Support NeoGAF

[qko]

These are speed holes, They make the console go faster.

Whatcha doing to your XBox neighboreeno?

 

[Doctre81]

Can someone please explain this gif to me, perhaps it's because English is not my native language but I don't get it

L = loss.

Taking "an L" means you lost or you lose, failed etc.

 

[Vic]

800 Gflops sounds like what a Pascal Tegra would be capable of reaching at maximum performance.

But the clocks shown in that picture don't make too much sense.

GPU at 275Mhz => 375 GFLOPS and at 1005Mhz => 875 GFLOPS?

How would this calculation work?

Yeah, the GFLOPS numbers doesn't make any sense.

 

[Smiles and Cries]

Enough for what

for Zelda

But I totally believe that Nintendo's 'Supplemental Computing Device' Patent will show up in some form for Switch's roadmap in a year or so

 

[Osiris]

Yeah, the GFLOPS numbers doesn't make any sense.

The misspelling of Core (Croe, haha) is more of a giveaway.

Bad fake is bad, why even bother?

 

[Doctre81]

So in the last day in this thread the very best from switch (based on it's form factor) that I said I could realistic expect hardwarewise has all been hinted at.

6GB of RAM
128bit bus
800 gflops single precision.

None of it is probably true however outside of maybe the bus. We'll see.

 

[LordOfChaos]

Not sure if anyone else has pointed this out, but the re-translated Foxconn leak said this:



Now, take a look at the heatsink, fan, and fins:

In addition to everything else the Foxconn leak had correct, I just can't see any way that this observation in particular was a wild guess. I'm now fully convinced that the Foxconn leak was 100% correct, which means we're currently looking at a Pascal chip X1 chip, likely using 4 A72 cores, and will be seeing a more powerful dock for 4K gaming later on.

It seems like they did physically see the device internals, but it's still a stretch for a factory production worker to know the specs inside the SoC that are under NDA.

A foxcon worker can leak the design of the next iPhone, but they don't know the CPU details just by looking at the internals without further examination under a microscope.

 

[Osiris]

So in the last day in this thread the very best from switch (based on it's form factor) that I said I could realistic expect hardwarewise has all been hinted at.

6GB of RAM
128bit bus
800 gflops single precision.

None of it is probably true however outside of maybe the bus. We'll see.

Fantasy, it's going to be 4GB, 64bit bus and gflops pretty much close to DF's reported clocks.

The constant attempts at talking it up beyond this is spectacularly amusing though, why the need to inflate into fanfiction what are already good specs for a portable mystify me.

 

[Vic]

So in the last day in this thread the very best from switch (based on it's form factor) that I said I could realistic expect hardwarewise has all been hinted at.

6GB of RAM
128bit bus
800 gflops single precision.

None of it is probably true however outside of maybe the bus. We'll see.

If this is really the July dev kit, none of this is true.

 

[LordOfChaos]

The constant attempts at talking it up beyond this is spectacularly amusing though, why the need to inflate into fanfiction what are already good specs for a portable mystify me.

Half of the Wii Us life was spent with some people waiting on a post-launch overclock because they said 45W average power draw one time, lol.

 

[Kissenkopf]

So in the last day in this thread the very best from switch (based on it's form factor) that I said I could realistic expect hardwarewise has all been hinted at.

6GB of RAM
128bit bus
800 gflops single precision.

None of it is probably true however outside of maybe the bus. We'll see.

Didn't Nintendo themselfes told us we get 4GB of Ram? And I thought we will end with about 400 gflops?

 

[primasaurus]

So does that battery size mean there was no way they could have made it charge faster? Or could there be a Switch revision with fast charge down the line?

Big batteries like this do take longer to charge

 

[Schnozberry]

Didn't Nintendo themselfes told us we get 4GB of Ram? And I thought we will end with about 400 gflops?

4GB is definitely how much RAM we're getting. The clocks are still somewhat unknown. The Eurogamer leak was based on information from July, and we know later revisions of the kit were faster overall. The real question is whether that attributed to a clock bump or to SDK improvements.

