DDR3 Memory Scaling Performance With AMD's Athlon 5350

[halfmanhalfamazing]

I'm not sure but i believe that for same number of cores and same frquency, Jaguar might indeed beat the c++p out of Kaveri,etc:

Thanks, that helps a lot.

http://www.anandtech.com/show/7933/the-desktop-kabini-review-part-1-athlon-5350-am1/4

[Spittie]

I'm not sure but i believe that for same number of cores and same frquency, Jaguar might indeed beat the c++p out of Kaveri,etc:

Just to point out a little thing, Jaguar has dedicated FPU for every core, while Steamroller/Piledriver have only a shared FPU for every two cores.
This is not to say that AMD didn't do a great job on Kabini (they did!), but don't expect results so good on integers-heavy tests too.

[AJSB]

Just to point out a little thing, Jaguar has dedicated FPU for every core, while Steamroller/Piledriver have only a shared FPU for every two cores.
This is not to say that AMD didn't do a great job on Kabini (they did!), but don't expect results so good on integers-heavy tests too.

I know , i know...makes us wonder however, why AMD continues to use only one FPU per two CPU cores in their main products....it?s kinda like KTM continues to push for their linkless rear suspension in their Enduro motocycles

[liam]

Wait until the A8-7600 comes out. Thats the 'mainstream' Kaveri chip. The 7850k is the unlocked flagship, so its even more of an apples-to-oranges situation.

The latest date I've heard for that chip is near the end of the year, for consumers, though OEMS will be getting it sometime this quarter.
That's a huge disappointment since that chip, running at 45w makes for a fantastic little htpc/gaming machine.

[_ONH_]

I know , i know...makes us wonder however, why AMD continues to use only one FPU per two CPU cores in their main products....

Good question why does not every car run up to 300kmph? The cheap answer ist no one will use it. So goes it with the FPU Particle simulations could also calculated on the GPU, it might needs different formulas but should be more efficient with higher numbers of particles.

On the 7zip Test you see the Modules is better with integer workloads, which should be used more often if I recall it correct.

Its the same as the discussion with belt driven bikes vs the ones driven with a chain. There are pro and con on both sides, typically you choose the one that fits your use case best.

[Ericg]

The latest date I've heard for that chip is near the end of the year, for consumers, though OEMS will be getting it sometime this quarter.
That's a huge disappointment since that chip, running at 45w makes for a fantastic little htpc/gaming machine.

Seriously? Damn it >< I was hoping itd be sooner.

An interesting read for people who like to examine CPU architecture... http://www.extremetech.com/computing...ad-performance

Its not totally up to date, so the information may not apply for Kaveri / Steamroller, but its a general overview of what went wrong with AMD's Bulldozer.


Also, would anyone be interested in some Windows 7 x64 vs Fedora 20 / 21 x64 Kaveri (7850k, 2133mhz RAM, 7200RPM drive) benchmarks in the coming months? I've got no problem running as much of the Phoronix Test Suite on my new desktop once it gets built as the first-run stress test.

[bridgman]

Is Kaveri "slower" than Kabini? I don't ask to be provocative, I honestly don't know.

Is there a 'mainstream' Kabini?

This is why I want to see benchmarks of the two together.(just these two)

Kaveri has Piledriver cores instead of Jaguar cores; Piledriver can run at much higher clocks than Jaguar. Kaveri also has 4x the GPU shader power (8 CUs vs 2 CUs) and >2x the memory bandwidth so in GPU-limited workloads Kaveri can be much faster than Kabini.

I don't think we have formally identified a "mainstream" Kabini but to me the Athlon 5150 seems like the best fit. For what it's worth, I would call the A10-7700K the corresponding "mainstream" Kaveri rather than the A8-7600.

I know , i know...makes us wonder however, why AMD continues to use only one FPU per two CPU cores in their main products....

I think it's because those products also include (or are used with) big-ass GPUs that have very high floating point throughput. Most of the server business is integer-intensive. Floating point and highly parallel work is gradually being pushed off to GPUs or to dedicated CPU arrays.

[Kivada]

So what about CAS timings at 1600Mhz? Can you force say 8-8-8-24 or tighter? If so what gains can be had from that?

[Kivada]

I can't help but be curious (even though the chips appear to be in different classes) how Kaveri and Kabini stack up against each other.

I have not seen benchmarks out there at all anywhere on this. Maybe I just missed them. If not, might make a good Phoronix article.

By the end of Q2 theres supposed to be an A4 Kabini APU out, rumors are it's supposed to only be dual core but have a faster GPU and will have the full dual channel memory controller that the rest of the desktop parts have.

[bridgman]

By the end of Q2 theres supposed to be an A4 Kabini APU out, rumors are it's supposed to only be dual core but have a faster GPU and will have the full dual channel memory controller that the rest of the desktop parts have.

If it's dual-channel it seems more likely to be a cut-down Kaveri than a bumped-up Kabini...