Nvidia`s upcoming Tegra 5 playing Battlefield 3.

[lagokc]

Yeah, it would be fine as just a feature set demo, but it's a tad silly to make a comparison with the A6X, when an actual Tegra 5 wouldn't run it either.

Hopefully it isn't seen as more than a feature set demo. It is a bit of a cheap trick but that's really the only way nVidia can show off the performance of Tegra 5's graphics component as games designed to run on the Tegra 5 have not yet been developed.

Come to think of it... are there any serious 3d games being developed for next gen phone chips?

 

[cmdrdredd]

Hopefully it isn't seen as more than a feature set demo. It is a bit of a cheap trick but that's really the only way nVidia can show off the performance of Tegra 5's graphics component as games designed to run on the Tegra 5 have not yet been developed.

Come to think of it... are there any serious 3d games being developed for next gen phone chips?

Not really because the market out there isn't big enough when you are pushed off the top 10 list by tons of free and $1 games. How do you advertise a serious 3D game that would cost quite a bit to develop on the system that is setup currently? Not easily.

 

[sontin]

Tegrazone.
People who buy a Tegra product have the Tegrazone. Developers who work with nVidia get free advertising.

It's a win - win situation for both sides.

Come to think of it... are there any serious 3d games being developed for next gen phone chips?

Look at all of the Tegra 4 games. Dead Trigger 2 for example. We will see a lot of games next year with Tegra 5: nVidia want to demonstrate all the graphis horsepower. So they need games with great graphics.

 

[NTMBK]

"Louis" the slugger?😵
You meant the famous bat made by the Louisville Slugger Company of KY right?😕

Nah, this guy:

 

[raghu78]

The expectation is that the Tegra 5 "Logan" GPU will have ~ 400 GFLOPS throughput (with an ultra low power Kepler GPU) in 2014.

the Tegra 5 Logan GPU is expected to be on par with PS3 / Xbox 360 at around 200 GLOPS. to even reach that they need a 1 SMX (192 cuda cores) GPU running at 500 mhz . 192 x 2 x 500 = 192 GFlops.

the current Tegra 4 GPU runs at 96 GFLOPS but it still does not feature unfied shaders. Tegra 5 is on the same 28nm node as Tegra 4. Tegra 5 features a Kepler gpu with unified shaders , DX11, opengl 4.3 and cuda support. to even get that kind of feature support and programmability and still double the performance is being a bit too optimistic.

http://en.wikipedia.org/wiki/Tegra#Tegra_4

the Temash A6-1450 tablet APU which features 4 Jaguar cores at 1 Ghz and 128 GCN cores at 300 mhz is supposed to draw 6w. 128 x 2 x 300 = 76.8 GFLOPS. With Turbo dock the CPU runs at 1.4 Ghz and GPU runs at 500 mhz but TDP is raised to 10w or more. 128 x 2 x 500 = 128 GFLOPS

I would be surprised if Nvidia Tegra 5 can get a 150 GFLOPS GPU while running 4 A15 quad cores at close to 2 Ghz and still being a sub 5w SOC . 4 A15 cores at 2 Ghz should draw similar power as 4 Jaguar cores at 1 Ghz.

Nvidia is clearly talking up performance more to drive hype than being a realistic estimate of whats going to be delivered.

 

[ams23]

Is this in the sense that all Maxwell GPUs will include an ARM core to assist them in some way or are they predicting that APUs will completely make standalone GPUs obsolete so they have to integrate the two in order to continue?

What they meant is that all NVIDIA GPU's will have at least some integrated components on die, including but not necessarily limited to the CPU.

 

[ams23]

the Tegra 5 Logan GPU is expected to be on par with PS3 / Xbox 360 at around 200 GLOPS. to even reach that they need a 1 SMX (192 cuda cores) GPU running at 500 mhz . 192 x 2 x 500 = 192 GFlops.

