nvidia tegra K1

[IntelUser2000]

I think that Intel's 14nm Cherry Trail won't be worried by a 28nm part...

The specifications shown by Nvidia with K1 already equals Intel's 20EU GT2 graphics HD 4400 on the Ultrabooks today. See, the successor isn't coming with Broadwell until late this year. That means, a chip that supposedly is in the "Atom range" has graphics performance of a much higher power Core chip.

Now, Intel's chips look really good only compared to AMD. Against ARM, they seem mediocre.

Unless Broadwell shows 3x performance increase on its iGPU(without using a version that needs a eDRAM or one that costs $90 extra), it may well be in 2015 that mobile ARM graphics catch up to the ones in Core chips.

Who cares if Intel is at "14nm" when it needs that to barely beat 28nm chips?

 

[Sweepr]

The specifications shown by Nvidia with K1 already equals Intel's 20EU GT2 graphics HD 4400 on the Ultrabooks today. See, the successor isn't coming with Broadwell until late this year. That means, a chip that supposedly is in the "Atom range" has graphics performance of a much higher power Core chip.

Now, Intel's chips look really good only compared to AMD. Against ARM, they seem mediocre.

Unless Broadwell shows 3x performance increase on its iGPU(without using a version that needs a eDRAM or one that costs $90 extra), it may well be in 2015 that mobile ARM graphics catch up to the ones in Core chips.

Who cares if Intel is at "14nm" when it needs that to barely beat 28nm chips?

They seems to be paying a lot of attention to iGPU performance as Cherry Trail will have 16 EUs based on a new (possibly radically different) Gen 8 architecture.
192 CUDA cores is half a GK108, depending on final clockspeeds I dont see them miles ahead of Intel or other late 2014 28/20nm ARM competitors.
Still, very impressive specs for a tablet chip and I hope NVIDIA manages more and better design wins than Tegra 4 had this time (perhaps a Nexus 10?).

 

[Ajay]

The specifications shown by Nvidia with K1 already equals Intel's 20EU GT2 graphics HD 4400 on the Ultrabooks today. See, the successor isn't coming with Broadwell until late this year. That means, a chip that supposedly is in the "Atom range" has graphics performance of a much higher power Core chip.

Now, Intel's chips look really good only compared to AMD. Against ARM, they seem mediocre.

Unless Broadwell shows 3x performance increase on its iGPU(without using a version that needs a eDRAM or one that costs $90 extra), it may well be in 2015 that mobile ARM graphics catch up to the ones in Core chips.

Who cares if Intel is at "14nm" when it needs that to barely beat 28nm chips?

Good point, especially compared to K1v2 (which will have higher CPU performance than v1). 'May you live in interesting times' looks to be coming true for Intel. It will be interesting to see if NV can win over any significant vendors on this go-around.

 

[liahos1]

The specifications shown by Nvidia with K1 already equals Intel's 20EU GT2 graphics HD 4400 on the Ultrabooks today. See, the successor isn't coming with Broadwell until late this year. That means, a chip that supposedly is in the "Atom range" has graphics performance of a much higher power Core chip.

Now, Intel's chips look really good only compared to AMD. Against ARM, they seem mediocre.

Unless Broadwell shows 3x performance increase on its iGPU(without using a version that needs a eDRAM or one that costs $90 extra), it may well be in 2015 that mobile ARM graphics catch up to the ones in Core chips.

Who cares if Intel is at "14nm" when it needs that to barely beat 28nm chips?

lets wait to see how these parts actually perform before coming to final conclusions here. jen hsun talks a massive game but has a less than stellar track record.

but honestly i get what you're saying. given the massive mfg lead intel has, you'd think they'd be able to build something out of the league of what the arm camp can build. silvermont cpu is stellar but gpu wasnt amazing. they need to step it up if they are going to get any real share on the phone side. tablet side i think they have a good opp.

 

[sontin]

That is one of Intel's biggest problem: If the ARM market envolves at the same pace they need to bring their fastest iGPU product down to the low-end.

CPU performances is a no matter. Custom ARMv8 cores will close the cap between now and midrange Intel CPUs like Haswell.

 

[NTMBK]

That is one of Intel's biggest problem: If the ARM market envolves at the same pace they need to bring their fastest iGPU product down to the low-end.

CPU performances is a no matter. Custom ARMv8 cores will close the cap between now and midrange Intel CPUs like Haswell.

If they close the gap in performance, they will also close the gap in power consumption. ARM has no special magic that lets it consume notably less power, and Intel has the superior fab on their side.

ARM chips are trying to encroach on a laptop and desktop market which has a deeply entrenched ecosystem based on native x86 binaries, whereas Intel are trying to break into a tablet and smartphone market running on a Java VM... who has the more difficult task here?

 

[IntelUser2000]

ARM chips are trying to encroach on a laptop and desktop market which has a deeply entrenched ecosystem based on native x86 binaries, whereas Intel are trying to break into a tablet and smartphone market running on a Java VM... who has the more difficult task here?

