NVIDIA Tegra Architecture

All device scores fluctuate heavily between different runs and typically the highest official score is taken for the final placement. Of course highly convenient that there's only ONE T4 based device now in the database and for a quite limited time, which serves your purpose pretty well to even further bend reality until it gets even more ridiculous.

See above. It's definitely not my fault that T4 has such a lame design win amount, is late or stars didn't allign in NV's favour. But thanks for actually proving my point over and over again.

It's not needed because the minority report of a Tegra market share can't enable it up to T3 since it isn't possible and while it's present on T4 it takes a quite heavy performance drop compared to any tile based GPU out there where it's in the majority of cases nearly for free.

There are top tier mobile games that either come with MSAA enabled for select devices, so it's definitely not anything absurd for that space. It's just that it isn't convenient obviously.

 

There are select racing sim mobile game cases where you can enable even 16xAA on a S3 which is actually a combination of 4x super and 4x multi-sampling; presupposition of course would be that there's fillrate to spare.

TAA stands for transparency anti-aliasing? If yes I'm afraid it's not the same as Multisampling, where it also besides all other AA method differences it also comes down to samples used, grid and a bunch of other details, where it's of obviously not that easy to drop them all into the same pot.

 

Well, I've seen a lot of Mali benchmarks, including micro architectural benchmarks, msaa is not the streight point of it, in fact Nexus10's Mali, as well as Adreno 320 are slower with MSAA compared to IMR such as Tegra 4 GPU, because you don't take into account the frame buffer compression, caches and NVIDIA expirience(fast MSAA algorithms), you could believe that tilers are faster with MSAA by default, but this is not anywhere close to thruth.

I thought it's obvious that I have been speaking for temporal anti-aliasing

 

Isn't the Adreno 320 pretty much an ATI design? What about their experience with MSAA?

 

It's been so many years that I wouldn't say you can call it an ATI design anymore.
AFAIR, the last Adreno sold by ATI was the Adreno 200, under the name Imageon Z460, and by then it was already a part of AMD.

 

I remember running Mafia with 16xS AA (same as you describe) on a geforce ti 4200. The game was not very GPU demanding and very CPU limited.

 

One would think that there's still quite a bit of former ATI employees left at Qualcomm? It's not the kind of company that employees run away from.

 

not too bad for a so called doomed SoC

even if it's late, T4 has now some interesting design wins. After HP Slateboook X2, we have
2 new tablets from Toshiba:
http://www.engadget.com/2013/06/03/toshiba-excite-pure-excite-pro-excite-write/

and the new Asus infinity pad:
http://www.techradar.com/news/mobil...rings-tegra-4-upgrade-eyes-q3-release-1156130

I see a trend on Computex. Low end tablets for Mediatek, mid range for Atom Clovertrail+ and high-end for T4. Does it mean that intel SoC gfx is not powerful enough for these new very high res panels (2560x1600) ?
Very few Qualcomm too...

 

Ironically early T4 results (with reduced clocks) showed a rather huge drop with MSAA enabled which was definitely over 40%, so I think your guesses grasped out of thin air are completely meaningless. Yes NV has magic wands within its MSAA algorithm which can magically replace the memory footprint and bandwidth consumption savings of a tile based solution.

Here's a tidbit of actual result for the Mali T604 in the Nexus10: http://gfxbench.com/device.jsp?benchmark=gfx27&D=Google+Nexus+10 which accounts for a 12% drop from 1xMSAA to 4xMSAA under 2560*1600 and here the same GPU in 1024*561: http://gfxbench.com/device.jsp?benchmark=gfx27&D=Samsung+Arndale+Board with a 5% drop.

How about you show me any Tegra4 result compared to that one since we're at it?

Unfortunately with the GLB2.5/2.7 merge the GLB2.5 MSAA results are gone, but I don't recall any unexpected big drop for MSAA on a Adreno320 either.

I think it's obvious that temporal AA and multisampling AA are quite different beasts. Lord knows what temporal AA in the given case even stands for and I'm not even digging any further since the answer could be too scary.


***edit: somewhat OT but I'd prefer Kishonti would use in the coming years a performance and high quality presets for its cross platform benchmark. Performance can remain 1080p 1xAA/1xAF and High quality something like 1080p 2xAA/4xAF. It might be too early for today's conditions, but in a while mobile SFF GPUs will be powerful enough to handle the latter with ease.

 

I just don't think any substancial part of the Adreno 320 was developed when the team was still under the ATI name.

BTW, another design win:

http://www.theverge.com/2013/6/3/4392554/toshiba-excite-pure-pro-write-android-tablets-hands-on

 

This statement makes no sense. Statistically speaking, when looking at a large distribution of scores, it generally makes the most sense to use the Median score (or Average score if the Median score is not available) for comparison to another single data point, because the Median score represents the mid-point around which all scores are distributed (ie. there are just as many scores below the Median score as there are scores above the Median score). In the case of the SGS4 S600 variant, the Median score ( http://gfxbench.com/device.jsp?D=Sa...+S4&testgroup=overall&benchmark=gfx27&var=med ) happens to match up very closely with the Average score ( http://gfxbench.com/device.jsp?D=Sa...+S4&testgroup=overall&benchmark=gfx27&var=avg ) and with the Anandtech Review score ( http://images.anandtech.com/graphs/graph6914/54302.png ). So compared to the Median score of S600-powered SGS4 and HTC One, the T4-powered SlateBook x2 is close to 25-37% faster, respectively, in GFXBench 2.5 Egypt HD Offscreen (1080p) test.