Actually Tegra 5 "Logan" is expected to exceed the performance of PS3 and Xbox 360, and is expected to have slightly higher performance than "Kayla".

the current Tegra 4 GPU runs at 96 GFLOPS but it still does not feature unfied shaders. Tegra 5 is on the same 28nm node as Tegra 4.

The fabrication process node for Tegra 5 is still unknown at this time.

Nvidia is clearly talking up performance more to drive hype than being a realistic estimate of whats going to be delivered.

It is what it is. The Tegra 4 GPU already has close to 100 GFLOPS throughput, so it makes sense that NVIDIA would aim for 200-400 GFLOPS throughput for the Tegra 5 GPU. Keep in mind that this new Kepler.M GPU used in Tegra 5 is probably even more energy efficient than the ULP Geforce GPU used in Tegra 4.

 

[blastingcap]

C'mon, you can do better than that, Nvidia. Video comparisons should be of the same scenes, not this apples and oranges garbage.

 

[ams23]

C'mon, you can do better than that, Nvidia. Video comparisons should be of the same scenes, not this apples and oranges garbage.

Most ultra low power mobile GPU's available today do not support DX11, including the ipad 4 GPU, so there would be no way to showcase Battlefield 3 on these products. It also wouldn't make sense to showcase DX9 graphics on a forward looking Tegra 5 GPU. I think all NVIDIA was trying to show (if it wasn't obvious enough to most people already) is that there will be a big leap in graphics quality and performance next year from a thin fanless tablet form factor compared to the ipad 4 which provides the highest graphics performance in a thin fanless tablet form factor today.

 

[Rezist]

Realistically however Apple will get all the early 20nm wafers and nVidia will be stuck waiting in line.

If this is supposed to be available in 2014 with the delays they have now as well as requiring design wins this will more likely be a 2015 product.

 

[AnandThenMan]

The fabrication process node for Tegra 5 is still unknown at this time.

Pretty much everything about Tegra5 is not known at this point, outside of PowerPoint slides.

 

[ams23]

If this is supposed to be available in 2014 with the delays they have now as well as requiring design wins this will more likely be a 2015 product.

The delay in Tegra 4 was reportedly due in part to bringing forward Tegra 4i and Tegra 5. 2015 will bring Tegra 6 "Parker", not Tegra 5 "Logan".

 

[raghu78]

The fabrication process node for Tegra 5 is still unknown at this time.

so you expect Nvidia's first 20nm tablet SOC a year after Nvidia's first 28nm SOC Tegra 4. keep dreaming. also the 20nm process looks delayed even for GPUs. so you can forget 20nm until 2015 Tegra 6 Parker.

It is what it is. The Tegra 4 GPU already has close to 100 GFLOPS throughput, so it makes sense that NVIDIA would aim for 200-400 GFLOPS throughput for the Tegra 5 GPU. Keep in mind that this new Kepler.M GPU used in Tegra 5 is probably even more energy efficient than the ULP Geforce GPU used in Tegra 4.

with increased feature set and programmability Kepler's energy efficiency will help it fit more transisitors to support the extra feature set. but doubling performance at the same 28nm node is just unrealistic.

 

[ams23]

Pretty much everything about Tegra5 is not known at this point, outside of PowerPoint slides.

If you read in between the lines, there is actually quite a bit that we've been told. We've been told that the Tegra 5 GPU has similar or even slightly better performance than Kayla, which itself has 384 CUDA "cores" and a GPU clock operating frequency of 540MHz according to Anandtech (which implies that the Tegra 5 GPU will have a maximum throughput close to 400 GFLOPS). We've been told that Tegra 5 will be offered in a much smaller package than Kayla (which implies that the Tegra 5 GPU has a relatively small die size area, and may indicate that 192 CUDA "cores" are being used at a relatively high GPU clock operating frequency). We've been told that Tegra 5 will not require a heatsink or fan (which implies that the Tegra 5 GPU has a reasonably low TDP). We've been told that the next click of ultra low power mobile GPU's will outperform the GPU's in PS3 and Xbox 360 (which implies that the Tegra 5 GPU has significantly more maximum throughput than 200 GFLOPS). We've been told that Tegra 5 has support for DirectX 11, OpenGL 4.3, and CUDA 5 (which implies that the Tegra 5 GPU will match the feature set of the latest and greatest Kepler GPU's).

