LordOfChaos
Member
PC, if we're going with the same allowance of "not exclusively for the top end hardware"
-
Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.
The iPhone 7 will the be the most powerful gaming system any Nintendo game has run on
- Thread starter nhlducks35
- Start date
PC, if we're going with the same allowance of "not exclusively for the top end hardware"
OttoSporteman
Member
Well..
Check out @L3GOSTEIN's Tweet: https://twitter.com/L3GOSTEIN/status/776865400952918016?s=09
Check out @L3GOSTEIN's Tweet: https://twitter.com/L3GOSTEIN/status/776865400952918016?s=09
What games did Nintendo ever release on PC? Like some old Japanese PC? iPhone 7 is probably faster than most old PCs unless they're just a few years old
LordOfChaos
Member
Mario is Missing, and yeah an iPhone 7 is way more powerful than PCs from when that came out, but Mario run also runs on much older iPhones, so applying the same rule both ways, PCs would be the most powerful gaming platform a Nintendo game ever ran on as a modern PC could still run it.
It makes the title utterly meaningless, but it already was as it wasn't built from the ground up for the 7. I still do get the OP though, it's an interesting tidbit that if Nintendo focused on top end iPhone hardware, on a hardware level it would be more powerful than any hardware they made.
Boombox On
Member
The oldest iPhone that can run Super Mario Run will almost certainly still be faster than any consumer-friendly PC that came out in 1992, the year the game came out..
AtomicShroom
Member
Oh come on that doesn't count. The MS-Dos version was published by Mindscape, not Nintendo, and I seriously doubt it would run natively today without Dosbox, meaning that it's emulated. And if we're going to just bundle emulation into the equation, then...
LordOfChaos
Member
Well, If Mario Run runs on a 4S or 5, it would invalidate "most powerful gaming system a Nintendo game has run on" by way of it not being exclusively for the 7 hardware, it's generally for the iOS platform which includes newer more powerful hardware if we apply the same rule of forward compatibility doesn't count.
32 bit Windows can run 16 bit applications, certainly you could find a 32 bit computer out there with more GPU grunt than an iPhone.
This is getting into uninteresting semantics games anyways, as I said I do get the OPs point, it is an interesting factoid that if they took full advantage, modern iOS hardware would be more powerful than any dedicated Nintendo hardware.
Quite a few games used deferred shading on the SPUs, especially later on in the console generation as the SPUs were better than xenos at deferred lighting passes.
Negotiator
Member
CISC vs RISC? LOL, where did that come from?!
I said MIPS and flops... you know, integer vs floating point operations?
I guess you still don't understand why Jaguar is 2.5-3x faster than the Xenon CPU in general purpose code, right?
You even ignored this site: http://7-cpu.com/
I wonder why...
Regarding flops, do you realize that Jaguar at 1.6 GHz (half the frequency of Cell) can perform 102.4 Gflops? Cell has 200 Gflops at 3.2 GHz. This says to me that Jaguar has a really efficient design (it reminds me of this tweet: https://twitter.com/marcan42/status/274179630599131136).
I'm not contradicting anything. For number crunching, we have GPUs these days. Cell is obsolete, but its legacy still lives in every AMD APU.
ND is even using the GPU for AI pathfinding in Uncharted 4. If the PS4 had a hypothetical Cell 2 with 32 SPUs, they would use those instead. Proficient developers will always use the most appropriate tool in their disposal, not the other way around.
ps: AMD doesn't have good ARM CPUs. There's a reason they put x86-64 cores in their APUs.
AFAIK, the bulk of pixel shader operations were performed by the RSX.
It was vertex shaders that the RSX was really weak at and it needed some Cell "assistance".
Negotiator
Member
http://www.fudzilla.com/news/processors/39179-jim-keller-was-not-a-big-fan-of-k12
x86-64 = better performance
ARM = more suitable for mobile devices
Also, I don't think ARM64 was ready back in 2012 when they finalized the APU designs... they wanted 8GB of RAM (Microsoft mostly and then Sony), so ARM32 wouldn't cut it.
I think you might have a fundamental misunderstanding of processor ISA's if that is what you believe it boils down to.
RSX mainly rendered the g-buffers, but in games that used deferred shading pretty much all of the pixel shading operations were done on the SPUs, including post processing and anti-aliasing. That's not to say RSX was horrible at it, but SPUs were much better.
Negotiator
Member
I think you don't know who Jim Keller is and why he dismissed K12 (ARM-based) in favor of Zen (x86-based). There's a reason ARM is restricted to mobile devices mostly.
