John Carmack Suspicious Of Nvidia’s Outlandish Tegra K1 Claims

John Carmack Suspicious Of Nvidia’s Outlandish Tegra K1 Claims

When Nvidia unveiled the Tegra K1 last month, they made a great manyimpressive claims about its performance, underlying architecture, and the fact that it would debut as two different parts — a quad-core based on the Cortex-A15r3 with the now-standard Companion Core for added energy efficiency, and a dual-core 64-bit chip that supports ARM’s 64-bit instruction set. John Carmack himself has just tossed water on these claims, telling people that Nvidia’s statements should be taken “with several grains of salt.”

That’s a fairly strong statement for a developer to make, particularly one with Carmack’s influence. So what, exactly, has Nvidia said? The Tegra K1 whitepaper states that “A vast majority of Xbox 360 and PS3 titles could be easily ported to Tegra K1 devices with minimal effort, delivering similar visual quality and gameplay experience.”

Nvidia has claimed that the Tegra K1 is substantially more powerful than last-generation consoles and has trotted out charts like the following to prove it.

Nvidia has a history of playing fast and loose with product comparisons, but it’s not the first company to claim a new mobile part would offer console-level graphics — Sony andQualcomm have made similar statements in the past when trotting out new products or GPUs. I also agree with Carmack, however, that the numbers alone aren’t going to tell the whole story on this one.

The Tegra K1′s FLOPS count is much higher than the Xbox 360 or PS3, but its texture fill rate and memory bandwidth are both significantly smaller. The PS3 had eight render mapping units, whereas the K1 has just four. However, the K1′s architecture is significantly more efficient than the older console chips, and DX11 introduced some efficiency-boosting capabilities that DX9 simply lacked.

There’s also the fact that resolution targets make a substantial difference, especially in mobile, where devices are bandwidth constrained. One way to level the playing field compared to older consoles would be to natively render a game in 720p, then upscale it for 1080p or higher. Another is to use more efficient texture compression, to reduce bandwidth load, and improve overall performance.

There are two issues muddying Nvidia’s claims to console-equivalent performance. First, there’s the fact that we have no idea what clock speeds Tegra K1 is using to hit these performance targets. Is this the performance the chip can offer in a tablet? In a smartphone? In a device with active cooling, like the Nvidia Shield? The second is the company’s blithe assurances that console titles can be “easily” ported. That’s not necessarily true — the ARM microarchitecture is nothing like what either console used, the control systems are different, and any mobile port would have to develop touchscreen-based controls as well as physical button mapping if devices like the Shield were the target.

Personally, I’d split the difference on this one. I think Nvidia is overstating the degree to which titles can be brought to Tegra K1, partly for reasons that have nothing to do with the actual chip and everything to do with economics, ergonomics, and return on investment for the publisher. While I agree that one should take the console performance claims with a grain of salt, though, the fact is Kepler and K1 move us closer to that parity than we’ve been in the past. Even if Tegra K1 can’t match older consoles altogether, it should still be able to come far closer than anything we’ve seen before.