Don't forget the madness that was hardware BC with Wii.
- Status
I don't buy that. You can get full on smart tablets for £50. A dumb screen with a couple of thumbsticks should not have cost the earth. As Globby says, the main issue was BC. Ditch that and use an OTS part and performance could have been far better without ludicrous pricing. Although I don't think the tech of that period was particularly good. APUs were very weak and discrete parts limited in availability and/or performance. I might start a discussion on what else Wuu could have been.
I see NX as being a Nintendo centric product certainly, but we have seen this succeed with the 3DS. 3DS has been successful without Western third party support, and I see NX as being more of a 3DS successor with the added bonus/selling feature being able to steam to the TV. Honestly, for many people, they didn't need a Wii U to get their Nintendo fix. The 3DS offered Mario Kart, Mario 3D Land, Smash Bros, and Zelda. 3DS was already blurring the lines for Nintendo's portable software compared to console. So while Nintendo certainly needs third party support, I see no reason to believe 3DS developers wont be ready to support NX immediately. Not to mention all of Nintendo's own resources now focus on a single platform, instead of splitting between two.
Nintendo sold something like 125 million DS in 5 years, and 50 million 3DS in 5 years. So 'successful' is relative. A new handheld going up against a stronger-than-ever mobile space doesn't smack of being a strong product. And if the specs don't allow a decent enough TV experience, it might be a jack-of-all-trades, master-of-none device.
Putting it another way, Nintendo got rich off the success (many, many millions of sales) of GB, NES, DS, and Wii. The handheld success accompanied the console success, and if the console failed, the handheld kept Nintendo going. 3DS isn't going to sustain Nintendo, and NX, it's replacement, won't have a second platform that can bolster it. So that NX needs to perform as well as a decent console+handheld combo for Nintendo, or they'll have to start downsizing. I seriously think they need a machine that'll start quickly (15+ million in first year?) and get to 80 million over 5 years, minimum. By comparison, they are living off 60 million N. gamers across 3DS and Wii U and that's two platforms.
Putting it another way, Nintendo got rich off the success (many, many millions of sales) of GB, NES, DS, and Wii. The handheld success accompanied the console success, and if the console failed, the handheld kept Nintendo going. 3DS isn't going to sustain Nintendo, and NX, it's replacement, won't have a second platform that can bolster it. So that NX needs to perform as well as a decent console+handheld combo for Nintendo, or they'll have to start downsizing. I seriously think they need a machine that'll start quickly (15+ million in first year?) and get to 80 million over 5 years, minimum. By comparison, they are living off 60 million N. gamers across 3DS and Wii U and that's two platforms.
I disagree, Nintendo can have the mobile+NX combo, and potentially use mobile as doorway to move people to NX. Hook them with a simple but addictive experience on mobile, and then offer them the real deal complete package on NX. Selling 50 million units in 5 years isn't anything to sneeze at really. I don't know if the PS3 or the 360 did much more than that in their first 5 years. Xbox One is probably tracking for similar numbers. Not to mention software is more profitable than hardware, and I would expect a very high attach ratio for Nintendo's first party games on NX.
So the rumor of the day seems to be that NX actually uses the Tegra X2 (Parker), and will be a lot more powerful than the X1. The real question is this, how powerful can the Tegra X2 feasibly be? We know there is reduced power consumption thanks to the 16nm Finfet process, but just how much headroom will that give them? You can only pack in so many GPU cores before heat/power consumption are too much for a portable unit.
It's not about "what represents a reasonable success in console terms." It's about how much Nintendo needs to sell to be able to remain the size they are.
http://i.imgur.com/G3Y8iCO.jpg
Look where they're at the last 5 years. They need something that'll make decent revenue, not barely sustain them, and 50 million consoles replacing their current 60 million won't do that. Look at where their big profits were, and how that equates to epic market penetration by reaching new audiences with Wii and offering the only real handheld option.
I guess it depends on just how bloated did Nintendo become through those DS/Wii years? Are there a lot more employees today than there were in the year 2005? You can see Nintendo stayed fairly constant for nearly 15 years, and then the boom of the DS and Wii skyrocketed income. I remember Iwata talking about how Nintendo considered doing layoffs in order to prop up the bottom line, but chose against it to keep employee morale up. So Nintendo has been willing to sacrifice short term profits for long term stability.