Not even Apple uses their (spectacular) ARM cores outside of mobile devices.
ps: Servers run Linux for the most part. Windows is a more popular OS in desktop computers.
Not when you have a cross-platform benchmark: http://7-cpu.com/
Are you still going to pretend that Cell PPE was a great CPU back in 2005? Because even a shitty, inefficient Pentium 4 from 2004 could beat it in plenty of benchmarks.
Do you understand the microarchitectural differences between in-order vs out-of-order? Microarchitectures have nothing to do with architectures.
Not to mention that the CISC vs RISC distinction is pretty much obsolete these days, since all modern x86 CPUs have RISC cores that process micro-ops. x86 is merely a frontend that translates/decodes x86 opcodes to micro-ops.
Now you're putting words in my mouth and that's not very cool of you...
I said gaming related SIMD code, like physics, destruction and stuff like that that can be GPGPU accelerated.
Do you even understand what SIMD is? Have you ever studied linear algebra and more precisely matrix multiplications? The 7-cpu benchmark has nothing to do with flops, if you bothered to read what it says: "The test code doesn't use FPU and SSE. Most of the code is 32-bit integer code."
I think you're confused by the "general purpose" term. GPGPU doesn't mean that GPUs can run everything, otherwise we wouldn't need a CPU. You'll always need a CPU to orchestrate the GPU and execute code with lots of branches.
A CPU is like the conductor in an orchestra, while GPU Compute Units are the equivalent of guys playing the violins. They're really good at it, but they can't really do anything by themselves.
Let me get this right... People are telling that the iPhone 7 will be stronger than Videogames like Wii U, Ps3 and Xbox 360?
That machine will be able to handle games like GTA 5 without a problem?
Is really that I reading?
Yeah, I've seen iPhone games that look better than the PS3 screenshot you posted
yes I replaced the PC screenshot you actually posted
blu
Wants the largest console games publisher to avoid Nintendo's platforms.
So 8 Jaguars @ 1.6 do 100 GFLOPS, and 8 SPUs @ 3.2 do 200 GFLOPS - where exactly is the FLOPS efficiency advantage of the Jaguars?
AMD had to prioritize their product lines and they naturally went after the better short-to-mid-term ROI option. Keller did not 'dismiss' K12 - he designed it. AMD had to decide what to launch first on the market, and they made the right choice.
And since you brought up apple:
http://daringfireball.net/linked/2016/09/14/geekbench-android-a10
Negotiator
Member
Maybe the fact that Jaguar has the same amount of flops per Hz? If it could run @ 3.2 GHz, it would be on par with Cell.
Not to mention power consumption (Jaguar @ 1.6 consumes 30 watts)...
That still doesn't explain why Apple has to buy CPUs from Intel, instead of putting their own ARM-based ones... they made the switch from PPC to x86, so it shouldn't be that hard to go full ARM.
blu
Wants the largest console games publisher to avoid Nintendo's platforms.
Leaving aside the fact Jaguar cannot run at 3.2GHz by design (it'd have a very different pipeline if it could), is CELL some measure for FLOP efficiency these days? I can think of a dozen CPUs that do the same (or higher) FLOPS/clock. CELL surely had lots of FLOPS for its time, but my 4-core desktop has those FLOPS and then some. So your impressive efficiency comment is puzzling. Heck, a 16-core Epiphany does 32GFLOPS at 2W and 65nm - now that's impressive for a non-GPU. Jaguar is a run-of-the-mill mobile-class design. Sure it has the best IPC in the x86 entry-level mobile word but that alone is not much of an achievement.
That's what lithography advancements can do to your designs. For reference, my intel IGP does 320GFLOPS @ ~8W @ 22nm fabnode, whereas my 2009 Tesla does 34GFLOPS @ ~20W @ 40nm. Woe is nvidia, I guess. Drawing efficiency conclusions out of thin air, across multiple lithography node generations, taking TDP ratings for linear (how many Watts would an 8-core Jaguar consume at 3.2GHz again?) does not really help make a point. Just saying.
Profit margins would be my uneducated guess.
Negotiator
Member
I never said that it would be able to run @ 3.2 GHz. I know that it has a short pipeline.
It's like comparing P3 vs P4. They were not designed with the same goals in mind.
I never said that consumption scales linearly. Jaguar @ 2 GHz consumes 50 watts (@28nm).
Lithography advancements do not always help that much (e.g. Polaris vs Maxwell/Pascal).
Apple has a higher profit margin by designing their own CPUs.
blu
Wants the largest console games publisher to avoid Nintendo's platforms.