 

[Doctre81]

Well if you needed any proof that this is most likely a fake...

They left the ip address at the top...

 

[JonnyDBrit]

Okay so now can someone summarize what, if anything, we've learned from these pictures?

Nothing, right?

I saw Thraktor's post but I don't speak technese.

We haven't confirmed anything, since ultimately the pictures aren't of the best quality, and even what we can see - like the code on the SoC - is lacking context.

From Thraktor's post, it's probable that the two RAM modules are each 2GB, providing a total of 4GB RAM. From Deeke, it may be that the modules are in fact FPRAM, depending on his understanding of the serial codes. Thraktor also theorised that the read speed of the Switch could be 250MB/s, outpacing the harddrive read speeds of PS4 and XBO.

But, again, none of that is outright confirmed - merely suggested by what we can see.

 

[Polygonal_Sprite]

It's all about the letter L today. But discussion has been rather lacking.

L is for lovely.

 

[Murdamonk]

Well, that degenerate quickly..

The more i read these threads, the more I think that the Switch will be even more powerful than a ps4 pro! The Switch is not even out and people are already talking about the Switch Pro in the other thread...

I'll be happy as long as the Switch has good games and can run Zelda without framedrops...

 

[Calm Mind]

Fantasy, it's going to be 4GB, 64bit bus and gflops pretty much close to DF's reported clocks.

The constant attempts at talking it up beyond this is spectacularly amusing though, why the need to inflate into fanfiction what are already good specs for a portable mystify me.

Are you honestly basing your theory on already proven old information? e.g. DF info is based on the not July devkit.

 

[Osiris]

Are you honestly basing your theory on already proven old information? e.g. DF info is based on the not July devkit.

Yes, because that information specifically quoted that the specs were final, from the article:

Documentation supplied to developers along with the table above ends with this stark message: "The information in this table is the final specification for the combinations of performance configurations and performance modes that applications will be able to use at launch."

The words "final" and "at launch" are used in a way that kind of reduces the need for speculation, IMO.

 

[Thraktor]

A quick note on the speculation that it may be a dev kit. From this close up photo of the eMMC module,

we can see the following markings:

16319AE
THGBMHG8C2LBAIL
J46336 TAIWAN

A quick search online shows that it's a Toshiba eMMC module (product lineup here) with a capacity of 32GB.

Given that we know the final product has 32GB of embedded flash, and the dev kits have 64GB, it would point towards it being a final product (or at least a prototype thereof).

 

[Apt101]

Makes me realize that we really, really need innovations in batteries to reduce that size. Imagine what would be possible with some other kind of battery that could pack the same juice at even 75% of the size (and 0% the explosiveness).

 

[Doctre81]

Wait that "leak" says 516 core tegra gpu? So we are expected to believe the switch soc will be xavier territory now? LOL nah.

 

[Christberg]

800 GFLOPs would have this thing nipping at the heels of XBox One. Numbers are off math wise too, going to say team fake on that one. Nothing we've seen indicates that to be the case. 4-500 GFLOPs docked jives with the software we've seen so far. And that's in no way a bad thing.

 

[Polygonal_Sprite]

if it was cpu limited then running at 1080p would not make the frame rate worse - and as mentioned, running at 30fps flawless in portable mode suggests gpu limitation as eurogamer reported that the cpu runs at full clocks while in portable mode.

The resolution limitation suggests a GPU bottleneck rather than a CPU one.

But don't higher resolutions work the CPU harder? I've gotten better framerates on my PC in modern engines by lowering the rendering resolution even though the game wasn't GPU limited.

 

[calc84maniac]

But don't higher resolutions work the CPU harder? I've gotten better framerates on my PC in modern engines by lowering the rendering resolution even though the game wasn't GPU limited.

I think that's more on memory bandwidth... which is increased in docked mode according to reports.

I guess if the CPU is doing stuff directly with framebuffers that could have an effect, but that should really be done on the GPU in modern games.

