prag16
Banned
"Magnitude" as in "order of magnitude? Do you understand what that means? Let alone the plural form?
You're essentially stating that a single xbone/PS4 core is at least 100x as strong as a single espresso core...
-
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.
Wii U CPU |Espresso| Die Photo - Courtesy of Chipworks
- Thread starter Fourth Storm
- Start date
"Magnitude" as in "order of magnitude? Do you understand what that means? Let alone the plural form?
You're essentially stating that a single xbone/PS4 core is at least 100x as strong as a single espresso core...
Locuza
Member
No, not as an "order of magnitude".
More of a size thinking.
To be rough:
One Jaguar Core is >30% higher clocked, each execution Pipe (Integer/Floating-Point) is twice as wide compared to espresso.
The core architecture of espresso still corresponds to the 90s, where as Jaguar is an modern architecture from many points of view.
So one Jaguar Core should be factors faster.
More of a size thinking.
To be rough:
One Jaguar Core is >30% higher clocked, each execution Pipe (Integer/Floating-Point) is twice as wide compared to espresso.
The core architecture of espresso still corresponds to the 90s, where as Jaguar is an modern architecture from many points of view.
So one Jaguar Core should be factors faster.
Corresponds to the 90s? What's that supposed to mean? You do know the x86 architecture, what the Jaguar uses, is even older right?
This doesn't answer my question at all really.
Please explain this matter of execution pipe being twice as wide, because according to the documents that were uncovered not to long ago, Espresso has boost in floating point capability as well as some other things compared to Broadway and the core architecture always had exceptionally good integer performance compared to most other CPU architectures.
Locuza
Member
I mean the core architecture, not the instruction set like x86 or PowerPC.
According to Héctor Martín (marcan42) the cores are nearly the same.
No major change from the one found in the gamecube or wii.
What is differrent and complety new, is the smp glue between the three cores and the L2-Cache.
The rest seems to be untouched.
I don't read the last pages, so which documents do you mean?
The execution pipes on Jaguar are 64-Bit wide on the Integer side and 128-Bit on the Floating-Point side.
On the paper the throughput is twice as high.
According to Héctor Martín (marcan42) the cores are nearly the same.
No major change from the one found in the gamecube or wii.
What is differrent and complety new, is the smp glue between the three cores and the L2-Cache.
The rest seems to be untouched.
I don't read the last pages, so which documents do you mean?
The execution pipes on Jaguar are 64-Bit wide on the Integer side and 128-Bit on the Floating-Point side.
On the paper the throughput is twice as high.
PPC750 is an extremely lightweight and simple CPU, there simply isn't much you could improve or change.
Also, 64bit CPUs aren't any faster than 32bit CPUs unless you actually need to work on 64bit integers - which is almost never. The reason amd64 chips are often a bit faster than x86 chips isn't because it's 64bit, it's because AMD also doubled the number of registers as part of the amd64 specification, and x86 always was a register starved architecture. Still is, compared to PowerPC. Espresso, like any ppc, has twice as many general purpose and floating point registers as Jaguar, and four times as many as any x86.
hi there, i dont have any clue or knowledge about chipsets and technical stuff, but we have smartphones, which outperform more or less the wii u already.
they cost nearly the same, but what iam wondering ist...
why is the wii u so gigantic???
is there a special reason to it?
its basically "comparable" to ps360, maybe a bit better with an advantage of having off-play and a second screen.
but still iam in awe that the wii u is so huge...
why is it not as big as a normal wii?
they cost nearly the same, but what iam wondering ist...
why is the wii u so gigantic???
is there a special reason to it?
its basically "comparable" to ps360, maybe a bit better with an advantage of having off-play and a second screen.
but still iam in awe that the wii u is so huge...
why is it not as big as a normal wii?
No, there are no smartphone chipsets that outperform a Wii U. There's a very simple explanation for that, and it's the same reason the system is bigger than a Wii: It consumes a lot more power, produces more heat, and therefore requires a more massive cooling solution.
the facts on the paper look so much more compelling... with i dont know how many cores, and at least 1gb ram already in a smartphone... thats why i feel like, the wii u is just too big for what it is... especially coming from the OG wii. which was a fucking master piece of a design.
lightchris
Member
Well the Wii U consumes >60% more power than the original Wii, so it's only natural that the case needs to be bigger to achieve similar kind of noise levels etc.
You're not all wrong about smartphones outperforming Wii U though, maybe not yet but in a year or so. If Nvidia's Logan will keep half of what it promises that might be enough. The entire comparison isn't really fair though since new high end smartphones/tablets are a lot more expensive than Wii U ever was.
prag16
Banned
Heh, best not to hold our breath there. Tegra 2, 3, and 4 all overpromised and underdelivered to varying degrees...
As for the "magnitudes" guy, I'm not even sure a single PS4 core is more than double a single Espresso core. Maybe, but I'd guess not, certainly not for all workloads. These are netbook cores after all I guess it'd be nice to see benchmarks. That'd be extremely interesting.
If only "on paper" always equaled reality then life would be much easier, but it rarely works like that. On paper you'd think Bobcat (Jaguars predecessor) would be faster than Espresso. In reality Espresso is shown to be significantly faster than Bobcat.
I really doubt Jaguar cores are twice as fast as Bobcat let alone Espresso.
prag16
Banned
I can't remember now. Did we have benchmarks? Was it 750CL vs. Bobcat or something? I seem to vaguely remember something. Maybe I'll dig through the thread later.
