Octo core Qualcomm

When the board software throws up an error message there's no need to re-post your post; it has been posted, you just don't see it due to the aforementioned error. :)

Anyway, isn't it pretty much par per course to poo on tech which you can't offer (at the moment), only to exalt it once you can. We've seen many examples of that in the past.
 
pytUqhZ.jpg


:LOL:
 
This doesn't look like the world's most convincing leak. And not just because Qualcomm said 8-core SoCs were dumb. If true this would be the first SoC with 8 legitimate higher end mobile cores. They don't even get an added flexibility benefit since they already have separate DVFS for each core. Which would add a lot of complexity

They also don't have the power budget to run 8 cores at anywhere remotely near 2.5GHz in something like a BlackBerry phone, even on 20nm, which will only have modest efficiency improvements over 28nm anyway. They'd get a bit of a boost in worthless benchmarks like AnTuTu but not a huge one, and that'd be about it. Not worth the huge premium in die area and complexity for both the SoC and PMIC. The process shrink does give them some more transistors to play with for similar package sizes, but they'd be better off spending it on enhancements that improve core perf/MHz and GPU performance.

I think when anyone goes for > 4 higher end, homogeneous cores in an SoC it'd make sense to hit 6 cores first.
 
SoC designers will need to go 8 cores eventually, once they aim for current-gen consoles' performance without pushing frequencies to ~3GHz.

I'm not saying I believe Blackberry will launch a smartphone with an 8-core Qualcomm SoC. But just because someone at Qualcomm said 8 cores were dumb in 2013, it doesn't mean it'll still be dumb in 2016 or later. Besides, Qualcomm could decide to implement their own big.LITTLE design.
 
There's still some headroom to push IMHO for CPU IPCs and more advanced multi-threading before they'll turn to 8 core configs.
 
There's still some headroom to push IMHO for CPU IPCs and more advanced multi-threading before they'll turn to 8 core configs.

Look at Apple and as a matter of fact Qualcomm's own Krait.

8-core is just a last resort / desperate move for the ppl having to use off-the-shelf ARM cores (Samsung, Broadcom, Mediatek etc).

Though it seems like Samsung, Nvidia and Broadcom have an ARM arch-license as well.
 
I obviously don't count a big.LITTLE 4+4 config as an "octacore" to avoid misunderstandings. In a proper implementation you can utilize all 8 cores but the primary purpose of the second low end core cluster is to save power.
 
I obviously don't count a big.LITTLE 4+4 config as an "octacore" to avoid misunderstandings. In a proper implementation you can utilize all 8 cores but the primary purpose of the second low end core cluster is to save power.

If it can use 8 cores at once, it's an octocore. Regardless of primary or secondary purpose of each cluster..
 
If it can use 8 cores at once, it's an octocore. Regardless of primary or secondary purpose of each cluster..

....and if I'd have now 4 cores, but each core capable of two parallel threads at all times what I do have then?
 
A non castrated Intel core chip?

:LOL:

My only other point is that I don't like the idea of 8 cores for ULP, because there's still enough low hanging fruit to pick. ARM doesn't seem to oppose in its public statements against hw multi-threading for the future.
 
....and if I'd have now 4 cores, but each core capable of two parallel threads at all times what I do have then?

You'd have a 4-core/8-thread CPU, like my Xeon E5.


:LOL:

My only other point is that I don't like the idea of 8 cores for ULP, because there's still enough low hanging fruit to pick. ARM doesn't seem to oppose in its public statements against hw multi-threading for the future.

Seeing how big.LITTLE chips with Cortex A7 doing the LITTLE part fail to have a substantially lower power consumption compared to the SoCs with a quad Cortex A9 or Krait 300/400, I have to agree.
 
The latter might be due to problematic big.LITTLE implementations. Let's see what it'll look like in the future.
 
SoC designers will need to go 8 cores eventually, once they aim for current-gen consoles' performance without pushing frequencies to ~3GHz.

