Codename_Steam
Cappy
- Pronouns
- He/Him
Are we still expecting Samsung 8nm? Or something better? Can we expect 2 Teraflops in Portable Mode?
6:30 is my favorite time, hands down1:00 on a 12-hour clock, 0:00 on a 24-hour. Or thinking more outside the box, 6:30 on an analog.
* Hidden text: cannot be quoted. *
You didn't! I think optimism is fine! It's taking incomplete data and then using it to become convinced your optimism is cold, hard truth that I think is worth avoiding. Not only does it set you up for disappointment, but it can make the thread kinda toxic.
One group expected 8SMs, and another expected 4SMs and we got 12SMs. Both were wrong - the optimist rationale for 8 didn't hold up. The pessimists reading of the power data did hold up, but Nintendo did something crazy.
Whatever the clock multiple is, I guess. For Switch that was 70-something MHz. Thr problem is that long before that, you don't save any power because you need to be sending a minimum amount of current for the chip to even workhow low can clocks go?
I honked like a goose6:30 is my favorite time, hands down
The crazy thing they did was go for an SOC with 12SMs. That's not in debate really.I feel even more confused now. All I'm getting is that Nintendo did the crazy, unimaginable thing, but that seems to have completely contradicted those test clocks? If Nintendo went in the opposite direction of what we expected on 12 SMs...where are expectations now? Is PS4 in handheld too optimistic now?
The February Direct is a Partner Showcase.
Merry Christmas, everyone.
From what I understand and correct me if I am wrong.I feel even more confused now. All I'm getting is that Nintendo did the crazy, unimaginable thing, but that seems to have completely contradicted those test clocks? If Nintendo went in the opposite direction of what we expected on 12 SMs...where are expectations now? Is PS4 in handheld too optimistic now?
"Merry Christmas" hits different when you end it with a period.I think it would be cool if any bad news on this forum was accompanied by "Merry Christmas, everyone."
e.g.
So...3 hour minimum battery best case scenario, I assume?From what I understand and correct me if I am wrong.
1) optimist were arguing for 8sm as they believe that's all Nintnedo can do.
2) pessimist were arguing for 4sm with a power budget around 12watts or something like that.
Nintendo did something out of the ordinary and went with 12sm going above what the most optimistic generally believe but still specing it to 12 watts that matched the most pestimestic people in terms of power consumption.
As for people doubting the NuSwitch ability...handheld mode should be in comparison to the PS4 and it should match the PS4 pro docked easy with out factoring modern techniques and DLSS at it's worst case scenario.
I feel even more confused now. All I'm getting is that Nintendo did the crazy, unimaginable thing, but that seems to have completely contradicted those test clocks? If Nintendo went in the opposite direction of what we expected on 12 SMs...where are expectations now? Is PS4 in handheld too optimistic now?
Compared to fewer SMs? More GPU power, more tensor cores, more RT cores.Ok, so then let's assume 8nm. Why would Nintendo go for 12 SMs. Cons are a big chip, right? What would the pros be of doing such a move?
Ok cool. So it would disappointing compared to the ultra high end expectations, but still better than the standard "Nintendo gonna Nintendo" expectations?Compared to fewer SMs? More GPU power, more tensor cores, more RT cores.
Basically much easier to get current gen ports.
Plus if they're interested in the tensor cores or RT cores for any of their software ideas (like AI Nintendogs or something) then the more the merrier.
The cons of course as you said are big chip and high power draw. Possibly higher cost in the long run but on 8nm probably cheaper up front.
Exactly my feeling, Nintendo original intentions was for the switch to be a 10 year console but with it's current specs 6 years is long in the tooth for the hardware under it's hood.Ok cool. So it would disappointing compared to the ultra high end expectations, but still better than the standard "Nintendo gonna Nintendo" expectations?
Like nothing is WUST levels of bad right?
Why? Because it’s the path of least resistance. It’s cheaper to go with one GPC over splitting up a GPC and creating a custom GPU.Ok, so then let's assume 8nm. Why would Nintendo go for 12 SMs. Cons are a big chip, right? What would the pros be of doing such a move?
No, it won't bewait, we've been over this possibility before...
christ maybe it really is bigger
that's the only dealbreaker for me so it'd be sort of cosmically funny to me if that's the concession they made
Raccoon playing his brand new Switch 2 on launch day:wait, we've been over this possibility before...
christ maybe it really is bigger
that's the only dealbreaker for me so it'd be sort of cosmically funny to me if that's the concession they made
Let's put it this way.Ok cool. So it would disappointing compared to the ultra high end expectations, but still better than the standard "Nintendo gonna Nintendo" expectations?
Like nothing is WUST levels of bad right?
* Hidden text: cannot be quoted. *
The floor's probably a copy and paste of current clocks for the Switch.how low can clocks go?
* Hidden text: cannot be quoted. *
I'm going to be speculating and making big assumptions below based on LiC's and oldpuck's discoveries (here and here) from the files on the NVN2 folder.* Hidden text: cannot be quoted. *
I would like to remind people of this thread in late February, right before the Lapsus$ hack. The most wildly optimistic of us were calling for 8SMs. The... more moderate wing thought that 4SM was as far as it could go power-wise.
