NVIDIA Tegra Architecture

[ams]

Putting costs aside makes the comparison irrelevant. Tegra 4-based tablets and smartphones will compete in price and performance against Atom-based tablets and smartphones, not against ULX_Haswell-based tablets and ultrabooks. FWIW, even at 22nm fabrication process, ULX_Haswell will be limited to dual CPU cores (four threads) and will need to run at reduced clock operating frequencies in order to fit a 10w power envelope.

 

[french toast]

Putting costs aside makes the comparison irrelevant. Tegra 4-based tablets and smartphones will compete in price and performance against Atom-based tablets and smartphones, not against ULX_Haswell-based tablets and ultrabooks. FWIW, even at 22nm fabrication process, ULX_Haswell will be limited to dual CPU cores (four threads) and will need to run at reduced clock operating frequencies in order to fit a 10w power envelope.

It doesnt make ot completely irrelevant. ..they will both likely make it into tablets..even of the intel one retail for more.

You could swap that for an amd alternative then...with a 10w tdp you could get a better fusion alternative I would suggest.

 

[ams]

Well, I think for a consumer, cost of the end user device is critical, irrespective of shared or similar form factors. Otherwise, what would stop a consumer from purchasing a Geforce Titan rather than a 7970 GHz Edition, even though they both have a maximum TDP of ~ 250w?

With respect to AMD, the situation is somewhat similar to Intel, but they don't have anything like Atom (or Tegra) to reach smartphones and very low cost tablets. Most of AMD's tablets will use dual Jaguar CPU cores running at a relatively low CPU clock operating frequency, while the quad Jaguar CPU core variants will be reserved more for higher cost convertible tablets.

 

[Alexko]

Reference tablet also carried high end sandisk nand flash storage and ddr3L 1866 to go along with a 9w tdp.

Forget about seeing nvidias marketing benchmarks inside any smartphone without handling it with a pair of oven gloves.


Edit..question. .putting costs aside and looking at purely performance metrics...why would you pick a full power tegra 4 in your tablet over say a 10w Haswell??..

As ams pointed out, those don't cost nearly the same. But Tegra 4 vs. Kabini/Temash, now that's a fight I can't wait to see.

 

[french toast]

As ams pointed out, those don't cost nearly the same. But Tegra 4 vs. Kabini/Temash, now that's a fight I can't wait to see.

Yes intels haswell will be out of reach in terms of cost...but still I asked a geniune and interesting question. ..how much more powerfull would a 10w Haswell be?

Temash in A6-5200 flavour would be a direct competitor to tegra 4...5.9w tdp in tablet...then turbo dock will allow 15w tdp when docked to key board ala asus transformer style.

 

[ams]

Temash-based [Windows 8] products will use a completely different operating system than Tegra 4-based [Android, Windows RT, Windows Phone] products, so I'm not sure that they would be direct competitors per se, but they will certainly compete with each other in general in the tablet marketplace.

 

[Alexko]

Yes intels haswell will be out of reach in terms of cost...but still I asked a geniune and interesting question. ..how much more powerfull would a 10w Haswell be?

With 1~2 threads, probably a lot. With 4 threads, maybe not that much, but it's hard to say without having Haswell's specs.

 

[french toast]

With 1~2 threads, probably a lot. With 4 threads, maybe not that much, but it's hard to say without having Haswell's specs.

If we take ivy bridge then...dont forget hyper threads. ..

 

[Alexko]

If we take ivy bridge then...dont forget hyper threads. ..

The closest thing is this: http://ark.intel.com/products/72015/Intel-Core-i7-3689Y-Processor-4M-Cache-up-to-2_60-GHz

2 cores, 4 threads, 1.5GHz base, 2.6GHz Turbo, 13W TDP.

It shouldn't even be a contest with 1~2 threads.

With 4 threads, assuming Ivy's performance per clock is twice that of Cortex A15, we get (4 ARM cores × 1.9GHz) / (2 Ivy cores × 2 IPC advantage ×1.5GHz) = 7.6/6, or a 27% advantage for Tegra 4, with 30% more power for Ivy Bridge.

That is, of course, a very very very rough estimate that doesn't take the memory performance under consideration, that assumes Ivy Bridge cannot Turbo at all when all cores are active, and of course the performance/clock estimate is very rough too.

All in all I'd be inclined to give Ivy Bridge a small performance advantage with all cores active, but with a roughly commensurate power consumption. Haswell should improve upon both performance and power by a little bit. But this is little more than guesswork.