Note that there are several ways to get higher-than-average or higher-than-median benchmark scores. One is to run short individual graphics tests so that the SoC does not heat up enough to thermally throttle frequencies. Two is to test the device in a freezer or in frigid temperatures so that the SoC does not heat up enough to thermally throttle frequencies. Three is to root the device to manipulate operating frequencies. Since the vast majority of consumers will not test their device under any of these conditions, it makes perfect sense to use the Median (or Average) score when it comes to statistical analysis versus another data point. The only time it would make sense to not use a Median (or Average) score is if there was a significant change in performance through a recent software update, but these phones have only been on the market for a few weeks, so it is unlikely that this would be the case (and even if this was the case, it would still be erroneous to focus on one top score out of many other lower scores for the reasons mentioned above).

The number of design wins is completely irrelevant to what was being discussed. Anyway, more important from a consumer perspective is the quality of design wins. I'd say that T4 design wins with HP SlateBook x2, Asus Transformer Pad Infinity, Toshiba Excite Pro, Toshiba Excite Write (and surely more to be announced at a later date such as the Vizio high res tablet) all represent quality tablet or tablet convertible designs at a reasonable price for consumers. The HP SlateBook x2 is $479 USD retail price with an attachable keyboard and with a battery integrated inside the keyboard. The Asus Transformer Pad Infinity is $399 USD retail price with a 2560x1600 resolution screen and with an optional attachable keyboard + battery integrated inside the keyboard available for only $100 USD more.

I do believe that you are very wrong here, because I seem to recall that Adreno 320 actually does have a significant drop in performance when moving from no MSAA to 4xMSAA in GLB2.5, and certainly a much larger drop in performance than something such as SGX 5xx.

 

It still remains a fact that there's only one T4 device on display with less than a handful of test runs yet. Now is that all you could come up with to bore me to death or is there more?

Another of those funky defense line I'm tired to read. Try to convince me with the usual let's beat around the bush tactic that market share or penetration is completely irrelevant

Agreed.

How significant exactly? Something like 20%? SGX MP is also a different beast since under ideal situations the difference might end up even at zero, because they're scaling entire cores and each of them happens to have 16 z check units: http://gfxbench.com/device.jsp?benchmark=gfx27&D=Samsung+GT-I9500+Galaxy+S4

 

It seems like Nvidia lost another fairly important design win in Surface RT 2. Anand is saying Qualcomm won the contract (wich fits with Nvidias silence about Tegra 4 and Win RT 8.1)

While the original was not a massive success, it still shipped/sold 1,5 million units wich is fairly decent considering it competed against iPad in price and better than Nexus 10 and Transformer Infinity. The 2nd generation version is supposedly an 8 incher and going to be fairly cheap. Assuming Qualcomm is also inside the Nexus 7 successor, thats 2 fairly important design wins Nvidia has lost out on this year

 

Microsoft sold 1,5 million Surface Tablets, of which 400k were Surface Pro. I also heard that the return rate for the Surface RT was extremely high because people would buy them thinking they could install windows x86 software in them... so the amount of people actually using a RT may be way less than 1 million.

To be honest, I'm not so sure nVidia lost the RT 2 or just didn't compete, as the platform seems like a complete waste of time and resources.

 

IMO the lacklustre sales have more to do with price points than OS confusion. I believe the Nexus 10 for example has only sold half a million units, In one of the patent trials it was mentioned Transformer Prime had sold less than 100k units after 6 months, i doubt the Infinity sold a record amount either. Im not sure what your definition of success is but over a million units makes it a better seller than any Android tablet at that price point so its still a fairly important design win. An 8 inch version for 299 could sell more than 1 million

 

Dead Trigger 2 image quality comparison between Tegra 3 and Tegra 4: http://www.youtube.com/watch?v=u_tQnK6-NAY&list=UUqSNM6HkOfv8JdGXniY2Pcw&index=1

 

Tegra 4 image processing for stylus support: http://www.youtube.com/watch?v=aaK8EAJ-IbY

 

As useless as the first game where you could just enable the "Tegra effects" by simply editing a configuration file, all while having smoother frame-rates than Tegra devices themselves.

 

I don't think you understand the comparison here. Dead Trigger 2 with Tegra 3 wouldn't even be playable with all the added effects shown on Dead Trigger 2 with Tegra 4. Whether it would be playable on other new and powerful SoC's with the added effects is not under debate here. And FWIW, I doubt that there is any ultra mobile SoC available now or in the very near future that will offer smoother framerates than Tegra 4 on Dead Trigger 2.