They don't need to... they just need to get the PC market chugging along for a decade or two and die off gradually while they focus on Android and iOS.

Eventually enough alternatives to Office and dwindling support on "full" application would mean the whole world would move to the mobile ecosystem.

There's LOT more for Intel to lose. If the ARM CPUs come to Core CPUs in per core performance that means Intel would need to sell previously $100 ASP CPUs for $20.

ARM has no special magic that lets it consume notably less power, and Intel has the superior fab on their side.

Broadwell brings Core to 4.5W TDP. There's no reason ARM can't do it in couple of years. Also we already established the latter is absolutely irrelevant. The fab advantage is all hypothetical.

 

[liahos1]

They don't need to... they just need to get the PC market chugging along for a decade or two and die off gradually while they focus on Android and iOS.

Eventually enough alternatives to Office and dwindling support on "full" application would mean the whole world would move to the mobile ecosystem.

There's LOT more for Intel to lose. If the ARM CPUs come to Core CPUs in per core performance that means Intel would need to sell previously $100 ASP CPUs for $20.



Broadwell brings Core to 4.5W TDP. There's no reason ARM can't do it in couple of years. Also we already established the latter is absolutely irrelevant. The fab advantage is all hypothetical.

i dont get this? baytrail's cpu is significantly faster than anything arm based less a7. its perf/watt in significantly better including a7s. I thought you were griping about gpu performance.

 

[IntelUser2000]

They seems to be paying a lot of attention to iGPU performance as Cherry Trail will have 16 EUs based on a new (possibly radically different) Gen 8 architecture.

Let's take the most optimistic projections of the Gen 8 GPU, based on CPU-World rumors.

Broadwell K is either 80% faster than GT3 Haswell, or GT2 Haswell in the K chips. Let's assume former is true.

That means, GT3 to GT3, Gen 8 is 80% faster. That means GT2 Broadwell, that is, the version with 24EU, will be 80% faster than 20EU GT2 Haswell. Extrapolate that out folks, how GT2 Broadwell with only 16 EUs will perform.

GTn HSW to GTn BRD = 1.8x
GT2 Haswell(aka HD 4400) to GT2 Broadwell(24EU) = 1.8x
GT2 Haswell to Broadwell 16 EU = 1.2x?
Tegra 5 = GT2 Haswell

At best, Cherry Trail's 16EU will be 20-30% faster than Tegra 5. This is the BEST Intel can show, with 2 FULL generation process advantage. Now the situation is a lot worse if the original extrapolation has to be based off GT3 Broadwell being 80% faster than GT2 Haswell.

i dont get this? baytrail's cpu is significantly faster than anything arm based less a7. its perf/watt in significantly better including a7s. I thought you were griping about gpu performance.

Intel cannot pull out anything other than measly 5-10% improvement with their Core line of chips. Apple chips are pretty close already, based on Denver, other ARM vendors aren't far behind.

 

[mikk]

Tegra K1 is 4x faster in GFXBench than a Bay Trail-T SoC with 4EU Gen7. Cherry Trail should match it with 4x the EUs and a major GPU Overhaul.

 

[liahos1]

Tegra K1 is 4x faster in GFXBench than a Bay Trail-T SoC with 4EU Gen7. Cherry Trail should match it with 4x the EUs and a major GPU Overhaul.

right. at a much better cost, and pef/watt advantage as well.

 

[Sweepr]

right. at a much better cost, and pef/watt advantage as well.

+The hability to put its improved iGPU performance to use running traditional Windows (x86) games, instead of being limited to whats available on the Google Play Store. Lots of Bay Trail tablets, 2-in-1s and convertibles (Windows, Android and dual-OS) are showing the potential of the new cheap low-power x86 chips.

 

[bullzz]

@mikk - how did u come up with the 4x number. apple A7 is 3x compared to BT in GFXBench. i think its more than 4x

 

[Sweepr]

@mikk - how did u come up with the 4x number. apple A7 is 3x compared to BT in GFXBench. i think its more than 4x

 

[mikk]

@mikk - how did u come up with the 4x number. apple A7 is 3x compared to BT in GFXBench. i think its more than 4x

In a 7-inch Tegra K1 reference tablet, Nvidia claims the ability to render 60 fps in the GFXBench 2.7.5 T-Rex offscreen test. In our benchmarks, the PowerVR G6430 in the iPad Air was the fastest ARM-based SoC GPU, which achieved only 27 fps.

http://www.notebookcheck.net/NVIDIA-Tegra-K1-SoC.108310.0.html


Z3770 or Z3740 scores 15-16 fps Offscreen.

 

[Enigmoid]

The specifications shown by Nvidia with K1 already equals Intel's 20EU GT2 graphics HD 4400 on the Ultrabooks today. See, the successor isn't coming with Broadwell until late this year. That means, a chip that supposedly is in the "Atom range" has graphics performance of a much higher power Core chip.

