NVIDIA Highlights Tegra K1 Denver 64-bit SoC

NVIDIA has claimed their "Denver" variant of Tegra K1 will be the first 64-bit ARM processor for Android. In order for that claim to remain true, the Denver version of Tegra K1 will have to launch in the coming months before competition such as Qualcomm and MediaTek can launch their Cortex A53 based SoCs.

Mile High Milestone: Tegra K1 "Denver" Will Be First 64-bit ARM Processor for Android

Our 32-bit Tegra K1 mobile processor has been racking up praise for bringing amazing performance and true console-quality graphics to the mobile space.

It “handily beats every other ARM SoC” in GPU performance benchmarks, according to Anandtech. And “the GPU performance is what stands out with the Tegra K1, nothing else on the market today is really able to get even close,” according to PC Perspective.

Now, eight months after unveiling Tegra K1’s 32-bit version, we’re providing further architectural details of the chip’s 64-bit version at HOT CHIPS, a technical conference on high-performance chips.

You can get more technical details here, while below is a general view of what we presented:

This new version of Tegra K1 pairs our 192-core Kepler architecture-based GPU with our own custom-designed, 64-bit, dual-core “Project Denver” CPU, which is fully ARMv8 architecture compatible. Further, Denver is fully pin compatible with the 32-bit Tegra K1 for ease of implementation and faster time to market.

With its exceptional performance and superior energy efficiency, the 64-bit Tegra K1 is the world’s first 64-bit ARM processor for Android, and completely outpaces other ARM-based mobile processors.

Credit NVIDIA Blog

Highest Single-Core CPU Throughput

Denver is designed for the highest single-core CPU throughput, and also delivers industry-leading dual-core performance. Each of the two Denver cores implements a 7-way superscalar microarchitecture (up to 7 concurrent micro-ops can be executed per clock), and includes a 128KB 4-way L1 instruction cache, a 64KB 4-way L1 data cache, and a 2MB 16-way L2 cache, which services both cores.

Denver implements an innovative process called Dynamic Code Optimization, which optimizes frequently used software routines at runtime into dense, highly tuned microcode-equivalent routines. These are stored in a dedicated, 128MB main-memory-based optimization cache. After being read into the instruction cache, the optimized micro-ops are executed, re-fetched and executed from the instruction cache as long as needed and capacity allows.

Effectively, this reduces the need to re-optimize the software routines. Instead of using hardware to extract the instruction-level parallelism (ILP) inherent in the code, Denver extracts the ILP once via software techniques, and then executes those routines repeatedly, thus amortizing the cost of ILP extraction over the many execution instances.

As part of the Dynamic Code Optimization process, Denver looks across a window of hundreds of instructions and unrolls loops, renames registers, removes unused instructions, and reorders the code in various ways for optimal speed. This effectively doubles the performance of the base-level hardware through the conversion of ARM code to highly optimized microcode routines and increases the execution energy efficiency.

The slight overhead of the dynamic optimization process is outweighed by the performance gains of already having optimized code ready to execute. In cases where code may not be frequently reused, Denver can process those ARM instructions directly without going through the dynamic optimization process, delivering the best of both worlds!

Dynamic Code Optimization works with all standard ARM-based applications, requiring no customization from developers, and without added power consumption versus other ARM mobile processors. That’s because the 7-wide superscalar design allows faster throughput than would otherwise be possible at the same clock speed.

Credit NVIDIA Blog

Denver’s remarkable design delivers great performance for both single- and multi-threaded applications, as well as multitasking scenarios. The dual-CPU cores can attain significantly higher performance than existing four- to eight-core mobile CPUs on most mobile workloads.

Denver also features new low latency power-state transitions, in addition to extensive power-gating and dynamic voltage and clock scaling based on workloads. Combining Dynamic Code Optimization, 7-way superscalar design and efficient power usage, Denver’s performance will rival some mainstream PC-class CPUs at significantly reduced power consumption.

