NVIDIA's Project Kal-El

[nyker96]

giving the x86s some competition in mobile market:

looks like NV and ARM is getting serious about entering the APU race:
http://www.anandtech.com/show/4181/...re-a9s-coming-to-smartphonestablets-this-year

 

[notty22]

Anand is very excited, I am to. Wish I had money to invest in Nvidia stock, its up 50% since 2011 and IMHO its going to keep going up. 6 month product cycles with these chips, quad core by the end of the year for smart phones. Demo'd playing blu ray quality to dual large resolution screen ? WoW

 

[tweakboy]

Wow thats crazy!

 

[LoneNinja]

Honestly I first laughed when I found out about Nvidia making ARM based processors, but I'm starting to think they may have something here.

 

[Idontcare]

Honestly I first laughed when I found out about Nvidia making ARM based processors, but I'm starting to think they may have something here.

If they fail it won't be for lack of effort. They are being as aggressive about this marketspace as JHH was when it came to the graphics market.

Any bets on who will be the 3Dfx, matrox, etc analogues? There are a lot of incumbents in this space, but the same was said of graphics 10yrs ago as well.

 

[Cogman]

Certainly is impressive from a performance standpoint. The real question, however, is what sort of power consumption are we talking about. Performance is great, but if the power consumption is too high then it is a lemon for the applications it is aiming at.

“It is not possible [that Atom Z600 will be competitive]. You could give an elephant a diet but it’s still an elephant. And when they think about power they think reducing from 20W down to 5W down to 4W down to 2W is really good. But you and I both know that in a mobile phone you need to be in a 100mW – 200mW. So they are still ten times away,” said Mr. Huang.

Color me extremely impressed if they are doing this at 100 - 200 mW. Heck, even if this is under 5W, we are looking at something that would make an excellent HTPC and netbook CPU. (operating system dependent)

 

[SlitheryDee]

100x faster than tegra 2 in 4 years... That's insane. They're already saying that kal-el will be in the neighborhood of a core2duo. At this pace won't they be closing the performance gap between mobile nvidia processors an desktop x86 processors even given the probable advances from amd and intel over the next 4 years?

 

[Cogman]

100x faster than tegra 2 in 4 years... That's insane. They're already saying that kal-el will be in the neighborhood of a core2duo. At this pace won't they be closing the performance gap between mobile nvidia processors an desktop x86 processors even given the probable advances from amd and intel over the next 4 years?

Well.... I wouldn't take that as the gospel truth on ACTUAL performance. Remember, nVidia likes to include things like GPU performance into their equations when they are talking about total system performance. It is quite possible that the ARM CPUs they couple onto this thing will be faster than intel's offering, but I have a sneaking suspicion that those numbers are derived from FLOPS of the total APU.

From a programmers standpoint, it is still pretty dang tough to get a good functioning GPGPU program going.

 

[Mopetar]

Certainly is impressive from a performance standpoint. The real question, however, is what sort of power consumption are we talking about. Performance is great, but if the power consumption is too high then it is a lemon for the applications it is aiming at.

I imagine this will be more aimed at netbooks and HTPCs, possibly tablets. A phone doesn't have a whole lot of use for four ARM cores right now. Most of the time they'll stay powered down while the dedicated hardware decoders on the SoC get most of the workout.

I'm a little suspect of their future performance claims as well. Even so, the rest of the industry has been able to keep pace with Nvidia. TI's new OMAP has it beat in a few benchmarks and Samsung's Exynos is almost even with it. Will be interesting to see how Qualcomm's next batch of SoCs perform.

 

[AtenRa]

Color me extremely impressed if they are doing this at 100 - 200 mW. Heck, even if this is under 5W, we are looking at something that would make an excellent HTPC and netbook CPU. (operating system dependent)

If im not mistaken, Tegra 2 Dual core ARM A9 CPU is at 150 to 200mW at full load so i will guess that a quad core will be double at 300-400mW.

What made a bigger impression for me is the Die Size of Kal-El at 80mm2 and that is the same as AMDs Bobcat Brazos Dual core APU at 40nm TSMC.

Two different micro architectures at the same 40nm manufacturing process in the same foundry with completely different technical characteristics and they end up at the same die size.

 

[Ajay]

If im not mistaken, Tegra 2 Dual core ARM A9 CPU is at 150 to 200mW at full load so i will guess that a quad core will be double at 300-400mW. .

That'll work out for high end phones (especially ones with a large screen and room for a larger battery). I read somewhere that < 500mW is necessary to get design wins for mobile phones. I imagine that OS will throttle down power use as much as possible. For a tablet design - this is fantastic news.