I am in no way saying that Nintendo wouldn't be in the best position to do well without western third party support, but remain committed to the believe that Nintendo's first party software and potentially third party exclusives will be the driving force, especially early on. If NX jumps out of the gate selling 10 million units in the first 6 months, developers will take a good hard look at NX, even if porting requires a bit more scaling down than they would like.
How well do Arm A57 cores compare to Jaguar cores? Developers seem to have been bottlenecked on PS4/X1 with games like AC4 and The Witcher 3 due to CPU limitations. would four A57 cores actually provide more performance?
I am in no way saying that Nintendo wouldn't be in the best position to do well without western third party support, but remain committed to the believe that Nintendo's first party software and potentially third party exclusives will be the driving force, especially early on. If NX jumps out of the gate selling 10 million units in the first 6 months, developers will take a good hard look at NX, even if porting requires a bit more scaling down than they would like.
How well do Arm A57 cores compare to Jaguar cores? Developers seem to have been bottlenecked on PS4/X1 with games like AC4 and The Witcher 3 due to CPU limitations. would four A57 cores actually provide more performance?
this article gives reference points (iwth Bay trail as a proxy) with older ARM designs. I suspect the lastest ARM cores (A72 and A73) are both faster per cycle and a lot more power efficient than Jaguar cores. They also both clock higher, I suspect the A73 is a bit tinier than the Jaguar too.
A cluster of 4 A73 2.4GHz is likely to behave better than 6 at 1.6GHz, as you can't always trump pure serial performances.
If ARM promises are true, a 4 core A73 @2.4GHz should beat 6 jaguar cores (leaving aside the Jaguar patchworked L2). It is pretty impressive for a CPU that should be significantly tinier (4 vs 6 cores, and a slightly tinier core to begin with) The save silicon would be well spent on some extra L2.
My assumptioon is that a properly design x4 A73 + x4 A35 CPU should beat nowadays consoles CPUs in pretty much all metric: more performances, more performances per Watts and per mm2.
One may wonder about how such a system would look like: x4 A73 x4 A35, 1 GPC, 4 SM, 16 ROPs, 128 BIT bus GDDR5, 6 GB of RAM (with a fucked set-up aka GT 970 but the slow ram part of the OS reservation). I suspect it would blast the leaving hell of the PS4 and XB1 in every metric including production costs. I would be tinier than 240mm2 that is for sure.
A cluster of 4 A73 2.4GHz is likely to behave better than 6 at 1.6GHz, as you can't always trump pure serial performances.
If ARM promises are true, a 4 core A73 @2.4GHz should beat 6 jaguar cores (leaving aside the Jaguar patchworked L2). It is pretty impressive for a CPU that should be significantly tinier (4 vs 6 cores, and a slightly tinier core to begin with) The save silicon would be well spent on some extra L2.
My assumptioon is that a properly design x4 A73 + x4 A35 CPU should beat nowadays consoles CPUs in pretty much all metric: more performances, more performances per Watts and per mm2.
One may wonder about how such a system would look like: x4 A73 x4 A35, 1 GPC, 4 SM, 16 ROPs, 128 BIT bus GDDR5, 6 GB of RAM (with a fucked set-up aka GT 970 but the slow ram part of the OS reservation). I suspect it would blast the leaving hell of the PS4 and XB1 in every metric including production costs. I would be tinier than 240mm2 that is for sure.
Last edited:
Firstly, look at the profits from 1990 to 2005. Even if N. stayed the same size from then to now, the current operating income isn't sufficient to maintain Nintendo of that size as well as generate enough profits and growth that shareholders won't revolt.
Secondly, look at the profits versus revenue. The last five years has seen higher average revenue than 1990 - 2005, but profits are averaging about zero. Clearly it's a lot more expensive to run the company now than 15 years ago.
D
Deleted member 13524
Guest
From the Drive PX2 details, there is some new information about TX2.
Namely, nvidia claims that each PX2 has 2 Tegra SoCs that can do up to 2.5 TFLOPs combined. They don't say it's half precision, so I think we can assume this is 2.5 TFLOPs FP32 among 2 iGPUs and whatever CPU cores they have (first Drive PX2 presentation claimed 8 TFLOPs FP32).
This would result in ~1.25 TFLOPs per TX2 chip. The GFLOPs that the CPU cores can do are rather residual compared to what the GPU can do, since at least the Cortex A53 only does 1 FLOP per core per cycle (as measured by Geekbench). I's say 20 GFLOPs max for the CPU cores (just in case each Denver2 has 2 FPUs), so we're looking at a ~1.23 TFLOPs iGPU.