Now, Intel's chips look really good only compared to AMD. Against ARM, they seem mediocre.

Unless Broadwell shows 3x performance increase on its iGPU(without using a version that needs a eDRAM or one that costs $90 extra), it may well be in 2015 that mobile ARM graphics catch up to the ones in Core chips.

Who cares if Intel is at "14nm" when it needs that to barely beat 28nm chips?

K1 has less bandwidth than core series CPUs and probably a worse memory controller as well.

If K1 is anything like Tegra 4 or Tegra 3 it will require significantly more power than baytrail.

I'm guessing that similar to tegra 4 the GPU can easily burn through 4+W under load. Not counting 4x A15 cores.

Battery test runs egypt HD onscreen capped at 30 fps. Do the math and the device is using about 4.8 W/hour running only the GPU at ~50% load (63 fps offscreen).

Alternatively, Asus T100 lasts 4.25 hours under brutal load (furmark + prime) at max brightness on a 10 inch screen on a 31 Whr battery. 7.3 W/hour. (notebookcheck).

Max power draw of Tegra 4 SKU is probably close to twice that of baytrail. I don't think K1 is going to drop power to any significant amount.

 

[jdubs03]

Wow that T-Rex score is outstanding. That really is phenomenal. 1.4x perf for CPU performance isn't that bad either, likely being 30% (around 1300 for ST GB3) faster than A7 for 32-bit geekbench.

I'm excited to see Denver, single-thread could be near 2000.

 

[scannall]

Hopefully, their claims will be closer to reality than they have been in the past. Looks cool and all, but I can wait for real numbers. I'm patient. ;-)

 

[NTMBK]

I am actually quite sad that NVidia didn't manage to get an x86 license out of Intel; if this were an x86 part as originally planned, it would actually be pretty interesting. It's cool to see a return of Transmeta-style code morphing technology (if Anand is correct), but a hot, gaming focused ARM part doesn't really excite me. The software just isn't there.

 

[BallaTheFeared]

If steambox takes off on the software side we could see arm based tablets compete with wintel for non phone gaming.

Until then though x86 is just too strong and if Intel continues to improve or better yet uses kepler IP I don't see android going anywhere from a gaming perspective in this form factor but I don't know anything.

 

[Qwertilot]

Tablets somewhat problematic for big gaming, tiny smart TV boxes and the like maybe much less so. There's a huge console market to try and take chunks out of and they'll be able to deliver some very cheap options.

Android maybe tricky because any game would have to aim at touch screen operation. StreamOS perhaps a more natural fit for consolish games - with it being based on Debain it'd be very managable to get the operating system ported to this sort of stuff.

The games would of course need recompiling whatever but less work than for Android. I wouldn't be at all surprised if we see it at some stage fairly soon.

 

[DaZeeMan]

lets wait to see how these parts actually perform before coming to final conclusions here. jen hsun talks a massive game but has a less than stellar track record.

but honestly i get what you're saying. given the massive mfg lead intel has, you'd think they'd be able to build something out of the league of what the arm camp can build. silvermont cpu is stellar but gpu wasnt amazing. they need to step it up if they are going to get any real share on the phone side. tablet side i think they have a good opp.

This is why I'm happy to see AMD throwing TSMC some business as well as NVidia, etc. More customers/product demand gives a fab more moolah to throw at the next die shrink. TSMC's revenues are currenly only a third of Intel's, so they can use all the customers they can get!

Global Foundries currently only manages about 10% of Intel's revenue, for comparison.

I don't really see TSMC catching Intel, but they have to find ways to stay somewhat competitive at least. Right now, NVidia's, etc. eggs are pretty much in the 'design smarter' basket, as I seriously doubt that Intel would lend them their 14nm manufacturing...

The days of 'Real Men have Fabs' are long behind us.

 

[Sweepr]

Fun fact - Mullins (AMD's next x86 tablet chip) scores 570 @ 3DMark11 according to AMD, thats a slightly lower score than Kabini A4-5000 (~590 pts). If Tegra K1 really does 60 FPS @ GFXBench 2.7 then it will outclass AMD's solution big time in the graphics department, Mullins should score around 35-36 FPS based on AnandTech's Kabini results.

 

[IlllI]

i always felt AMD missed the boat not getting into this sooner like nvidia did. With the pc market continuing its shrink, I wonder how AMD will keep afloat, since they don't have many other established avenues of income like nvidia.

 

[NTMBK]

i always felt AMD missed the boat not getting into this sooner like nvidia did. With the pc market continuing its shrink, I wonder how AMD will keep afloat, since they don't have many other established avenues of income like nvidia.

Tegra has cost NVidia millions of dollars; I don't think I'd call it an "avenue of income" just yet. Breaking into the consumer ARM SoC market is seriously hard; even companies like TI, which practically owned the market at one point and have their own fabs, have bowed out.