 

[LordOfChaos]

But don't higher resolutions work the CPU harder? I've gotten better framerates on my PC in modern engines by lowering the rendering resolution even though the game wasn't GPU limited.

The two aren't completely orthogonal, but the impact of resolution on the CPU is much smaller than that of the GPU.

 

[Nanashrew]

L is for lovely.

And a great detective.

 

[Polygonal_Sprite]

I think that's more on memory bandwidth... which is increased in docked mode according to reports.

I guess if the CPU is doing stuff directly with framebuffers that could have an effect, but that should really be done on the GPU in modern games.

The two aren't completely orthogonal, but the impact of resolution on the CPU is much smaller than that of the GPU.

OK, a real shame they couldn't get BotW to 1080p when docked :'( I suppose 900p is still what a 50+% increase in pixels though? Hopefully there are other graphical improvements over the Wii U version too.

And a great detective.

Aww. He sure is!

 

[ShadowFox08]

The two aren't completely orthogonal, but the impact of resolution on the CPU is much smaller than that of the GPU.

Definitely explains the resolution and frame rate differences between PS4 and Xbone for multi plat ports, despite Xbone having a slightly stronger CPU. That being said, the PS4 the lead multiplat console and the other consoles are just ports..

Makes me wonder if there are any multi plat games where xbone displays some kind of advantage using its higher clocked CPU in anything like better A.I. Probably not.

Anyway back on topic..

 

[Apparition]

This very well could the production prototype the FCC received in August.

The question is: if any of the specifications changed in the final revision, wouldn't they needed to resubmit for final approval? Dunno how the FCC works.

 

[Muzicfreq]

Big batteries like this do take longer to charge

Also limitations on USB type c not supporting turbo/quick charge

 

[Buggy Loop]

800 GFLOPs would have this thing nipping at the heels of XBox One. Numbers are off math wise too, going to say team fake on that one. Nothing we've seen indicates that to be the case. 4-500 GFLOPs docked jives with the software we've seen so far. And that's in no way a bad thing.

I'm wondering something, do these kind of tests include also CPU GFLOPs ? They typically don't count for much but it would fuck up the math for just GPU.

 

[Muzicfreq]

Makes me realize that we really, really need innovations in batteries to reduce that size. Imagine what would be possible with some other kind of battery that could pack the same juice at even 75% of the size (and 0% the explosiveness).

There is... But nowhere near efficient in production due to the way it is made.

 

[Deeke[VRZ]]

RE: Switch RAM

I think the K4F6E304HB-MGCH code is right. Searched the web and found this.

http://www.alldatasheet.com/view.jsp?Searchword=K4F6E304HBMGCH

AVNet distributes the RAM, and Samsung Electronics (SEC) makes it.

According to this link, K4F6E304HB-MGCH modules are 16Gb DDR4 modules (Gb = gigabit)

link: https://products.avnet.com/shop/en/...56?CMP=EMEA_FindChips_inventoryfeed_VSE&c=EUR

16 GIGABIT = 2 GIGABYTE

DRAM Chip Mobile LPDDR4 DRAM 16G-Bit 512Mx32 1.8V 200-Pin FBGA

Here's the datasheet.

http://www.samsung.com/semiconductor/products/dram/mobile-dram/K4F6E304HB-MGCH?ia=3107

EDIT: I'm told 3Gb/s isn't that fast. But it's LPDDR4 which is more efficient than previous RAM.

 

[Schnozberry]

This very well could the production prototype the FCC received in August.

The question is: if any of the specifications changed in the final revision, wouldn't they needed to resubmit for final approval? Dunno how the FCC works.

That depends on what changed. If nothing changed regarding the generation of electrical or radio interference, I don't believe it would require re-submission. Devices have changed clock speeds post launch before and the FCC didn't come crashing down.

 

[antonz]

Until we get scans of the SoC and official numbers all that can be done is endless speculation.

 

[Oresama]

Well, that degenerate quickly..

The more i read these threads, the more I think that the Switch will be even more powerful than a ps4 pro! The Switch is not even out and people are already talking about the Switch Pro in the other thread...