This means that future mobile devices using our 64-bit Tegra K1 chip can offer PC-class performance for standard apps, extended battery life and the best web browsing experience – all while opening new possibilities for gaming, content creation and enterprise apps.

Look forward later this year to some amazing mobile devices based on the 64-bit Tegra K1 from our partners. And for hard-core Android fans, take note that we’re already developing the next version of Android – “L” – on the 64-bit Tegra K1.

Credit NVIDIA Blog

So one year later we have a true 64bit announcement from someone else. It takes me back...

“The 64-bit Apple chip hit us in the gut,” says the Qualcomm employee. “Not just us, but everyone, really. We were slack-jawed, and stunned, and unprepared. It’s not that big a performance difference right now, since most current software won’t benefit. But in Spinal Tapterms it’s like, 32 more, and now everyone wants it.”

At the time Qualcomm CMO Anand Chandrasekher called the chip “a marketing gimmick," though Qualcomm quickly put out a statement in which it walked back Chandrasekher’s comment and called it “inaccurate.” Soon after that, Chandrasekher was reassigned. Whoops.

But once Apple introduced a 64-bit processor, all the other phone-makers wanted one too. “Apple kicked everybody in the balls with this. It’s being downplayed, but it set off panic in the industry.”

Credit HubSpot

That whole article is just beautiful. It's like reading how far ahead of the game Apple really is. That's why I always say that most of what the competition does feels like catch-up, even before Apple releases anything of the sort (read: smart watches).

@musicalmac

Intel released their x86-64 Silvermont cores last year on a 22 nm FinFET process and will be releasing Broadwell (Core M) and Airmont (Cherry Trail) on 'real' 14 nm FinFET this year.

Apple released an ARMv8 CPU core on a 28 nm process, while the majority of companies had plans for ARMv8 on a 20nm process. Depending on the application, the move to 64-bit on Apple's A7 didn't offer that much of a performance increase. Futuremark only measured a 7% gain between 32-bit 3DMark Ice Storm, and 64-bit 3Dmark Ice Storm on Apple's A7. The impressive part was the move to ARM's ARMv8 architecture.

Denver represents a first for NVIDIA, a complete in-house ARMv8 CPU core. Although NVIDIA is releasing it with a Tegra K1 variant, we'll likely see it play a larger role in NVIDIA's 2015 Erista (20 nm, Maxwell GPU).

No ARM SoC in 2014 will be able to match the capabilities of NVIDIA's Kepler GPU used in Tegra K1. The PowerVR 6XT GPU used in Apple's A8 will be on a 20 nm process (Tegra K1 is on a 28 nm process), depending on the clocks, it might beat out Tegra K1 in raw performance, but it will still be limited to OpenGL 3.x (and similar) features.

Apple released their first OpenGL ES 3.0 capable SoC with the A7, while companies like Qualcomm had GLES 3.0 support in their Snapdragon S4 Pro SoC a full year before. Qualcomm launched LTE-A compatible modems a year ago, with Intel launching their LTE-A modems this year, and there is still no LTE-A support (or even LTE 2600 support) in Apple's A7. Qualcomm has confirmed Cortex A57 (ARM's in-house large ARMv8 CPU cores) and LPDDR4 RAM for their 20 nm Snapdragon 810 SoC coming in H1 2015, so they're not exactly falling behind.

For a company that you claim is far ahead of the game, It would appear Apple still has some catching up to do.

Hopefully this will be in the Surface 3 RT or whatever they will call it and will run Windows Threshold

Hopefully not. With Core M and Cherry Trail in Q4, there is no reason any Windows tablet should have to run on ARM hardware. Windows RT was a mistake.

@NVIDIATI: You find me an example of Apple being caught so off guard by any new development that people likened it to a kick in the balls that set off an industry-wide panic, and you'll have my attention.

I agree. If you're going to commit to something, commit to it. Don't fire blindly into the night, Samsung-saturating everything to find a foothold. Make a decision. RT was a mistake.

