AMD's jaguar cores

P

pelo

2[H]4U
Joined
Apr 23, 2011
Messages
2,911
slide-1-728.jpg


AMD-Intros-Jaguar-Based-Kabini-and-Temash-APUs-Next-Month-Not-Next-Year-3.jpg


Jaguar_power_gating.png


Jaguar_core_pipeline.png


Jaguar_core_diagram.png


http://semiaccurate.com/2012/08/28/amd-let-the-new-cat-out-of-the-bag-with-the-jaguar-core/

Questions are still abound regarding the perf-per-watt it's able to offer, but providing a monumental AMD screwup, it's bound to be a good bump over Bobcat's perf-per-watt. This should bring some decent performance for Temash at the 4.5W TDP level and make it into tablets so long as OEMs are on board.

As an architecture, expect AMD's Kabini/Temash to outperform Atoms given the small improvements in the Atom line since the original Bonneville architecture. With the Atom Intel has opted to go with decreasing the overall TDP and power usage to fit the form factor while retaining the overall CPU performance (and increasing GPU performance withe the newer PowerVR GPU).

AMD's approach appears to be improving on Bobcat as a whole through the entirety of the TDP ranges. Their 9-25W TDP SKUs should do quite well and be a great upgrade over Bobcat, but they're also too big to fit into a typical tablet and may instead be stuck into the same designs that the ULV Intel Ivy/Haswell product lines. Thus Kabini should bring better than Bobcat performance, worse than Intel's ULV chips but substantially cheaper (maybe ~$100-$150 cheaper?).

The big question is how Temash will do given its very tight TDP restrictions (4.5W) and whether it can compete on a perf-per-watt basis with the likes of ARM and Intel's Atoms. Of all AMD's chips they've designed and released since the X2, the Bobcat was the only one that really put a hurting on Intel, essentially beating it in every single aspect. If AMD can provide a successful followup to the Bobcats (which happen to be their bestselling APUs even today) they'll provide a great second-tier option to the expensive >$1000 ULV Intel hybrids
 
I think if AMD keeps doing well with thier graphics, and keeps upping performance of thier various low power chips, they might be able to continue competing. They just need to stay focused, as long as they are making good design choices on the lower end, they might make enough $ to stay competitive on the mid and possibly even high end(once they get everything sorted out that is)
 
That was rumored, but that's unnecessary at the moment. ARM+x86 might help somebody like Google or Apple if they wish to transition from one ISA to the other, but for AMD themselves it offers little benefits. It would just mean a bigger die with the unused cores sitting idle thus higher price tag
 
why?
If you can use the ARM for surfing the web or even just word processing and such things and the heavy duty cores for more heavy duty tasks, you might be able to save a good deal of energy.
 
junky said:
why?
If you can use the ARM for surfing the web or even just word processing and such things and the heavy duty cores for more heavy duty tasks, you might be able to save a good deal of energy.
Click to expand...

It really doesn't make a difference in this case as the power consumption of Jaguar cores at 4.5W TDP for the entire SoC would be very slim. Going forward that might make more sense but it's still more likely in ARM than it is in x86. Mixing the two together requires an OS and software to handle both, which is something only Apple would consider a possibility as a transition type of architecture if they wish to make the jump from OS X and iOS into a single OS with ARM powering it (remember they've got their own engineers now designing custom ARM architectures)

What you're getting at makes more sense in the strictly ARM ecosystem with it's big.LITTLE approach to multi-core architectures, whether A15+A7 or A57 and A53, but these are different core designs sitting on the same die. It would be like Intel using Haswell + Atom on the same die.

http://www.arm.com/products/processors/technologies/biglittleprocessing.php

Where a big core kicks in for strenuous workloads and the little core is the more efficient one taking care of everything else. While that makes sense for ARM, the x86 makers don't have that option as it would require using an OS and ecosystem that can take advantage of both ISAs and architectures fluidly. I guess Win8 could potentially do that, but using x86+ARM would be like admitting defeat for both AMD and Intel in the perf-per-watt segment. Intel has already proven that x86 doesn't necessarily have to be power hungry, but the price is another matter entirely.
 
