Nintendo Switch Event 2017-01-12 and Switch Launch discussion

Status
Not open for further replies.
Assuming you mean "rich", it's irrelevant. The Switch won't use DirectX.
The DX level is just a sort of generational indicator in terms of features implemented in hardaware. They don't need DX as an API to access them. PS4 is using a 'DX12 level' part without using Direct X. Quality of implementation isn't guaranteed either and two DXn GPUs could operate different on certainly workloads, especially the latest features where one IHV hasn't worked out how to get the best of the hardware implementation.
 
And what are they doing then? Calculating some benchmark logic or what?
If you really have no idea what a CPU does during 3D rendering then why are you questioning if it's doing any work or not?

They are not only rough, but don't make any sence at all
It's unfortunate that you are incapable of making sense of it, yes.

Don't get it, if it's TX1 then they can simply port it on 16nm any time they want
Yes. You definitely don't get why Nintendo is unlikely to just launch a new Switch with "2x performance" any time they want.
The only time where it would be important to do something like that would be at release.
300-768MHz on the GPU is clearly pointing to a 20nm from the start.
It took Sony and Microsoft 3 to 4 years until they thought it would be appropriate to launch a mid-gen upgrade with better performance.



The DX level is just a sort of generational indicator in terms of features implemented in hardaware. They don't need DX as an API to access them. PS4 is using a 'DX12 level' part without using Direct X. Quality of implementation isn't guaranteed either and two DXn GPUs could operate different on certainly workloads, especially the latest features where one IHV hasn't worked out how to get the best of the hardware implementation.
I know it can be used as an indicator of featureset, but you really have to wonder if it was meant that way when it came in the same sentence as "has the best image quality in GFXBench, and has the potential for 2x perf improvements with 16nm shrink, does S820 have something of this? Why didn't you mention this?:rolleyes:".
Render quality's meaning in GFXBench is questionable at best (the GTX1080 has worse results than Tegra 3 smartphones) as it could be related to something like a specific driver setting (e.g. Vibrant color setting).
"2x perf improvement with 16nm over S820" doesn't make sense, both because the TX1 doesn't use 16FF and because the S820 actually uses Samsung's 14FF.
The "roll eyes" smiley is just the cherry on top as to the true purpose of that post, really.


As for featureset, Qualcomm had been rather silent about Adreno's 500 support for DX12. Probably because Windows 10 Mobile was shelved after finally being considered a market failure and AFAIK DX12 was never implemented in that platform.
Now that Microsoft announced the "full" Windows 10 to run in Snapdragon 835 with win32 emulation, Qualcomm has announced the Adreno 540 has DX12 support. Meaning all other 500 series GPUs probably do support DX12 as well. It just wasn't relevant to mention that until now:

All we're missing is which feature level.

EDIT: missing AT's credit.
 
Last edited by a moderator:
The DX level is just a sort of generational indicator in terms of features implemented in hardaware. They don't need DX as an API to access them. PS4 is using a 'DX12 level' part without using Direct X.

I would say that "DX12 level part" is a totally pointless classification. Aside IO it is not really the hardware feature that separates what features newer and older hardware can support, it is the drivers.
 
If you really have no idea what a CPU does during 3D rendering then why are you questioning if it's doing any work or not?
I'm asking exactly because I do know what CPU is doing in these benchmarks. In GFXBench, there is no sound, no game logic, no runtime physics, there is however a scene setup, draw calls and driver, all on a single CPU thread with 50% utilization.

It's unfortunate that you are incapable of making sense of it, yes.
Yes, I'm incapable of making sense out of some random napkin notes, which don't prove anything anyway.

You definitely don't get why Nintendo is unlikely to just launch a new Switch with "2x performance" any time they want.
It obviously doesn't make sense to launch a new Switch with 2x perf right now because 16nm is the most expensive tech process right now, they will wait until others will move to 10nm and 16nm prices go down

The only time where it would be important to do something like that would be at release.
Why so? TX1 in Switch has a good perf right now, it's faster than S820 in console mode and we don't know whether S820 would have been any faster in Switch in handheld mode, we also know that Maxwell does well against GCN, there are no major perf holes, so it's possible to port many things to TX1, we don't know whether S820 is capable of handling modern games without major renders rewriting. Personally, I've already tried to port some UE4 demos on Shield Tabled with close to Switch handheld mode perf and it was super easy
This demo is mostly bound by a single CPU thread perf because of 2000-3000 draw calls per frame, there are 20 FPS on average with 720p, 4x AF, TAA, HDR, Bloom, dynamic shadows, nice filtering for shadows, screen space reflections, relatively complex materials and depth buffer physics for particles, this demo uses the general ES3.1 path without any NV specific extensions, so somebody can try to launch it on something with S820 to see if it's any faster.