@musicalmac That still doesn't equate to Apple being "so far ahead". The industry knows what a company like Intel is producing a year or two ahead of an actual product release (most hardware companies release a timeline of future products). They're an established industry leader and no one is surprised when they push the envelope beyond what is currently possible. The same way no one is surprised that Intel is releasing a Broadwell SoC with a TDP < 4.5 Watts on a true 14 nm FinFET process.

@NVIDIATI: ...and yet industry insiders compared Apple's move to a kick in balls that set off an industry panic. There's no way to downplay the significance of such a reaction. Everyone was surprised.

@musicalmac Apple surprising industry insiders does not justify the claim that Apple is "so far ahead". Especially when a competition, such as Intel, has hardware and developments that are ahead of Apple.

@NVIDIATI: Well, okay, however you want to look at it.

I wonder when someone will manage to surprise Apple the way Apple has surprised the industry. Like the article says, you can downplay it, but it really set off a panic in the industry. Panic. Their words, not mine. You don't set off a panic without being a few steps ahead.

@musicalmac I'm not interested in their words, or downplaying the "surprise", so stop trying to hide behind an article.

The reality is, Apple is not at the leading edge in many aspects of SoC development, nor is it in a position for you to claim, that Apple is "so far ahead". Again, I've pointed these factors out to you in a previous post.

@NVIDIATI: It's not just having the hardware ready, it's having the software ready, too. iOS and the apps fully support the 64bit A7. How far away from a fully 64bit mobile OS are we in terms of the other major players? How long until we see an android phone release with the required 64bit hardware/software combo?

How quickly we forget how Apple has launched nearly every major consumer tech revolution since the original iPod 13 years ago -- or has trampled the competition in nearly each scenario. It's very convenient to forget that Apple has been leading for more than a decade.

EDITS: Added "consumer", comments about the relationship between 64bit hardware and software

Apple has definitely been way ahead of the game. I just went from my new 4S to a 5S, and this fingerprint scanner is something I haven't seen anyone else able to even remotely imitate it or come close. The other fingerprint scanners work so horribly that I believed these things were just gimmicks up until I used Touch ID. Still no one out there in sight with a solution as elegant and easy to use.

We're discussing mobile SoC's in a thread about a new mobile SoC, so I don't see the relevance of Apple's previous market success.

If you're curious about other 64-bit software, Android L (with ART) is coming this fall and will offer support for x86-64, ARM64 and MIPS64. According to Google (June 2014), ~85% of current Play Store apps are immediately ready to switch over to 64-bit. This means that only ~15% of apps need some sort of recompiling to make use of 64-bit.

Qualcomm's ARM64 Snapdragon 410 (4x Cortex A53) and Snapdragon 610/615 (4x/8x Cortex A53) will come in phones in Q3 and Q4 of this year. I don't know if Tegra K1 with Denver will make it into a phone. I'm not sure which other vendors will have Cortex A53 or Cortex A57 SoCs out before the end of 2014 (Samsung?, Mediatek?). Qualcomm has been spotted testing their flagship Snapdragon 810 (4x Cortex A57, 4x Cortex A53, Adreno 430, 20 nm process, LPDDR4 RAM, CAT. 6 LTE-A world modem), but it should only come to market in early 2015.

Any phone that uses Intel's x86-64 Silvermont cores (Merrifield or Moorefield SoC), or a future SoC with Airmont / Goldmont cores, will have 64-bit support with Android L.

Imagination Technologies should have a MIPS64 SoC in 2015, which is interesting as both Intel and Apple are part owners (5% and 8.6%).

@NVIDIATI: So the short answer is we don't actually know when a non-Apple phone will be fully 64 bit? Maybe early 2015?

And Apple's decade long leadership is relevant because they've introduced the most meaningful advances in consumer technology and the culture of consumer technology since the original iPod. It's convenient to forget how much Apple has impacted the lives of everyone who enjoys phones and other mobile devices, computers, music, games, travel, business, and so much more.