My guess would be that the TX2's GPU (Parker?) has 4 SMs for 512 cuda cores operating at ~1.2x GHz. Maybe it's really a 2x TX1 GPU in everything else, so 512 cuda cores, 32 TMUs and 32 ROPs. Perhaps 32 ROPs is too much for a SoC so I guess they could have stayed with 16 ROPs or maybe an unusual 24 ROP configuration for an odd number of memory channels like 3*32bit LPDDR4.
Of course, these would be clocks for the Drive PX2 which is a 250W watercooled beast carrying a pair of GP107 chips which are probably around 75W each (so 75W for each SoC too).
There's also that very odd CPU core configuration which doesn't seem to make it very console-friendly, unless either the A57 or the Denver modules aren't accessible to the developers (for example, by having them dedicated to manage an external GPU (e.g. through a dock add-on like suggested by rumors and patents)).
Anyways, a TX2 would be very interesting for the NX if it is indeed a handheld, even if the GPU clocks come down to ~800MHz and a CPU module is disabled. Even more if it is a handheld with an add-on (sold separately) dock that carries an external GPU that would put it close to a PS4 Neo in processing power.
And yes, I know it's not going to happen because this is Nintendo.
Namely, nvidia claims that each PX2 has 2 Tegra SoCs that can do up to 2.5 TFLOPs combined. They don't say it's half precision, so I think we can assume this is 2.5 TFLOPs FP32 among 2 iGPUs and whatever CPU cores they have (first Drive PX2 presentation claimed 8 TFLOPs FP32).
This would result in ~1.25 TFLOPs per TX2 chip. The GFLOPs that the CPU cores can do are rather residual compared to what the GPU can do, since at least the Cortex A53 only does 1 FLOP per core per cycle (as measured by Geekbench). I's say 20 GFLOPs max for the CPU cores (just in case each Denver2 has 2 FPUs), so we're looking at a ~1.23 TFLOPs iGPU.
My guess would be that the TX2's GPU (Parker?) has 4 SMs for 512 cuda cores operating at ~1.2x GHz. Maybe it's really a 2x TX1 GPU in everything else, so 512 cuda cores, 32 TMUs and 32 ROPs. Perhaps 32 ROPs is too much for a SoC so I guess they could have stayed with 16 ROPs or maybe an unusual 24 ROP configuration for an odd number of memory channels like 3*32bit LPDDR4.
Of course, these would be clocks for the Drive PX2 which is a 250W watercooled beast carrying a pair of GP107 chips which are probably around 75W each (so 75W for each SoC too).
There's also that very odd CPU core configuration which doesn't seem to make it very console-friendly, unless either the A57 or the Denver modules aren't accessible to the developers (for example, by having them dedicated to manage an external GPU (e.g. through a dock add-on like suggested by rumors and patents)).
Anyways, a TX2 would be very interesting for the NX if it is indeed a handheld, even if the GPU clocks come down to ~800MHz and a CPU module is disabled. Even more if it is a handheld with an add-on (sold separately) dock that carries an external GPU that would put it close to a PS4 Neo in processing power.
And yes, I know it's not going to happen because this is Nintendo.
That Tegra SoC TFLOP figure though is without considering Pascal GPU capability, unlike TX1 processor (which has 256 Maxwell Cuda cores) it is not integral but discrete with the Drive PX2 Tegra precessor.
So it would in theory be higher as it is CPU 2.5 TFLOPs and discrete Pascal GPU.
Of course it comes down to how much of the Drive PX 2 HW design translates to other Tegra Pascal solutions.
Cheers
Edit:
Look at the Drive PX2 with its Tegra processors in comparison to Jetson TX1 and also Drive PX; latter two have the Maxwell cores integral to the Tegra X1 processor.
So it would in theory be higher as it is CPU 2.5 TFLOPs and discrete Pascal GPU.
Of course it comes down to how much of the Drive PX 2 HW design translates to other Tegra Pascal solutions.
Cheers
Edit:
Look at the Drive PX2 with its Tegra processors in comparison to Jetson TX1 and also Drive PX; latter two have the Maxwell cores integral to the Tegra X1 processor.
Last edited:
No, that's GPU compute flops. You don't get 2.5 teraflops of CPU processing outside of massive Itanium-based supercomputers!
Not sure I follow as we are just talking about the values Nvidia reports.
And they give 2.5 Teraflops for the Dual next gen Tegra chips and 5 TFLOPs for the discrete dual Pascal GPU in the PX 2.