 

[french toast]

The closest thing is this: http://ark.intel.com/products/72015/Intel-Core-i7-3689Y-Processor-4M-Cache-up-to-2_60-GHz

2 cores, 4 threads, 1.5GHz base, 2.6GHz Turbo, 13W TDP.

It shouldn't even be a contest with 1~2 threads.

With 4 threads, assuming Ivy's performance per clock is twice that of Cortex A15, we get (4 ARM cores × 1.9GHz) / (2 Ivy cores × 2 IPC advantage ×1.5GHz) = 7.6/6, or a 27% advantage for Tegra 4, with 30% more power for Ivy Bridge.

That is, of course, a very very very rough estimate that doesn't take the memory performance under consideration, that assumes Ivy Bridge cannot Turbo at all when all cores are active, and of course the performance/clock estimate is very rough too.

All in all I'd be inclined to give Ivy Bridge a small performance advantage with all cores active, but with a roughly commensurate power consumption. Haswell should improve upon both performance and power by a little bit. But this is little more than guesswork.

Oh cheers yea thats pretty much what I thought..except x86 winning by more.
Arm and x86 are not too far off one another now.

 

[Alexko]

Oh cheers yea thats pretty much what I thought..except x86 winning by more.
Arm and x86 are not too far off one another now.

Well, Intel's memory controller and all-core Turbo could give Ivy/Haswell significantly more performance than my probably conservative estimate.

But Tegra 4 (or any ARM SoC, for that matter) versus Ivy/Haswell is a bit apples to oranges: Intel's CPU are manufactured on a much better process, with a lot of hand-tuning, and a big, expensive die. They also have very different optimal power ranges, so the winner will depend a lot on the power target you choose.

I really don't think Haswell could come anywhere near Snapdragon S600 within 3W, for instance.

ARM SoCs versus Kabini/Temash would be a better comparison: small, cheap, relatively synthesizable cores all made on the same (or almost the same) 28nm TSMC process. The different ISAs will still make benchmarking difficult, and the power targets are a bit different too, but the results are still guaranteed to be very interesting. And then there's Atom, of course.

 

[Turbotab]

Oh cheers yea thats pretty much what I thought..except x86 winning by more.
Arm and x86 are not too far off one another now.

I believe there is still a huge performance gap, look what an older 32nm Sandy Bridge based Celeron 847, with no HT & 2 cores downclocked to 800 MHz does to a dual-core A15 @ 1.7 GHz.

http://www.anandtech.com/show/6476/acer-c7-chromebook-review/3

 

[Turbotab]

Well, Intel's memory controller and all-core Turbo could give Ivy/Haswell significantly more performance than my probably conservative estimate.

But Tegra 4 (or any ARM SoC, for that matter) versus Ivy/Haswell is a bit apples to oranges: Intel's CPU are manufactured on a much better process, with a lot of hand-tuning, and a big, expensive die. They also have very different optimal power ranges, so the winner will depend a lot on the power target you choose.

I really don't think Haswell could come anywhere near Snapdragon S600 within 3W, for instance.

ARM SoCs versus Kabini/Temash would be a better comparison: small, cheap, relatively synthesizable cores all made on the same (or almost the same) 28nm TSMC process. The different ISAs will still make benchmarking difficult, and the power targets are a bit different too, but the results are still guaranteed to be very interesting. And then there's Atom, of course.

Haswell doesn't have to match Snapdragon 600 on power usage, Clovertrail + @32 nm is already competitive and Bay Trail is designed to fight that battle.

 

[Alexko]

Haswell doesn't have to match Snapdragon 600 on power usage, Clovertrail + @32 nm is already competitive and Bay Trail is designed to fight that battle.

I wasn't saying that it needs to match Snapdragon on power usage, merely that the result of an Ivy Bridge / Snapdragon duel would depend a lot on the targeted power envelope. They're obviously very different designs with different power targets, but there is some overlap. At the lower end of this overlap, Snapdragon should win, and at the upper end, Ivy should.

As for Clover Trail+ being competitive, I haven't seen any reviews, and I'm a little skeptical, but perhaps. I'm sure we'll know soon enough. Bay Trail looks promising, but if I'm not mistaken, it's not expected until Q1'14, which means it will have to compete with Snapdragon 800.

 

[Laurent06]

The closest thing is this: http://ark.intel.com/products/72015/Intel-Core-i7-3689Y-Processor-4M-Cache-up-to-2_60-GHz

I had never noticed the price of this chip, I guess heavy binning has to be done.

Wouldn't the i5-3439Y, or i5-3339Y be a better match for your comparison? Or even i3-3229Y or the castrated Pentium 2129Y?