Last edited:
I understand, although it is a question of how low in power consumption they can get.
 
junky said:
I understand, although it is a question of how low in power consumption they can get.
Click to expand...

If you believe ARM's own statements regarding the big.LITTLE approach, the benefits could prove to be substantial:

http://www.eetimes.com/electronics-news/4400437/Big-little-test-results-show-promise--ARM-reports

SANTA CLARA, Calif. – ARM reported a test chip using its A15 and A7 cores saved 50 percent on power consumption without sacrificing performance. It also revealed a road map for enabling software for the so-called big.little initiative which pairs a large and small core.

The technology looks promising enough that even some chip makers outside the ARM ecosystem—such as Intel—might consider adopting it, said one analyst. Samsung is expected to sample the first big.little SoC, perhaps before the end of the year. Several chip makers have similar SoCs in development, said ARM.
Click to expand...

It could be valid for both AMD and Intel, but they would need even more frugal core designs for it to fit certain TDP thresholds for phones/tablets. Currently, it just isn't feasible for either of them in either segment. For PCs - desktops/laptops and even servers- it could be a future strategy. Small Temash cores plus Vishera modules? two Atom cores and two Haswell cores? Who knows :)
 
They will be doing an APU type approcah first where the gpu and cpu are different construction method like the current fusion cpus, going forward they will be making both in a bulk process so as they can share thier memory and such better.

The first ARM amd combo will be arm amd core with amd graphics if I recall, but overall, they want an opteron arm type design, but as they have stated, they have to take a step at a time, and thier fusion type approach as they are doing with thier own cpus, will be the easiest way to achieve this. Opteron, is server and workstation, not desktop or mobile, for that they will do jaguar and similar chips for the time being.

Basically scale power down and more chips in a server should transition to higher overall performing(like the atom used in seamicros design and ARM is close to atom as far as power/performance was concerned) standard chis for mobile and lower power environments and standard desktop chips for well, desktop.
 
pelo said:
Joel's article on ET

http://www.extremetech.com/gaming/1...-notebooks-and-tablets-if-it-launches-on-time

Though he appears to be mistaken regarding Kaveri. The leaked roadmaps show Kaveri has disappeared and that Steamroller has either been pushed back or cancelled altogether, instead the Steamroller-based APU will be replaced by a Richland Trinity-derived APU on GloFo's 28nm bulk.
Click to expand...

From what I read on another forum Kaveri isn't scrapped, but won't be out for awhile. 2014 I think, but isn't that when 14nm is going to be down by GL? Maybe we'll see Kaveri on that instead of 28nm. Richland is going to be a piledriver cores on the 28nm supposedly offering 40% higher clocks within the same TDP, while having GCN cores. Sounds like this improvement would be better than going for Kaveri currently due the higher clocks gaining more performance vs increase IPC. Sounds like RIchland could be decent and won't be a laughing stock compared to Haswell hopefully.

Read another snippet (forgot where) that the next MS Surface "Pro" could have the jaguar cores in it too. Could be interesting to see AMD pull that off.
 
Last edited:
I highly doubt we'll see 40% higher clocks on the same TDP unless there was something drastically wrong with the 32nm FD-SOI process and Trinity. 40% isn't a small number, it's absolutely huge.

Given that 28nm bulk will provide roughly the same performance benefits as 32nm FD SOI, that number is unrealistic to say the least.

It might be that Richland sees more aggressive turbo or simply rearranged turbo and they've managed to gain some TDP headroom for the CPU side with more aggressive clock gating, but even still a 40% gain is highly unlikely.
 
It was based on speculation from this. Should of mentioned it probably is with use of the turbo, and the guy there probably misinterpreted that. All speculation of course. Nothing official.
 
You must log in or register to reply here.
Back
Top