This is where we are coming from when discussing those values and my response to ToTTenTranz.
Cheers
Isn't that figure counting the flops of the integrated graphics? 2.5TF, I wonder if even the upcoming 22-core Skywell-E Xeon can hit close to that performance with AVX512...
Silent_Buddha
Legend
Maybe this will help.
http://techgage.com/article/intels-skylake-core-i7-6700k-a-performance-look/
An Intel i7-6700k does a bit over 110 GFlops (a bit over 27.5 GFlops per physical core). You're thinking Nvidia has made a CPU core that is well over a magnitude faster than any CPU core Intel has ever created?
Regards,
SB
They might be working on their next title: Luigi; Mansion of the skyvers.
The Tegra chips incorporate GPUs. The dual Tegras provide 2.5 TF processing power (mostly GPU compute). The Pascal GPUs deliver 5 TF and over 24 DL TOPS (whatever they are!). The total Drive PX2 system processing power is 7.5 TF.
http://www.nvidia.com/object/drive-px.html
OK Guys I think you are reading too much into what I am saying.
I should had said processor instead of CPU, my point was to differentiate between the design change of Tegra K1-X1 to that of the Tegra X2.
Tegra X2 is the 1st design to have discrete GPUs where the othes with the Kepler and Maxwell cores were integral to the SoC processor and not able to use discrete.
Check the design of the Drive PX (before PX 2) or the Jetson TX1.
Hence why I am saying it has more power than the 2.5TFlops that some are basing the power possibly of Nintendo NX because that is only the dual X2 Tegra processors.
This is further compounded by the design options of the Drive PX2 that will not necessarily be fully comparable to other next gen Tegra designs; it does not need to be 2xnext gen processors and depends upon flexibility of how the GPU architecture connects with the Denver cores.
Sorry for the confusion as used CPU just to differentiate between previous designs as there are fundamental aspects only applicable to next gen such as discrete GPUs, would had been better to say just processor.
Edit.
It is also worth noting that Nvidia stated the dual TX1 had 2.3 Teraflops of performance in the Drive PX.
Cheers
I should had said processor instead of CPU, my point was to differentiate between the design change of Tegra K1-X1 to that of the Tegra X2.
Tegra X2 is the 1st design to have discrete GPUs where the othes with the Kepler and Maxwell cores were integral to the SoC processor and not able to use discrete.
Check the design of the Drive PX (before PX 2) or the Jetson TX1.
Hence why I am saying it has more power than the 2.5TFlops that some are basing the power possibly of Nintendo NX because that is only the dual X2 Tegra processors.
This is further compounded by the design options of the Drive PX2 that will not necessarily be fully comparable to other next gen Tegra designs; it does not need to be 2xnext gen processors and depends upon flexibility of how the GPU architecture connects with the Denver cores.
Sorry for the confusion as used CPU just to differentiate between previous designs as there are fundamental aspects only applicable to next gen such as discrete GPUs, would had been better to say just processor.
Edit.
It is also worth noting that Nvidia stated the dual TX1 had 2.3 Teraflops of performance in the Drive PX.
Cheers
Last edited:
I'm still confused. 
No-one's saying NX is going to use Drive PX2. The idea is NX can use a single Tegra SOC. Drive PX2 has two of these with a combined TF os 2.5. Ergo, one of these is 1.25 TF which would be the ballpark for NX, no? Are you suggesting a discrete GPU coupled to this is NX? How'd that work in a handheld?
No-one's saying NX is going to use Drive PX2. The idea is NX can use a single Tegra SOC. Drive PX2 has two of these with a combined TF os 2.5. Ergo, one of these is 1.25 TF which would be the ballpark for NX, no? Are you suggesting a discrete GPU coupled to this is NX? How'd that work in a handheld?I am not sure how Nvidia work this out because they also use ISPs-DSPs for part of the image classification and not just traditional GPU cores.
But this is Nvidia so I do not equate their figures to that of literally
To emphasise why I feel the discrete GPU is critical in the next gen Tegra design here is an important figure that I added as an edit earlier but will be missed by some as it was so late.
The Drive PX has 2.3 Teraflops of performance with its dual TX1 processors.
The Drive PX2 has 2.5 Teraflops of performance with its dual next gen Tegra processors.
That is not much of an improvement if one also does not consider that latest next gen Tegra Drive PX2 is designed around discrete GPUs.
Cheers
- Status
Similar threads