Anyway it's likely all of these chips except perhaps the Pentium, will beat A15 @1.7GHz for single threaded tasks. Their prices are very high though and I wonder if Intel has been able to tune its 22nm process to get more low power Haswell dies.

 

[french toast]

I wasn't saying that it needs to match Snapdragon on power usage, merely that the result of an Ivy Bridge / Snapdragon duel would depend a lot on the targeted power envelope. They're obviously very different designs with different power targets, but there is some overlap. At the lower end of this overlap, Snapdragon should win, and at the upper end, Ivy should.

As for Clover Trail+ being competitive, I haven't seen any reviews, and I'm a little skeptical, but perhaps. I'm sure we'll know soon enough. Bay Trail looks promising, but if I'm not mistaken, it's not expected until Q1'14, which means it will have to compete with Snapdragon 800.

Yes intel chips are manufactured at a far higher cost so they are not not direct comparison...if you used intels technologies for the cortex A15s that would be interesting to see how much more performance could be squeezed from it.

I too would like to see some benchmarks before judging which one would be better, I would be very suprised if Snapdragon s600 didnt win that battle.

 

[french toast]

I believe there is still a huge performance gap, look what an older 32nm Sandy Bridge based Celeron 847, with no HT & 2 cores downclocked to 800 MHz does to a dual-core A15 @ 1.7 GHz.

http://www.anandtech.com/show/6476/acer-c7-chromebook-review/3

Yea I see what you mean..and that was sandybridge clocked at half the frequency! ..although anand does make a brieft mention of maybe the software being a bottle neck in certain scenarios.

 

[Alexko]

I had never noticed the price of this chip, I guess heavy binning has to be done.

Wouldn't the i5-3439Y, or i5-3339Y be a better match for your comparison? Or even i3-3229Y or the castrated Pentium 2129Y?

Anyway it's likely all of these chips except perhaps the Pentium, will beat A15 @1.7GHz for single threaded tasks. Their prices are very high though and I wonder if Intel has been able to tune its 22nm process to get more low power Haswell dies.

Oh, I didn't know there were so many Y chips, I thought the i7 was the only one. Yes, price-wise, the Pentium is the best comparison, and still far more expensive than any ARM SoC.

I guess right now this Pentium might be appealing because it's x86, and that fills a need, but once Kabini comes out, I can't think of a good reason to buy that for anyone.

 

[Blazkowicz]

I think this Pentium is nice, very similar to the Celeron 847 but with a TDP seven watts lower.
Note that the Celeron has a "recommended TRAY price" of $134, whatever that means ; and you can buy it on a whole ITX motherboard for a whole less than that, taxes included, buying one not one thousand.

 

[ams]

I believe there is still a huge performance gap, look what an older 32nm Sandy Bridge based Celeron 847, with no HT & 2 cores downclocked to 800 MHz does to a dual-core A15 @ 1.7 GHz.

http://www.anandtech.com/show/6476/acer-c7-chromebook-review/3

The dual core Sandy Bridge Celeron 847 fabricated on a 32nm fabrication process has a max TDP of ~ 17w (when running at 1.1GHz CPU clock operating frequency). So even with a shrink to a 22nm fabrication process and even with a newer architecture, there may not be too much room to increase performance beyond this since the max TDP of the 22nm ULX_Haswell part is much lower at ~ 10w.

Comparing the dual core Sandy Bridge Celeron 847 (operating at 800MHz) in the Acer C7 Chromebook to the dual core Cortex A15 (operating at 1.7GHz) in the Samsung Chromebook XE303 (with both CPU's fabricated on a 32nm fabrication process at Intel and Samsung fabs, respectively) the Sandy Bridge Celeron 847 processor has a ~ 0% performance advantage in RIABench Focus Tests, a ~ 10% performance advantage in BrowserMark, and a ~ 30% performance advantage in SunSpider 0.9.1 and Kraken, but a ~ 50% disadvantage in the web browsing battery life test: http://www.anandtech.com/show/6476/acer-c7-chromebook-review/5 (and this battery life metric is even more lopsided when you consider the fact that the dual core Cortex A15-equipped Chromebook has ~ 20% less battery capacity than the dual core Sandy Bridge Celeron 847-equipped Chromebook!). So even though the dual core Sandy Bridge Celeron 847 CPU clearly has higher IPC (instructions per clock), the dual Cortex A15 CPU appears to have significantly better performance per watt and significantly lower power consumption than the dual core Sandy Bridge Celeron 847 CPU, even when the latter is operating at only 800MHz.