 

[drizek]

If im not mistaken, Tegra 2 Dual core ARM A9 CPU is at 150 to 200mW at full load so i will guess that a quad core will be double at 300-400mW.

What made a bigger impression for me is the Die Size of Kal-El at 80mm2 and that is the same as AMDs Bobcat Brazos Dual core APU at 40nm TSMC.

Two different micro architectures at the same 40nm manufacturing process in the same foundry with completely different technical characteristics and they end up at the same die size.

I would love to see some linux benchmarks comparing the two. I can't really think of anywhere where tegra can win though. Bobcat will handle memory better, have much better 3D and video acceleration. Tegra 3 will need some pretty serious clockspeeds in order to stay competitive, and that probably wont happen until 28nm.

Obviously they can't be compared directly, bobcat doesn't even come close to fitting in a smartphone since it is not an SOC. I think in 10" tablets there can be some overlap between the two though. Nobody is interested in a tablet running windows xp tablet edition, but a Honeycomb tablet that can competently dual boot windows 7 might find a niche.

 

[Idontcare]

If im not mistaken, Tegra 2 Dual core ARM A9 CPU is at 150 to 200mW at full load so i will guess that a quad core will be double at 300-400mW.

What made a bigger impression for me is the Die Size of Kal-El at 80mm2 and that is the same as AMDs Bobcat Brazos Dual core APU at 40nm TSMC.

Two different micro architectures at the same 40nm manufacturing process in the same foundry with completely different technical characteristics and they end up at the same die size.

You'll notice this happens in consumer CPU's as well and it happens for a fundamental reason - design targets.

Die size relates to a targeted manufacturing cost. X dies per wafer sort of targets.

 

[cbn]

If they fail it won't be for lack of effort. They are being as aggressive about this marketspace as JHH was when it came to the graphics market.

Any bets on who will be the 3Dfx, matrox, etc analogues? There are a lot of incumbents in this space, but the same was said of graphics 10yrs ago as well.

I don't think Nvidia will lose, but I wonder when they will announce their own Cortex A15 version of Qualcomm APQ8064 and OMAP 5?

At this time it appears they would rather spending time designing quad core Cortex A9 when the rest of the industry has moved on and is proceeding with A15?

Or maybe Project Denver is their plan for Cortex A15, but with a giant GPU instead of something typical?

It will be interesting to see how the Nvidia roadmap and its timing turns out?

 

[khon]

I don't believe this part at all. Not even a little bit.

They aren't going to have 100x the performance in 4 years, that's simply not happening. And they're not going to get performance well above that of a C2D while maintaining smartphone compatible power consumption, at least not as soon as this year.

 

[Idontcare]

I don't believe this part at all. Not even a little bit.

They aren't going to have 100x the performance in 4 years, that's simply not happening. And they're not going to get performance well above that of a C2D while maintaining smartphone compatible power consumption, at least not as soon as this year.

Considering that A15 scales to 32 cores out-of-the-box if a company were so inclined to build such a behemoth chip, I don't see why you think a 100x increase is feasible.

As already noted the real question, dare I say the ONLY question, that matters is what does the performance/watt graph look like?

If it scales anywhere close to the published performance graph then the chips are non-starters. If Nvidia aims to increase performance 100X while keeping power-consumption at current levels then I'd agree it is total bullocks.

The truth will naturally be somewhere in the middle there. The question we have every right to ask is "where" in the middle does it fall?

If Stark consumes >800mW then it is a non-starter. QED.

 

[cbn]

I don't believe this part at all. Not even a little bit.

They aren't going to have 100x the performance in 4 years, that's simply not happening. And they're not going to get performance well above that of a C2D while maintaining smartphone compatible power consumption, at least not as soon as this year.

I'll bet the Stark claim (with Core 2 duo CPU performance) corresponds to how they think ARM 64 bit will turn out.

As far as the 5x performance jumps go (and the 2x performance jump between logan and stark) I believe those refer to CPU and GPU added together. (with power gating would this kind of increase be possible? Multiple CPU and GPU cores able to be shut down during idle conditions so as to not kill battery.)

 

[cbn]

If they fail it won't be for lack of effort. They are being as aggressive about this marketspace as JHH was when it came to the graphics market.

Any bets on who will be the 3Dfx, matrox, etc analogues? There are a lot of incumbents in this space, but the same was said of graphics 10yrs ago as well.

I'm mostly worried about AMD.

They are too small of a company to miss the boat on something like this.

 

[AtenRa]

First have a look at this video

http://www.youtube.com/watch?v=VATNyBTPltI&feature=related

They test Kal-El with coremark benchmark.