300-768MHz on the GPU is clearly pointing to a 20nm from the start.
Switch will be based on 20nm chip for obvious reasons, nobody is arguing with this. What I'm talking about is the possibility of mid life 3DS like console update with updated cost reduced 16nm SoC.

It took Sony and Microsoft 3 to 4 years until they thought it would be appropriate to launch a mid-gen upgrade with better performance.
Because these consoles are based on new hardware.

(the GTX1080 has worse results
There are always some questionable entries in every DB of every benchmark, why do you even compare with the most suspicious one?

Meaning all other 500 series GPUs probably do support DX12 as well
DX12 support really doesn't mean anything, feature levels is what matter because GPU can support "DX12" even without such basic FL12 stuff as bindless textures
 
Preorders and early sales mean zilch. There's a basically guaranteed 5+ million unit sales to Nintendo fans; it's sales beyond these that matter. In other words, there are two markets being sold to. The first is those who want a product with Nintendo on it*. Switch satisfies their requirements. The second is the general consumer, who'll be looking for a handheld gaming device, or some digital curio on an impulse buy, or gamers looking for something a little different. The first market gets you no more than Wii U levels of fail. The second market gets you PS4/Wii/DS levels of win.

* :p People who value Nintendo's IP and execution and are happy to have any new Nintendo product on faith alone.
 
I'm asking exactly because I do know what CPU is doing in these benchmarks. In GFXBench, there is no sound, no game logic, no runtime physics, there is however a scene setup, draw calls and driver, all on a single CPU thread with 50% utilization.
There is sound in Car Chase. It has moving objects -> it has game logic. You seem to be mixing up GFXBench's GPU centric benchmark with some single-frame rendering benchmark for the GPU.
"A single CPU thread with 50% utilization" is a meaningless information without knowing which CPU at which frequency. Even the % of utilization can be deceiving because it usually refers to the amount of registers being used and not to the whole pipeline.


Yes, I'm incapable of making sense out of some random napkin notes, which don't prove anything anyway.
You were unable to understand that no one was trying to prove anything even after I explicitly said so, much less the meaning of what you call napkin notes.


It obviously doesn't make sense to launch a new Switch with 2x perf right now because 16nm is the most expensive tech process right now, they will wait until others will move to 10nm and 16nm prices go down
(...)
TX1 in Switch has a good perf right now
So you're saying there's no reason to launch a 16FF chip right now because apparently the Switch already has good performance.
Now this is just pure wishful thinking. That or you definitely didn't see the same footage of launch titles as the rest of the world. Or the fact that FIFA is using the PS360 engine.


it's faster than S820 in console mode
You have no idea how a S820 could perform given the same TDP and cooling in "docked mode".

we don't know whether S820 would have been any faster in Switch in handheld mode
If the TX1 needs to go down to 1GHz GPU and 300MHz GPU in handheld mode, I have little doubt the S820 with disabled baseband processors could do substantially more.


it's possible to port many things to TX1, we don't know whether S820 is capable of handling modern games without major renders rewriting. Personally, I've already tried to port some UE4 demos on Shield Tabled with close to Switch handheld mode perf and it was super easy
(...)
this demo uses the general ES3.1 path without any NV specific extensions, so somebody can try to launch it on something with S820 to see if it's any faster.
You reduced shadowmap and texture resolution from UE4 demo assets, compiled them for Android in apk and ran it in a Tegra K1 tablet.
And somehow this convinced you that, compared to Tegra X1, the Snapdragon 820 might need "major renders rewriting" to run it even though you state the demo is using the general OpenGLES 3.1 path.

Talk about "random napkin notes"...



we also know that Maxwell does well against GCN
Not sure if you're trying to make this into an AMD vs nvidia war here, but that's irrelevant because there are zero handheld SoCs using GCN today in the market.


DX12 support really doesn't mean anything, feature levels is what matter because GPU can support "DX12" even without such basic FL12 stuff as bindless textures
Yet you don't know what feature level is supported by the Adreno 500 series, rendering your previous comparison moot.
 