That still leaves me puzzled about your 'impressive efficiency' comment. What exactly is impressive in Jaguar's 4-way SIMD * 2 ops /clock in this day and age?
But you quoted a power rating @ a given clock, in the context of Jaguar having the same FLOPS/clock as CELL. And where are you taking those TDPs from? I hope you're not extrapolating from APU numbers, as those never give the CPU ratings.
Lithography advancements always help. Clearly NV have had the architecture TDP edge with Maxwell/Pascal, but AMD cannot blame lithography for that.
Apple are already designing their own CPUs. But they'd also have to produce them. Whereas Intel are masters of production but they do not quite design stellar tablet CPUs. It'd be curious to know why you think TSMC/Samsung can provide better prices per transistor than Intel.
mm hmm
Toadthemushroom
Member
So yeah, the Apple A10 fusion is an absolute beast. It only ever uses two cores at once, but the fact that the two low power cores (which are clocked at 1.1Ghz) are faster than my iPhone 5s, which is already competitive with flagship Android phones in single core performance, blows my mind.
I don't think anything Nvidia can provide Nintendo with, whether it be the Tegra X1 or X2, will be able to match the A10.
Supply and demand. Every A10 fabricated could do towards a new iPhone 7 or 7 Plus, instead of the lower-margin Apple TV.
It's a shame, I was hoping the new Apple TV would have, at the time, had an A9X on board. But A9Xes had much more worth in the iPad Pros.
I don't think anything Nvidia can provide Nintendo with, whether it be the Tegra X1 or X2, will be able to match the A10.
Supply and demand. Every A10 fabricated could do towards a new iPhone 7 or 7 Plus, instead of the lower-margin Apple TV.
It's a shame, I was hoping the new Apple TV would have, at the time, had an A9X on board. But A9Xes had much more worth in the iPad Pros.
Negotiator
Member
@blu
It all boils down to the fact that people are so eager to shit on Jaguar, just because "Cell has moar flopz!!!!!!1111ONEONE"
Yeah, it has double the amount of flops, but it also has double the frequency of Jaguar. That's all I was saying.
Honestly, what's so impressive about Cell minus the SPU part? Would Cell be a memorable processor if it had no SPUs at all? I appreciate the legacy it left behind, but that's it.
PPE was mediocre back in 2005 (let alone in 2016) and it should not be praised at all.
Regarding Intel, if what you're saying was true, then all smartphones/tablets would have x86 SoCs. Intel could outsell ARM if they wanted to long time ago, but they never cared about selling cheaply, even though they always had the lithography advantage.
Apple does not produce chips, they merely design them. nVidia does the same.
It all boils down to the fact that people are so eager to shit on Jaguar, just because "Cell has moar flopz!!!!!!1111ONEONE"
(even though most of them don't really know what flops are, but I guess it's a cool buzzword)
Yeah, it has double the amount of flops, but it also has double the frequency of Jaguar. That's all I was saying.
Honestly, what's so impressive about Cell minus the SPU part? Would Cell be a memorable processor if it had no SPUs at all? I appreciate the legacy it left behind, but that's it.
PPE was mediocre back in 2005 (let alone in 2016) and it should not be praised at all.
Regarding Intel, if what you're saying was true, then all smartphones/tablets would have x86 SoCs. Intel could outsell ARM if they wanted to long time ago, but they never cared about selling cheaply, even though they always had the lithography advantage.
Apple does not produce chips, they merely design them. nVidia does the same.
ultimatebestusername
Member
The iPhone 7 benchmarks are really impressive, but the problem with smartphone games is that they are usually made targeting a min spec that is far less powerful than the current flagship, so we don't often get to see what the latest hardware is capable of outside of tech demos in press conferences.
Maybe they just don't want to get into the hardcore gaming side of the market and go against consoles. They seem largely cautious about going in big on markets they don't see immediate returns on. Just speculation on my part though.
Also I don't think they want to put the AppleTV on a yearly upgrade cycle yet. I suspect it will be more like Applewatch with a year and a half or two between updates.
mangochutney
Member
Min spec for Super Mario Run seems to be the 5S. It won't run on anything older than an A7 it seems.
Chû Totoro
Member
Yes but not because the game is too demanding but because of iOS constraints.
An iPhone is not a dedicated gaming system so you can't use the resources like you would do it on a console.
Do people really think that a runner game needs the power of an iPhone 7 (or even 5S) to run properly? You can't be serious...