 

[ams23]

so you expect Nvidia's first 20nm tablet SOC a year after Nvidia's first 28nm SOC Tegra 4. keep dreaming. also the 20nm process looks delayed even for GPUs. so you can forget 20nm until 2015 Tegra 6 Parker.

28nm fabrication process is certainly a strong possibility for T5 for more than one reason, but unless you are privy to some inside information, there is no good way to know for sure for an SoC that is at least one year off. Keep in mind that by the time T5 starts shipping, Kepler will have already been shipping on a 28nm fabrication process for 2 full years.

with increased feature set and programmability Kepler's energy efficiency will help it fit more transisitors to support the extra feature set. but doubling performance at the same 28nm node is just unrealistic.

Right, so Kayla must be a figment of our imagination too? At this point, based on what has been reported, there is little doubt that Logan will easily achieve 2-4x the GPU throughput performance of Wayne. The only real question that remains is how NVIDIA will be able to do it while keeping TDP under control if they stick with a 28nm fabrication process for Logan.

 

[notty22]

These chips have dedicated teams that have been working on them for years. A delay in T4 for industry features does not equate to T5 being delayed.
Do a search for T5, you will find articles about it, going back years. Some by Charlie at S/A.
http://en.wikipedia.org/wiki/Project_Denver

The existence of Project Denver was revealed at the 2011 Consumer Electronics Show.[2] In a March 4, 2011 Q&A article CEO Jen-Hsun Huang revealed that Project Denver is a five year 64-bit ARM architecture CPU development on which hundreds of engineers had already worked for three and half years and which also has 32-bit ARM architecture backward compatibility.[3]

http://en.wikipedia.org/wiki/Tegra#Upcoming_releases

 

[Exophase]

We don't know anything about Kayla's perf/W. All we know about it is it needs an HSF. There's no guarantee that it offers even the slightest additional efficiency over existing mobile Kepler parts. Has nVidia even claimed that?

I don't know why it's making a lot of waves either, it's not the first time nVidia paired a standard Tegra SoC with current-gen GPU hardware.

On the topic.. I really don't know where nVidia gets off using running this kind of game on iPad 4 when pretty much all reasonably specced tablets out today can handle vastly superior visuals. Ask yourself this - would nVidia use this game to showcase a Tegra 3 product? If not, why use it to showcase the much more powerful iPad 4? I'm surprised they even believe they can get away with this without scrutiny..

For what it's worth, I'm pretty sure the general consensus on current-gen tablet gaming is far beyond "vintage 1999."

 

[Stuka87]

Am I the only one doubting we will see this in 2014? Tegra 4 has already been pushed out, and we may not see it in products until 2014. 4i is supposed to get released this summer last i checked.

I will be curious to see how Tegra 5 will do everything they say it will with the power envelope they say it will have.

Also, not sure how valid that iPad test is. As BF3 was never released to run on an A6. nVidia's test is with an x86 CPU. So I think the A6 may be running poorly also due to the fact that there is zero optimization for non-x86 processors in BF3. That or nVidia purposely set the graphics much lower. Plus the whole difference scene that was pointed out before.

 

[raghu78]

28nm fabrication process is certainly a strong possibility for T5 for more than one reason, but unless you are privy to some inside information, there is no good way to know for sure for an SoC that is at least one year off. Keep in mind that by the time T5 starts shipping, Kepler will have already been shipping on a 28nm fabrication process for 2 full years.

its known that mobile SOCs are always behind GPUs in adopting the bleeding edge. the fabrication capacity (or volume) required for mobile SOCs is much more than GPUs.