When 12SMs hit, it wasn't the optimists who were proved "right", though. The optimists based their theory on the absolute certainty that redesigning Orin's GPC was so costly, that if Nintendo were to have less than Orin, 8 was the only option. The optimists were wrong.
Meanwhile, Orin's power data showed the pessimists to have been correct across the board about Orin power consumption. But very definitive sources said that, despite those accurate predictions, Nintendo were doing something unexpected.
It's Christmas. The unexpected is cool, it's fun to be excited by what's under the tree, and I love to shake every present to see what's inside. But if you were the kid who assumed they were getting a puppy because one of your presents was clearly a soft, squishy dog toy - well, sometimes it's socks. It could be cool Teenage Mutant Ninja Turtle socks that say "radical!" and "bodacious" every time you walk! You would have loved that... if you weren't expecting a puppy.
I don't know how much I'm gonna be around between now and the New Year, but I'll keeping poking at what I've got (and another source, as it happens) and if something definitive comes up I'll share it. But I hope this little tidbit satisfies folks who are asking "where are the December leaks?!?!?"
For those of you who celebrate Christmas, I hope you get a puppy. For those of you who don't, well, I hope you're warm, well fed, and loved. Hopefully we get a reveal next month, it will be my birthday after all.
But knowing my luck, I'll hear "you just got presents for Christmas. You loved your socks!" Ah well.
* Hidden text: cannot be quoted. *
There's a possibility Samsung's 7LPP process node (and Samsung's 6LPP process node) is discontinued.Wasn't someone saying that Drake could be on Samsung 7nm? Which was apparently more like an 8nm+ or something? Maybe that will actually be what we end up getting.
I just got this.6:30 is my favorite time, hands down
The logical conclusion then, is that it's Samsung, but better than 8nm. What Samsung node would be needed to reach that target? 5nm?It was always going to be pretty damn good at 8nm. I don't know what bug people got that made them think 8nm is some totally unusable, garbage node from the stone age. It isn't. It's perfectly fine. I mean the main concern is really TV mode, since that's where the power will be needed to push 4K. Which isn't a usecase where the power consumption is so much of a concern.
If those targets are anywhere near indicative of final hardware, and I'd say they are given that's what Devs were told to target earlier this year when Samsung and Nvidia would have hammered this processor out years ago, it'll be fine. Launch Switch battery life or better, with Series S-like performance with a higher output res after DLSS. That is more or less the bottom end of what Drake could do. And it's still very, very good.
I mean, the process node was important only in deriving estimates based on what each process node could theoretically achieve at a certain thermal envelope and SoC die size (which we assumed would be similar to Erista). If you have the clocks and the TFLOPs and the number of SMs and cores, the only thing the process node will tell you at that point is how big the chip is and how power-hungry it is, which means nothing to compute performance. It's only relevant when talking about how they'll cram everything into the chassis based on how large the SoC die is and how big the battery will have to be to power it all. Good questions, sure, but nothing that impacts compute performance.LiC's info is far more important than the process node. Over 2tflops handheld, that's much faster than the Steam Deck, and has DLSS and Ray Tracing hardware. It's not a let down, regardless of whatever the process node is, that doesn't actually mean anything to us. It was always the clock that was important, also docked it would exceed the series s thanks to DLSS, 3.456tflops is really good.
* Hidden text: cannot be quoted. *
* Hidden text: cannot be quoted. *
I'm just going to poke one big hole in this. Those frequencies that you are discussing, like thermal throttling and ignore the 660mhz clock altogether, those wouldn't be in a test for DLSS, those are targeted frequencies for use, that's why they come with power consumption results.I'm going to be speculating and making big assumptions below based on LiC's and oldpuck's discoveries (here and here) from the files on the NVN2 folder.
* Hidden text: cannot be quoted. *
* Hidden text: cannot be quoted. *
* Hidden text: cannot be quoted. *
* Hidden text: cannot be quoted. *
So yeah, I very rarely do long speculations. But with LiC's and oldpuck's discoveries, I have to spew out my thoughts before I go crazy. But as I mentioned before, I'm making some big assumptions.
Imo the odds that they somehow made Drake twice as efficient as any other ampere card, without changing the process is slim to none.Nah this makes 8nm far more likely IMO
Type [h I d e] without the spaces.
GPD already did that.I'm convinced all of you guys are going to build the Switch 2 long before Nintendo ever does!
Samsung by itself really isn't bad news. Sure, not as efficient as tsmc, but we don't know which Samsung node.I think it would be cool if any bad news on this forum was accompanied by "Merry Christmas, everyone."
e.g.
GPD already did that.
Valve did too.
Aya NEO too.
It would be truly incredible if it supported DLSS/FSR 3 on top.LiC's info is far more important than the process node. Over 2tflops handheld, that's much faster than the Steam Deck, and has DLSS and Ray Tracing hardware. It's not a let down, regardless of whatever the process node is, that doesn't actually mean anything to us. It was always the clock that was important, also docked it would exceed the series s thanks to DLSS, 3.456tflops is really good.
Can confirm1:00 on a 12-hour clock, 0:00 on a 24-hour. Or thinking more outside the box, 6:30 on an analog.
To many nms. Fewer is better.