LordOfChaos
Member
How was that written for Cell? As I was getting at in earlier pages, code complexity blew up dramatically with it, but it could also do some good pushing once all was said and done.
This routine for edge detection
Code:
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <unistd.h>
/* ... */
/* the graph */
vertex_t * G;
/* number of vertices in the graph */
unsigned card_V;
/* root vertex (where the visit starts) */
unsigned root;
void parse_input( int argc, char** argv );
int main(int argc, char ** argv)
{
unsigned *Q, *Q_next, *marked;
unsigned Q_size=0, Q_next_size=0;
unsigned level = 0;
parse_input(argc, argv);
graph_load();
Q =
(unsigned *) calloc(card_V, sizeof(unsigned));
Q_next =
(unsigned *) calloc(card_V, sizeof(unsigned));
marked =
(unsigned *) calloc(card_V, sizeof(unsigned));
Q[0] = root;
Q_size = 1;
while (Q_size != 0)
{
/* scanning all vertices in queue Q */
unsigned Q_index;
for ( Q_index=0; Q_index<Q_size; Q_index++ )
{
const unsigned vertex = Q[Q_index];
const unsigned length = G[vertex].length;
/* scanning each neighbor of each vertex */
unsigned i;
for ( i=0; i<length; i++)
{
const unsigned neighbor =
G[vertex].neighbors[i];
if( !marked[neighbor] ) {
/* mark the neighbor */
marked[neighbor] = TRUE;
/* enqueue it to Q_next */
Q_next[Q_next_size++] = neighbor;
}
}
}
level++;
unsigned * swap_tmp;
swap_tmp = Q;
Q = Q_next;
Q_next = swap_tmp;
Q_size = Q_next_size;
Q_next_size = 0;
}
return 0;
}While a mere 60 lines on a general processor (P4, Jaguar, whatever) and an exacerbating 1200 on the Cell, you also went from the Pentium 4 HT running at 3.4 GHz checking edges at 24-million edges per second. On the Cell, 538-million edges per second. A 22x speedup with a lot of elbow grease.
Remember when devs figured out to move antialiasing to the Cell? That would be why. Modern GPUs are now into the billions of edges per second, but it was impressive for a 2006 CPU.
Would a P4 have produced better performing games at the start of the gen? Surely. Would it have carried through as much to the end of the gen? Probably not.
You're absolutely right in that a general benchmark written for generic processor X will have even a P4 run circles around the Cell, but it's ignoring everything Cell was built in mind with, it's a deliberate move from hardware complexity (prefetching, branch prediction, caches) to the programmer having a lot more complexity, but in turn it dedicates more transistors to stuff that makes it fast.
I don't agree with preferring it over the Jaguar, but I quite disagree with looking at a generic 7Zip benchmark as a way of seeing its potential. All it says is that it's not great at unmodified general purpose code.
Nor, by the way, will I agree that ARM is bad for large processors. Developing large cores is an ordeal that can cost into the billions, the market has to show a strong demand for ARM cores. The ISA isn't the limit to scaling up, in fact its sane front end is a saving grace.
Would Cell be notable without its primary feature of note? Er, no, I guess?
LordOfChaos
Member
How I wish, especially as they dropped the requirement for games to support the touchpad/wand controller. I guess they want to separate the iPad Pro and iPhone 7 as the powerful ones though, Apple has been going for chip appeal a lot of late I find.
If they did start putting A9X in there and more storage, and updated it to the newest A*X every year, the consoles may not need to be afraid but certainly be looking behind their backs...A yearly A*X update cadence could rapidly bridge gaps, though then again the 3 year console update cadence may be in response to devices like it and keep them ahead.
Chû Totoro
Member
I don't know too for real but from what I saw the game doesn't seem that demanding... maybe high res assets are something to take into consideration but that's about it.
But maybe 5S is the minimum because of this (iOS Metal feature sets)
Which GPU Family is it?
I'm not a mobile game dev nor a console game dev though.
Toadthemushroom
Member
Chû Totoro;217358109 said:
iPhone 5s was the first consumer device to ship with a Power VR Series 6 GPU if I remember correctly. The iPhones 4s-5 had Series 5 GPUs, with similar architecture to the PS Vita's GPU.
As for Super Mario Run, 5s is probably the minimum to ensure 60fps gameplay. Given it's multiplatform I'm not expecting any low level programming using Metal etc., and the visuals are pretty basic too. Miitomo was built using Cocos3D, I expect something similar with Super Mario Run.
Negotiator
Member
The majority of game code is written in high-level languages (C/C++), just like the 7z benchmark. It's mostly vectorized code that is written in hand-tuned assembly.