I'll be happy as long as the Switch has good games and can run Zelda without framedrops...

Haha Same here!

 

[RoyaleDuke]

Are we looking at a zero hour hardware change that is slightly better or slightly worse than before? Some of this is dizzying to follow.

Anyway, it's neat.

 

[newbong95]

Real of fake? You decide:

4 sm = 128 x 4 = 512 cuda core maybe

 

[AmyS]

So in the last day in this thread the very best from switch (based on it's form factor) that I said I could realistic expect hardwarewise has all been hinted at.

6GB of RAM
128bit bus
800 gflops single precision.

None of it is probably true however outside of maybe the bus. We'll see.

Not going to be true.

4GB RAM - 3.2 GB available for games
64-bit bus / ~25 GB/sec bandwidth
just under 400 GFlops when docked - single precision.

 

[calc84maniac]

4 sm = 128 x 4 = 512 cuda core maybe

516 adding together CPU and GPGPU cores? Man, that would be weird.

 

[Smurfman256]

4 sm = 128 x 4 = 512 cuda core maybe

Where do you see number of SMs?

 

[AmyS]

I thought each SM had 64 cuda cores. 4 SM = 256 cuda cores.

Nvidia won't have 512 cuda cores in a mobile chip until Xavier, which is still roughly a year away from production (Nvidia said Xavier would be sampling at the end of the year).

 

[eighty(one)]

L is real

L is love, L is life.

Spoiler

Team real, I'm on

 

[WhiteRabbitEXE]

Everyone complaining about the battery time and I'm just here being impressed that they were even able to get 2.5-6 hours for it.

Look. One can complain about the battery even if it's the best they can possibly do. Yes, they can. To take it to extremes, if they made a console that was 12TF in portable mode with a 3D 4K screen and 4G LTE at the same size as the Switch, and then by some miracle they got it to last 15 minutes on a charge, that would be insane, sure. But it would still be worthless as a handheld because what the fuck are you going to do in 15 minutes?

So, bringing back to reality, yes, the battery life is the best they can do with those specs at that size. But again, that's still not good, just as it wasn't good for the 3DS, even if again, it was the best they can do. At the end of the day Nintendo designed a device that inherently necessitated several sacrifices, battery life included, and just because they were necessary doesn't make then non-issues. It is perfectly valid to complain about.

 

[Buggy Loop]

I thought each SM had 64 cuda cores. 4 SM = 256 cuda cores.

Nvidia won't have 512 cuda cores in a mobile chip until Xavier, which is still roughly a year away from production (Nvidia said Xavier would be sampling at the end of the year).

Naw TX1 is 2SM, 256 cores. 4SM would be 512 cores.

Personally i think it's not possible, at least not in the same form factor as the new shield android TV (2017) that we saw in this thread. 2SM -> 4SM with nearly identical sizes? Maybe someone more technical could clear things up, but i don't think that's even possible.

16nmFF is roughly the same density as 20nm, where would they fit 2 more SM in there?

 

[newbong95]

Where do you see number of SMs?

Back calculation of 516cores

I thought each SM had 64 cuda cores. 4 SM = 256 cuda cores.

Nvidia won't have 512 cuda cores in a mobile chip until Xavier, which is still roughly a year away from production (Nvidia said Xavier would be sampling at the end of the year).

Nope for maxwell and pascal its 128 per sm and for gcn its 64 per cu

 

[bomblord1]

We're slowly encroaching on a new hardware fanfic thread

 

[LordOfChaos]

I'm wondering something, do these kind of tests include also CPU GFLOPs ? They typically don't count for much but it would fuck up the math for just GPU.

That discussion comes from a paper calculation of GPU flops. In this case,

ALUs (shaders) * frequency in thousand MHz * 2 operations per core per clock = paper GPU Gflops.

Not that I'm throwing my weight behind the 800 number

 

[JonnyDBrit]

We're slowly encroaching on a new hardware fanfic thread

Slowly? We're at over 20 pages here!