK Hynix which requires the memory die to be stacked right on top of the GPU or APU within a multi-module package.

The memory controller for HBM is located inside the APU/GPU die which raises cost but has very clear cut advantages in latency and power consumption. Since the flash chips are controlled by an APU/GPU on-die memory controller the communications overhead is very small and the APU can have very fine grain control over power characteristics of the memory which reduces power consumption.
 

[TERAFLOP]

Interestingly, when Intel released the C2Q, AMD claimed it wasn't a real quad core because it was 2 C2D die in one package. Now with the FX line, they act all smug and claim they have the first 8-core CPU when even their patents list them as quad cores.

Those are entirely different things you're talking about, AMD shares resources between cores to reduce total area.

Intel on the other hand used two separate dual cores and connected them via a northbridge which posed a severe bottleneck.

Each time one of the cores wanted to communicate with the core on the other die several ms of latency would be wasted on the communications overhead.

The impact on performance can be very clearly seen now that we have good multi-threaded games.

If you run a core 2 duo vs a quad on the same frequency in games you will see significantly lower gains compared to running a dual core i3 or dual core phenom vs a quad core i7 or a quad core phenom.

[Vitalius]

GDDR5 as system memory is too expensive and latencies will be quit a bit high. However...buying an FM3 moderboard with like 2 GB GDDR5 onboard soldiered to it reserved for HSA besides DDR4 slots sounds pretty damn good. I fully expect an 4GHZ+ 8core APU at Kaveri or better IPC.

 

The actual cost difference between having 4GB of VRAM and 2GB of VRAM is very little ($20-$30 or so). I can't imagine that'd be too expensive.

AMD's answer to everything is to add more cores. Look where that has gotten them.

Say that in 2 years when everything is multi-core optimized. Fun Fact: That's a trick because it still applies. On top of the industry.

[kuddlesworth9419]

The actual cost difference between having 4GB of VRAM and 2GB of VRAM is very little ($20-$30 or so). I can't imagine that'd be too expensive.

Say that in 2 years when everything is multi-core optimized. Fun Fact: That's a trick because it still applies. On top of the industry.

Hopefully although we have had multiple-core processors for a while know yet not much has been done. Just so you know though having a CPU with multiple week cores won't be as good as a CPU that has fewer cores that are stronger, unless you have a 16 core CPU vs a 4 core CPU that are both running a good multi-core optimized processor. AMD should work on finding an efficient architecture first and then start adding cores to it rather than just adding cores to a sh*ty architecture that is Steamroller. Kaveri is an even more efficient architecture then what Steamroller is. Steamroller is holding back AMD's APU's and the CPU's. If they can make a strong CPu architecture that is on par with Intel's then AMD will be the top dog once again.

[Suika]

An octacore with decent performance per core would be nice. Nothing like that FX 9590 though. That was stupid.

[TechFan@ic]

Going forward with HSA processors you might see an APU product from AMD which has 2 or even 3 different x86 cores.
One with very high single-threaded performance for branchy code or gameplay code inside game engines, another with significantly less single-threaded performance but significantly higher total throughput for productivity and HPC and a third low power Jaguar core which is used in the idle states of the machine or when watching videos to save power.

[Lord_Kitty]

Those are entirely different things you're talking about, AMD shares resources between cores to reduce total area.

Intel on the other hand used two separate dual cores and connected them via a northbridge which posed a severe bottleneck.

Each time one of the cores wanted to communicate with the core on the other die several ms of latency would be wasted on the communications overhead.

The impact on performance can be very clearly seen now that we have good multi-threaded games.

If you run a core 2 duo vs a quad on the same frequency in games you will see significantly lower gains compared to running a dual core i3 or dual core phenom vs a quad core i7 or a quad core phenom.

 

 

Yes, I know about that on the Core 2 Quads. But that wasn't the point. Intel made a quad core and AMD was quick to say otherwise. Now AMD made a quad core with 2 clusters but they are all high and mighty with their 8-core CPUs.

 

I could say the same thing for APUs. Intel was the first one to put a GPU on a CPU package. Though AMD's approach is different and better with HSA, they really should not say things like why doesn't Intel call their chips APUs.

[Vitalius]

Yes, I know about that on the Core 2 Quads. But that wasn't the point. Intel made a quad core and AMD was quick to say otherwise. Now AMD made a quad core with 2 clusters but they are all high and mighty with their 8-core CPUs.

 

I could say the same thing for APUs. Intel was the first one to put a GPU on a CPU package. Though AMD's approach is different and better with HSA, they really should not say things like why doesn't Intel call their chips APUs.

Again, it's not the same thing. 

What Intel did was basically package two dual cores as a quad core when if you did use all 4 cores, it will heavily bottleneck. 

AMD packaged an octo core CPU that was slightly handicapped due to how it was made. 

Intel did the equivalent of me taking two dual core CPUs and soldering them to a Motherboard together beside each other. 

AMD actually made an 8 core CPU that shared resources. 

Not the same thing.

[Vitalius]

Repost. 

[Kilerbomb]

I just hope the single threaded performance is better... That's seriously holding them back. They've pretty much got the GPU portion down, it's CPU that's the problem.