From www.coremark.org

http://www.coremark.org/faq/index.php?pg=faq

Will CoreMark work for analyzing multicore performance?

You can run multiple instantiations of CoreMark (a rate-type benchmark) and the source code includes a flag that will allow you to compile in code that supports a ‘create and destroy’ context functionality (where a context can be either threads or processes with shared memory or with sockets). Regardless, it’s expected that adding more cores will yield an almost linear performance behavior. For analyzing multicore performance, we highly recommend that you use the EEMBC MultiBench software.

what do you guys think ??

 

[BD2003]

Im starting to finally see how we get to the point I keep hearing about where your smartphone is your primary computer.

A Tegra 2 having half the performance of a C2D is wishful thinking. A 100x performance increase within 4 years seems like wishful thinking as well.

But this isnt an also ran like VIA or S3 saying it, its NVIDIA. When they say it, its actually worth listening to. So even if its wishful thinking, they clearly plan on bringing it one way or another.

Besides just being excited about what NVIDIA can do, it seems like competition is really going to heat up in the SoC space. Sounds like this is where all the interesting things are going to happen in the consumer market for the next couple years. New top of the line desktop chips from intel are great every year, but I find that as each day goes on, I use my smartphone and tablet more and more, and my desktop less and less.

 

[Ajay]

I'm mostly worried about AMD.
They are too small of a company to miss the boat on something like this.

Yes, and that is probably why Dirk Meyer resigned.

 

[Dribble]

Personally I think Nvidia are a bit of a wolf amongst the chickens in the mobile soc market.

Nvidia are used to fighting with Intel, AMD and here they have Qualcomm? Sure Apple and Samsung make soc's but that's not their primary business, they'll quite happily use other companies chips if that suited them better.

JHH must be over the moon - he's got his company into position and pushing hard in the big new arena for personal computing and Intel + AMD are no-shows. Nvidia are likely to win big here.

 

[cbn]

Nvidia are used to fighting with Intel, AMD and here they have Qualcomm?

Isn't Qualcomm approaching the size of Intel?

 

[Idontcare]

Personally I think Nvidia are a bit of a wolf amongst the chickens in the mobile soc market.

Nvidia are used to fighting with Intel, AMD and here they have Qualcomm? Sure Apple and Samsung make soc's but that's not their primary business, they'll quite happily use other companies chips if that suited them better.

JHH must be over the moon - he's got his company into position and pushing hard in the big new arena for personal computing and Intel + AMD are no-shows. Nvidia are likely to win big here.

http://www.arm.com/products/processors/licensees.php

A15 Licensees: Texas Instruments, ST-Ericsson, nVIDIA, Samsung Electronics

A9 Licensees: Broadcom Corporation, NEC Electronics, nVIDIA, STMicroelectronics, Texas Instruments, Toshiba, Mindspeed Technologies

A8 Licensees: Broadcom Corporation, Freescale Semiconductor, Matsushita, Samsung Electronics, STMicroelectronics, Texas Instruments, PMC-Sierra, 3Dlabs

These guys are experienced ARM licensees, Nvidia is a nobody to their customers in the working relationship environment.

I don't think the likes of TI, Samsung, Qualcomm are just going to roll over and sit idly by while Nvidia "wins big".

There is a reason Intel and AMD have stuck with their core competency.

Nvidia may do well in ARM, but their victory is not at all assured. This industry is all about customer/supplier relations (and by customer I don't mean us, Nokia is a customer of ARM processors) and they take years if not decades to build.

Nvidia could have the best thing since the pentium processor and it won't see a retail shelf if they can't get socket wins. Intel learned this in 2005, they are kinda learning much hasn't changed since then too.

And the reason for this is fairly straightforward too, with ARM being so easily accessible and the fact that most of these guys are fabless results in them all using TSMC as their foundry so process tech is not a differentiator means they fight tooth and nail to lock-in customers (sockets) as they've got little else to "add value" to the supply chain.

Nokia isn't looking for the wonder chip that will vault them to the top of the smartphone market for 3-6 months, they are looking for the supplier with a proven track record of not dropping the ball and leaving their customer chip-less or 6 months behind schedule with their 4G rollout.

Consistency wins the day in this industry, you can't be fickle, all gung-ho one year and nowhere to be seen the next. This has been Intel's undoing, it remains to be seen if Nvidia is really committed to this market or not and you can bet many customers are intentionally sitting on the sidelines waiting to see as well. (and Nvidia knows this too, hence all the roadmappery and open-ness about their aspirations for the next 4 yrs)

 

[OCGuy]

Well IDC supposedly nV already has some signifigant contracts....