Preorders and early sales mean zilch. There's a basically guaranteed 5+ million unit sales to Nintendo fans; it's sales beyond these that matter. In other words, there are two markets being sold to. The first is those who want a product with Nintendo on it*. Switch satisfies their requirements. The second is the general consumer, who'll be looking for a handheld gaming device, or some digital curio on an impulse buy, or gamers looking for something a little different. The first market gets you no more than Wii U levels of fail. The second market gets you PS4/Wii/DS levels of win.
I'd go as far as say that increasing production based on a couple of hundred thousand pre-orders is ridiculous for something on the scale of a new Nintendo console.
Either the production increase had been part of the plans for factory contracts a long time ago or there's no production increase. Either way, I call bullcrap on that production statement.


It could be extremely well accepted in Japan and wealthy gamers that feel let down by the Vita demise and the stagnating state of (core) gaming on iOS/Android.
I feel let down by my Vita's demise and the stagnating state of core gaming on Android (which is why I had high hopes for the Switch, to be truthful).
Yet the last thing I want is a bigger and more expensive Vita that may get an even darker future.
At least the Vita is still getting a steady stream of JRPGs. If the Switch, like the Wii U before it, is unable to come down in price, then those japanese developers will just keep making games for the Vita and/or the 3DS.
 
Last edited by a moderator:
I'd go as far as say that increasing production based on a couple of hundred thousand pre-orders is ridiculous for something on the scale of a new Nintendo console.
Either the production increase had been part of the plans for factory contracts a long time ago or there's no production increase. Either way, I call bullcrap on that production statement.
It was a call to investors. You think Nintendo outright lied about what they are doing?
 
It was a call to investors. You think Nintendo outright lied about what they are doing?
I saw it in a tweet and wasn't aware it referred to an investor call.
And looking at that tweet again, the production ramp and pre-orders' status are separated by a comma, so the production part may not be related to the pre-orders at all (it could be linked to some other market indicators, though).
 
There is sound in Car Chase
The talk was about Manhattan, why did you all of a sudden remember Car Chase now?

t has moving objects -> it has game logic
It has animation, why on earth one would need game logic in GPU benchmark

You seem to be mixing up GFXBench's GPU centric benchmark with some single-frame rendering benchmark for the GPU.
Yes, GFXBench is GPU centric benchmark that's why its tests don't have physics, game logic or whatsoever else

"A single CPU thread with 50% utilization" is a meaningless information without knowing which CPU at which frequency
I've already said the meaningful information about CPU, you just don't read

Even the % of utilization can be deceiving because it usually refers to the amount of registers being used and not to the whole pipeline.
Frequency can be higher than 50%, sure, but it doesn't change the fact that only one core is used

So you're saying there's no reason to launch a 16FF chip right now because apparently the Switch already has good performance.
I didn't say anything like this. I said that there's no reason to launch a 16FF chip right now because such chip would cost more than 20nm chip. There is no improvement in density for 16nm over 20nm, which means higher price for a 16nm chip with the same or slightly smaller die area

You have no idea how a S820 could perform given the same TDP and cooling in "docked mode".
I have because there is no S820 with the same TDP, but I have no idea why do you think that S820 can scale to the same TDP in the first place

I have little doubt the S820 with disabled baseband processors could do substantially more.
That's just your wishful thinking

And somehow this convinced you that, compared to Tegra X1
TegraX1@300 MHz has similar params to TK1 at 700 Mhz, TK1 is faster at texturing, which might not be that important considering significantly faster rasterization rate, blending rate and better bandwidth and caches of TX1, my regression based perf model shows 10% higher overall perf for TX1 @ 300 MHz

the Snapdragon 820 might need "major renders rewriting" to run it even though you state the demo is using the general OpenGLES 3.1 path
I'm not sure whether it will work on S820 at all considering bugged ES3.1 driver of qualcomm. It's render which might require rewriting since there are many things which can cause app crash on S820, I am not even talking about performance, performance can be awful on custom mobile architectures without careful tuning

Yet you don't know what feature level is supported by the Adreno 500 series
That's simple, if Qualcomm's marketing department doesn't claim FL12 support then there is no FL12 support. All other guessed are based on wishful thinking
 