To accept that reality is to understand on a more fundamental level how Apple decides what it believes is and isn't important, right down to the choice to go 64bit while everyone else scrambles to catch up. To lead by example is the most powerful way to lead, and the industry's reaction to Apple's decision to go 64bit absolutely indicates their position as the leader of the brave new mobile world.

It helps if you actually read my post:

"Android L (with ART) is coming this fall and will offer support for x86-64, ARM64 and MIPS64."

"Qualcomm's ARM64 Snapdragon 410 (4x Cortex A53) and Snapdragon 610/615 (4x/8x Cortex A53) will come in phones in Q3 and Q4 of this year."

"Any phone that uses Intel's x86-64 Silvermont cores (Merrifield or Moorefield SoC), or a future SoC with Airmont / Goldmont cores, will have 64-bit support with Android L."

Again, this is about SoC development, and Apple's previous market success has no relevance on that. Singing a story about Apple's leadership, bravery, heroics, what ever rhetoric you want to call it, does not make their SoC development any more advanced.

I don't think you understand, in this case we already knew about NVIDIA's Denver core and the launch timeline long before Apple announced their ARMv8 core. As far as NVIDIA is concerned, everything is still on schedule. Intel had already released their 64-bit SoC last year. So don't make claims that everyone else is scrambling to catch up, because that's just not true.

@NVIDIATI: If people aren't scrambling to catch up, why did the industry react the way it did? Liken it to a kick in the balls, talking about how "we'll have 64 bit phones in the market, too" and generally sweating out this surprise shift Apple put them all through. If people aren't scrambling to catch up, why is it that nobody has released or announced a phone for release that can match what Apple did now a year ago?

Is there a phone that has been announced or is it just something you assume will happen?

Looking into Samsung a little bit more, according to AnTuTu, Samsung's Exynos 5433, which is to be used in the Galaxy Note 4 and Meizu MX4 (both launching this fall), is using Cortex A57 and Cortex A53 cores in a big.LITTLE configuration. It's also coming with ARM's Mali-T760 GPU and should be using Intel's XMM 7260 modem (CAT. 6 LTE-A). Like the Exynos 5430, the 5433 should use Samsung's 20 nm HKMG process.

When a company with the vast market saturation, such as Apple, starts to put "64-bit" in their advertisements, other companies are going to react with press statements. The media was all over this at the time, but again, that didn't necessarily make the competition's 32-bit ARM SoCs any weaker or less capable.

The same reason it took Apple a full year to announce an SoC with support for OpenGL ES 3.0 after it had released in Qualcomm's Snapdragon S4 Pro. You don't need to release a 64-bit SoC or an ARMv8 SoC to match or in some cases exceed the performance of Apple's A7 (Snapdragon 801, 805, Tegra K1). NVIDIA's Kepler GPU on Tegra K1 has support for OpenGL 4.x, DirectX 12, and features such as tessellation. This is why Tegra K1 devices can play a number of PC ports utilizing full OpenGL. Assuming Apple is using a PowerVR 6XT GPU in the A8, they won't be able to match these capabilities.

Excluding NVIDIA, most of the companies had already committed to ARM's stock Cortex A57 (large core) on a 20 nm process. This was not possible when Apple released their custom ARMv8 core on a 28 nm process.

As for phones coming this year, Samsung's latest Galaxy Mega, although still unannounced, will be using Snapdragon 410 which uses 4x Cortex A53 cores. The phone in ASUS' Transformer Book V is using Intel's Moorefield SoC and was announced at Computex 2014. Assuming the Antutu data posted in the post above is correct, then both the Galaxy Note 4 and MX4 will be using an ARM64 SoC. Most devices will only be announced or made official at, or around the time of, IFA (early September).

@NVIDIATI: I suppose so, early September is an exciting time for our beloved industry. I think we may have the most exciting one on record to look forward to this year.