Right, so Kayla must be a figment of our imagination too? At this point, based on what has been reported, there is little doubt that Logan will easily double or even quadruple the GPU throughput performance of Wayne. The only real question that remains is how NVIDIA will be able to do it while keeping TDP under control if they stick with a 28nm fabrication process for Logan.

the specs of Logan GPU and Kayla are not the same.

http://www.anandtech.com/show/6847/more-details-on-nvidias-kayla-a-dev-platform-for-cuda-on-arm

"NVIDIA was quick to note that Kayla is a development platform for ARM on CUDA as opposed to calling it a development platform for Logan; though at the same it unquestionably serves as a sneak-peak for Logan. This is in big part due to the fact that the CPU will not match what’s on Logan – Tegra 4 already is beyond Tegra 3 with its A15 CPU cores – and it’s unlikely that the GPU is an exact match either"

 

[Stoneburner]

I thought Tegra 4 products like the Vizio tablet were supposed to be out by now. Tegra has been an okay platform so far, nothing earth shattering though. It's difficult to gauge the relevance of the demo unless we compare it to Tegra 5's competitors at release.

 

[NTMBK]

These chips have dedicated teams that have been working on them for years. A delay in T4 for industry features does not equate to T5 being delayed.
Do a search for T5, you will find articles about it, going back years. Some by Charlie at S/A.
http://en.wikipedia.org/wiki/Project_Denver

http://en.wikipedia.org/wiki/Tegra#Upcoming_releases

Project Denver is going into T6, not T5.

 

[notty22]

Project Denver is going into T6, not T5.

I had my links out of order. My point was that these projects are in development, some in tandem.

 

[Exophase]

Am I the only one doubting we will see this in 2014? Tegra 4 has already been pushed out, and we may not see it in products until 2014. 4i is supposed to get released this summer last i checked.

If Tegra 4 doesn't make it in a product this entire year then nVidia is in big trouble. And would have to be going through some big problems internally. Look at Shield, it looks more or less ready to me. I don't see what could hold off T4 for such a long time.

I didn't really understand what nVidia's argument about holding off Tegra 4 so they could release Tegra 4i earlier meant. Now that I think about it more the only thing that comes to mind is they have a limited number of wafer starts acquirable from TSMC. And the reason why sacrificing one quarter for 4 wins two for 4i is that 4i is substantially smaller so they can get a lot more per-wafer.

Similar argument for Tegra 5. nVidia is saying Q1 2014 for first devices. I think this is way too optimistic and don't buy it. But not making it out in 2015 at all? These aren't process node timelines, nVidia may traditionally be well off their estimations but never by that much. I expect to see it by Q3 2014 at the latest.

 

[sontin]

If Tegra 4 doesn't make it in a product this entire year then nVidia is in big trouble. And would have to be going through some big problems internally. Look at Shield, it looks more or less ready to me. I don't see what could hold off T4 for such a long time.

Six months from produktion to real products. They are now finishing Shield. So they are on time.

I didn't really understand what nVidia's argument about holding off Tegra 4 so they could release Tegra 4i earlier meant. Now that I think about it more the only thing that comes to mind is they have a limited number of wafer starts acquirable from TSMC. And the reason why sacrificing one quarter for 4 wins two for 4i is that 4i is substantially smaller so they can get a lot more per-wafer.

Their investors moan about their Opex. Tegra was a one product per cycle business. With Tegra 4i they have another one. A lot of the new features needed to be ready to go further: PRIMS 2, 2nd generation 5th core, Computional photography etc. Instead of finishing Tegra 4 on time they delayed it to bring 4i faster to the market. And don't forget: Huang mentioned that they changed their business structure last year.

Similar argument for Tegra 5. nVidia is saying Q1 2014 for first devices. I think this is way too optimistic and don't buy it. But not making it out in 2015 at all? These aren't process node timelines, nVidia may traditionally be well off their estimations but never by that much. I expect to see it by Q3 2014 at the latest.

Huang said they will "easily go into production earlier next year". So first products should appear in Q2 2014.

 

[badb0y]

I expect T4 devices to be out by the end of the year. T5 I am expecting in the middle of next year or end of next year.