Hacker News Reader

Comment by Symmetry | Hacker News Reader

Wait, they're using the 64 bit K1 as their processor? That's Project Denver[1]! I'm super excited to see how that thing benchmarks. This is basically Transmeta's Efficeon (same IP and team, even) with a bunch of improvements. They mentioned in the Hot Chips talk that they had made improvements like enabling native execution while the optimizer works - they benchmarks they provided said that it was very effective but I'm eager for third party tests.

[1]http://en.wikipedia.org/wiki/Project_Denver

EDIT: And here's Anandtech's article on Project Denver http://www.anandtech.com/show/7622/nvidia-tegra-k1/2

View on HN

I would expect it to bench pretty much the same as the Shield Tablet. In fact, they are probably almost identical hardware-wise. I'll be interested to see if there is a price difference.

Anand's CPU benchmarks are very browser-oriented. The Shield usually beat all other Android devices, but still lost to the iPad Air. I suspect that has more to do with the quality of Mobile Safari than the differences in hardware. In the GPU benchmarks, the Shield does very, very well. http://www.anandtech.com/show/8296/the-nvidia-shield-tablet-...

Overall, the K1 has both CPU and GPU paper/marketing specs that are theoretically a little better than an Xbox 360. This is demoed by the 360 game Trine 2 running very nearly as well on a Shield as on the 360. http://www.eurogamer.net/articles/digitalfoundry-2014-can-tr...

edit: Looks like the Nexus has a 2048x1536 screen vs the Shield's 1920x1200. 51% more pixels on the same GPU will be a much prettier at the cost of more GPU|battery strain. Given the high power GPU, that seems like a good trade off for a non-gaming-oriented tab.

> I would expect it to bench pretty much the same as the Shield Tablet.

It's not the same chip as in the Shield Tablet (which has an ARMv7 CPU). The GPU is the same but the CPU is the 64 bit Denver CPU (ARMv8). According to benchmarks should be a fair bit faster, especially in single thread execution.

Which one should be faster, the N9 or the Shield?

The N9.

That's possible, but the Shield has a K1 varient with 4 32-bit A15 cores on the chip while the Nexus 9 has a varient of the K9 that has 2 64-bit Project Denver cores. The graphics are identical between the two versions of the K1 so I don't expect much difference in the GPU benchmarks, but I expect the CPU benchmark comparison to be interesting. On one hand we'll have four 3-wide conventional OoO cores. On the other we'll have two cores that are initially narrow and in order but are 7-wide VLIWs once the code morphing software has had a chance to run.

Is that big little cores or four symmetric cores?

Four symmetric cores - all A15s. I don't think NVidia has ever done big.LITTLE even though they had a fifth core with the same architecture but different process/gate sizes for lower power in some previous Tegra chips. But you can see from them all having the same max frequency that all four cores are the same.

32-bit K1 also has the companion core[1] (probaly Cortex A7).

[1] http://www.extremetech.com/wp-content/uploads/2014/01/tegra-...

The companion core of the ARMv7 TK1 is a low power Cortex-A15, like the grandparent comment says. This is different from ARM's big.LITTLE which uses different microarchitectures for the two clusters, i.e. Cortex-A15 + Cortex-A7 or Cortex-A53 + Cortex-A57.

It's too small to be A15. I found the information that the companion core was A15 in Tegra 4, but it's similar size to other cores[1].

[1]http://cnet4.cbsistatic.com/hub/i/2013/01/06/318d69ea-fdb7-1...

Sorry, but that looks more like an artist's vision of a die shot than an actual die shot. The TRM is pretty clear that it's an A15: http://i.imgur.com/Fvh2KVT.png

Core size isn't just a matter of the number of gates, but also their size. One important tool for reaching high frequencies in chips is to have gates that drive more capacities load be wider and therefore higher power. But if your goal is to have an architecturally identical core that's lower power one of the tools you have to accomplish that is to not do that. There's probably other ways to shrink a core at the cost of performance, as well.

Reply on news.ycombinator.com