And yes, I know that the Cell could beat P4 flops-wise. It was made to be an SIMD beast at the expense of PPE's IPC.
~
People have to understand that this trend of moving CPU tasks to the graphics processor is not something new... it has been there for almost 20 years. So, unless you're really young, you should not be surprised by this trend.
Before 3D accelerators (3DFX Voodoo etc.) became the norm, it was the CPU (more precisely the FPU) that had to render 3D graphics. There was even an era (386SX) when CPUs didn't even have a built-in FPU (it used to be a separate chip).
When GeForce 256 came out (the first "GPU"), it moved T&L from the CPU to the GPU with the advent of the DX7 API. People were sceptical of that, since 3DFX reigned supreme back then and it didn't support T&L. People are now sceptical because only AMD offers proper Async Compute support. Déjà vu, right?
All I'm trying to say is I don't understand why some people think that the CPU must do everything. This isn't consistent with the historical trend I'm describing here.
So, why are people so fussed about physics code being processed by the GPU in this generation of consoles instead of the CPU in the previous generation?
~
Regarding ARM, what's the benefit of developing a large core? ARM became popular because of mobile devices that need small, less power-hungry cores. Contrary to the popular belief, even the Jaguar is not a good fit for tablets (5-10W devices), let alone smartphones (2-3W).
ARM is adopting features (like OoO) that x86 had long time ago and this increases the transistor count. Increasing the complexity always comes with a price.
The CISC vs RISC battle is a moot point these days. Why would the market show a strong preference for ARM cores in the desktop/laptop/server segment? x86-64 cores are already a perfect fit for the job.
LordOfChaos
Member
You posted AMD putting ARM on the backburner as proof ARM isn't good for high end cores, I said high end cores cost a lot to make and the market has to show a strong demand or else few will risk it, that's all. You're going somewhere else with what I said that is rather complicated to get into, but I'd point to the simplified front end of ARM as a boon for developing high end cores while x86 takes increasingly Intel-ian R&D to shove forward. Sure, ARM would have to get more complex lengthy pipelined cores that eat some of its advantage, but the ISA is no barrier to scaling up, is my main point.
Similarly I'm not sure what I said that launched you into the GPU talk - I'm full aware of it, I've programmed for it, and my post itself supported the fact that GPUs are better at doing what Cell was good at for a CPU in the year it was launched. Which was a saving grace for the lackluster RSX, in the end.
Yes, it was SPE heavy. You seem keen to ignore its defining feature. The PPE was nothing special, sure. The idea was to move anything that an SPE would take more time on than a transfer to the SPE. The SPEs were in fact also capable of running C/C++ code, but a copy pasta job obviously wouldn't do all they could.
From the Metro dev,
Negotiator
Member
The GPU part wasn't directed at you.
There are people who still believe that we need faster CPUs to process physics for example, instead of offloading them to the GPU.
I guess they're not really satisfied by the current approach and they'd rather have this instead:
http://www.geek.com/chips/intel-predicts-10ghz-chips-by-2011-564808/
https://www.reddit.com/r/pcmasterrace/comments/2ryhty/in_2000_intel_said_we_would_have_10_ghz/
There are people who still believe that we need faster CPUs to process physics for example, instead of offloading them to the GPU.
I guess they're not really satisfied by the current approach and they'd rather have this instead:
http://www.geek.com/chips/intel-predicts-10ghz-chips-by-2011-564808/
https://www.reddit.com/r/pcmasterrace/comments/2ryhty/in_2000_intel_said_we_would_have_10_ghz/
Can you imagine to have a pocket "console looking" nintendo zelda or mario galaxy?! The big problem is the controller (touch controls <<<<<< buttons and sticks)
LordOfChaos
Member
This feels great, and was reminiscent of what the concept of the NX appears to be, though the issue is still the assumption of how many people have a physical controller like it so most games are hampered by thinking about virtual controls for control depth.
Also app install sizes, I think still a limit? And prices. No one is making AAA games for 6.99 the same as they would for moving the decimal place over.
Personally if Apple TV could run all of the indie games that are on my PS4, I would question keeping my PS4 as it is mostly indie games I'm playing. iCloud saving when playing between phone, iPad and Apple TV would be fantastic.
Hi fidelity games aren't very important. Around 450 million PlayStations sold over 22 years. Over 1 billion iPhones sold over 9 years. And the iPhones are sold at a way higher profit margin.
The new Apple TV was damn near marketed as a direct competitor to dedicated consoles. It fell flat though.