I'm not saying I believe Blackberry will launch a smartphone with an 8-core Qualcomm SoC. But just because someone at Qualcomm said 8 cores were dumb in 2013, it doesn't mean it'll still be dumb in 2016 or later. Besides, Qualcomm could decide to implement their own big.LITTLE design.

To match consoles they'd also need to have GPUs that would consume at least 50W on today's processes, but there's no way they'll have that in 2016 either.

If it can use 8 cores at once, it's an octocore. Regardless of primary or secondary purpose of each cluster..

Technically yes, but the point here is that multi-cluster ARM SoCs - be it real big.LITTLE or homogeneous - can have more flexibility by offering per-cluster DVFS. But this rationale doesn't make sense to Qualcomm who already has per-core DVFS.

Seeing how big.LITTLE chips with Cortex A7 doing the LITTLE part fail to have a substantially lower power consumption compared to the SoCs with a quad Cortex A9 or Krait 300/400, I have to agree.

Seeing how Cortex-A15s SoCs are much more powerful than Cortex-A9 ones I don't see how the comparison is relevant - if anything, a big.LITTLE solution getting good

But the other side of this is that under many typical utilization tasks like web browsing, messaging, simple 2D games, videos, and calls, the SoC power consumption takes back seat to a lot of other parts of the phone. So the difference is not nearly as critical as the SoC makers make it sound.
 
This doesn't look like the world's most convincing leak. And not just because Qualcomm said 8-core SoCs were dumb. If true this would be the first SoC with 8 legitimate higher end mobile cores. They don't even get an added flexibility benefit since they already have separate DVFS for each core. Which would add a lot of complexity

They also don't have the power budget to run 8 cores at anywhere remotely near 2.5GHz in something like a BlackBerry phone, even on 20nm, which will only have modest efficiency improvements over 28nm anyway. They'd get a bit of a boost in worthless benchmarks like AnTuTu but not a huge one, and that'd be about it. Not worth the huge premium in die area and complexity for both the SoC and PMIC. The process shrink does give them some more transistors to play with for similar package sizes, but they'd be better off spending it on enhancements that improve core perf/MHz and GPU performance.

I think when anyone goes for > 4 higher end, homogeneous cores in an SoC it'd make sense to hit 6 cores first.

"New to the Roadmap: 64-bit and Octa-core"

28nm

"I do have to point out that the Snapdragon 615 violates Anand Chandrasekher's Things that are Dumb list, but again it seems like Qualcomm is simply doing what the Chinese OEMs want. I could have a longer discussion about whether or not it's smart to listen to your customers if they are leading you astray, but let's see how this one pans out once Qualcomm shifts back over to its own CPU core IP next year. "

http://www.anandtech.com/show/7784/...-64bit-warpath-with-48core-cortex-a53-designs

It appears to me that mediatek is putting Qualcomm under unbeleivable pressure, if they drop their own cores in place of "standard" CPU cores, just so they can launch socs that have the "8 core" bullet point.
 
Last edited by a moderator:
if they drop their own cores in place of "standard" CPU cores, just so they can launch socs that have the "8 core" bullet point.
I think it has more to do that their own cores are not competitive at the low end. A7's and A53's are simply massively more efficient and smaller than Krait.

Anyway the 615 seems to be doing a HMP little.littler configuration, afaik if targeted, the A53's can achieve some high clocks.
 
Interesting tidbits:

They named these new chips as Snapdragon 610 and 615 (and not 500, 480, whatever).
If Qualcomm is consistent with their previous naming schemes, this is a sign that they have better performance than the Snapdragon 600, meaning:

- They consider the quad Cortex A53 (whatever their clockspeed) in the S610 to be faster than the quad 1.7GHz Krait 300 in S600
- They consider the Adreno 405 to be faster than the Adreno 320. Just like the Adreno 305 ended up quite a bit faster than the 220.


This should put the Cortex A53's performance in perspective. Unless they're clocking the CPUs to over 2GHz (which I find unlikely in what seems to be a mid-range SoC), it looks like the A53 should have higher IPC than Cortex A9
 
Back
Top