I saw it in a tweet and wasn't aware it referred to an investor call.
And looking at that tweet again, the production ramp and pre-orders' status are separated by a comma, so the production part may not be related to the pre-orders at all (it could be linked to some other market indicators, though).
Why struggle to parse a Tweet's punctuation when it's not hard to find a better source?

https://www.bloomberg.com/news/arti...fit-lifts-forecast-on-pokemon-ahead-of-switch
That’s putting the focus squarely on the Switch, which goes on sale March 3 and will seek to fuse home and mobile gaming. Nintendo said it expects to sell 2 million units by the end of March, with analysts predicting about 10 to 11 million units in the 12 months after that. While game consoles are usually sold at a loss early in their life cycles, the Switch will be profitable from the start, Nintendo said.

“You can tell customers have huge expectations based on how Switch reservations are doing,” Kimishima said at a briefing. “We want to increase production as much as we can.”
 
I doubt they will sell 11mill units the first year. Nintendo fanatics will buy it so it will sold out at first, but then who ?
 
Even the % of utilization can be deceiving because it usually refers to the amount of registers being used and not to the whole pipeline.

It doesn't refer to either of those things, it's basically the percentage of time the CPU core is actively running that task (vs running some other task or being idle) multiplied by the ratio of current clock speed to max clock speed.

So if something needs 50% CPU it could in the best circumstances (depending on some other factors) run with the CPU at half the maximum clock speed, or at maximum clock speed with the CPU spending half of its time clock or power gated. Or doing twice as much stuff.

In the case of games or graphics benchmarks that means it's GPU limited and/or frame limited, at least on average.

If your power budget for 4 corse is 4xFMax_power and you only need 50% of one core it means you may be able to use significantly below 1/8th of the total CPU power budget. Very, very roughly of course.
 
Online gaming with switch to cost roughly 25 USD per year
The online service to make multiplayer possible on the new Nintendo Switch will cost roughly 25 euro/usd per year. The CEO of Nintendo stated in an Asian newspaper.

The upgrades will likely apply to multiplayer online gaming and the library of downloadable classic games, and carry an annual fee of 2,000 yen to 3,000 yen ($17.60 to $26.40). "With paid [services], we will be able to fully commit to customers," said Nintendo President Tatsumi Kimishima.

Online multiplayer and other services are available for free on the predecessor Wii U console. For the Switch, Nintendo is introducing a smartphone chat app as well as monthly downloads of classic games. The Switch will hit stores in March.

Kimishima also said Nintendo is "studying" adding virtual reality functionality to the new console. "If we are able to resolve the issues with playing [VR] comfortably for long hours, we will support it in one form or another," he said, without providing a time frame. Nintendo is planning to release two or three smartphone games in the fiscal year ending March 2018, the president also said. The launch date for the "Animal Crossing" series will be pushed back to fiscal 2017 from the current financial year.
http://www.guru3d.com/news-story/online-gaming-with-switch-to-cost-roughly-25-usd-per-year.html
 
It makes the "free rental" model more palatable now that they're not trying to charge the same as Sony/MS, still not attractive to me but less objectionable. I find reading articles from non-specialist publications very frustrating (let alone those fan translated versions of the same), I mean the phrase "library of downloadable titles" implies a netflix style arrangement but this could just mean library in the sense I have a library of PS+ titles that I've built up over time. With launch still so close though I'm kind of surprised that we haven't seen any actual details on how chat, etc are to be handled other than it involves a phone app somehow. Whatever else changes at Nintendo their (non-)communication strategy remains the same.
 
Half the price of XBL/PSN does make it a little better - though they'll still need a strong library of multiplayer games. Smash and Mario Kart might be fun, but the quality, long tail MP games will need to keep coming.

In terms of sales I'm guessing somewhere between WiiU and 3DS. It's Nintendo's handheld crowd that'll be bringing in the larger numbers.

Still think lower cost dedicated handheld and home console models would be a good idea at some point down the line.
 
Depends on the quality of service, but the price isn't bad at all. 'Library of classic games' is being a bit generous though as you only get a monthly rental of one title at a time as I understand it. You neither have a library you can pick and choose from, nor build up a personal games 'library' (should be called collection) like the others.
 
Online gaming with switch to cost roughly 25 USD per year

$26 is based on current approximate exchange rate, but then we have a 29,980 yen console that translates to $299 USD, so it could just be $30USD.
 
Status
Not open for further replies.
Back
Top