The WiiU is the same design as the Wii with an optical drive in the front, a heatsink behind it and a fan behind that. It isn't very efficient, but cheap and works. Without the optical drive it could easilly be the size of the Ouya with a better cooling solution.
Since we know that Iwata said that they are using a single architecture for their next handheld and home console, what are the chances it will be based on the GC-Wii-WiiU processor line?
Note: single architecture refers to the fact that there will be just one OS, one set of tools, etc., not that there will only be one hybrid device. There will be a handheld and there will be a separate more powerful home console.
Note: single architecture refers to the fact that there will be just one OS, one set of tools, etc., not that there will only be one hybrid device. There will be a handheld and there will be a separate more powerful home console.
Blu did a SIMD heavy benchmark on various CPU's. Two of those were 750CL (identical core to Wii CPU) and Bobcat. 750CL was 35.7% faster clock for clock.
IMO if we go for a worst case scenario and assume Espresso has the same clock for clock performance as 750CL/Broadway and that Jaguar is 50% faster clock for clock than Bobcat then:
Espresso core @1.24Ghz = 100
Jaguar core @ 1.6Ghz = 143
Of course in reality an Espresso core is going to be faster than a Broadway core clock for clock and I really doubt a Jaguar core is quite 50% faster than Bobcat clock for clock.
All of this also depends on the final clock speed of the CPU in consoles such as One and PS4 of course (was it ever confirmed?).
For what it is, Wii U is tiny. It's much smaller than any 360 or PS3 and obviously much smaller than PS4 or X1. It consumes more power and is more powerful than the Wii of course, so it needs to be bigger, but I fail to see how it's huge.
prag16
Banned
Wow, and SIMD is supposed to be a weak point for the 750 (unless I didn't sleep well last night and am getting mixed up). Interesting.
Yeah SIMD was always supposed to be the area where this CPU would struggle the most, but it seems if you use paired singles correctly you can get very respectable performance.
BTW edited my post to make it clear I was referring to per core performance. Obviously One/PS4's CPU's have more cores (3 for games on WiiU vs 5-6 available for games on One/PS4?) which will widen the gap in overall performance.
BTW edited my post to make it clear I was referring to per core performance. Obviously One/PS4's CPU's have more cores (3 for games on WiiU vs 5-6 available for games on One/PS4?) which will widen the gap in overall performance.
prag16
Banned
Off topic, but I'm heavily disappointed that the "xbone" moniker has not stuck better. I see usage of "One" or "the one" or "XB1" tracking way ahead of "xbone" lately.
I'm still gonna say "xbone".
lightchris
Member
So is bobcat with its 64-bit vectors per cycle. Jaguar has them doubled to 128 bit.
In any case synthetic benchmarks shouldn't be translated to real world performance. My guess is that average IPC won't be much different between Espresso and Jaguar cores as long as SIMD isn't involved.
Locuza
Member
And I'm pretty sure that someone benchmarked one espresso core, which is 20% faster per clock, than one haswell core.
Sorry, but this posting is a good reason for me to stay out of this thread.
blu
Wants the largest console games publisher to avoid Nintendo's platforms.
With compelling arguments such as yours, that'd be a very wise course of action.
blu
Wants the largest console games publisher to avoid Nintendo's platforms.
Yes. It's a per-core test* and the workload is constant (that's why I keep such a large iteration count - so it would not finish too quick on the much faster CPUs, and variance in measurements would not become too large). Also, all the data is in L1, so it's basically purely an op- and ALU-throughput test.
* Except for the hyper-threaded results (i.e. Cell), where N threads (2 in the case of Cell) crunch over the same constant workload.
Locuza
Member
I will not believe that such an old architecture could get close or even be above in real life performance to something like Bobcat or even newer Jaguar.
I can roughly compare one old 750 with an Pentium 3 or K7, which should according to specmark comparable performance wise and later on i found some restults looking at bobcat vs. K8/K10, which are really not compelling looking at the FPU-Side, but both designs are new out-of-order architectures, physical register files, advanced mechanism.
Jaguar eliminated the huge weakness in bobcats FPU-Design.
Source:
http://www.macinfo.de/bench/specmark.html
http://www.xtremesystems.org/forums/showthread.php?272541-Bobcat-Core-performance-analysis-Bobcat-vs-K10-vs-K8
Shin Johnpv
Member
You're looking at things wrong. This is why pure paper specs don't always tell the whole story. Not all cores are created equally, Jaguar may have 8 cores, but my 4 core i5 will smoke it with out even trying. Plus things like clock speed are only a valid comparison of power when talking about the same architecture. You had Pentium 4's clocked between 3 - 4 ghz, but there are brand new chips clocked around 1.5 - 2 ghz that would kill those P4s.
blu
Wants the largest console games publisher to avoid Nintendo's platforms.
Well, frankly, it all started as a compiler test - I was testing how well various compilers would handle an old template vector lib of mine. For that I needed a 'reference' speed test, so I wrote a 4x4 matmul via intrinsics for a few architectures I had access to at the time. Surprisingly (or not), some of the late compilers have been able to produce identical results from the latest iteration of the template lib, as from the reference routine. Also, it bears noting that llvm's cortex-a8 and bobcat backends got serious improvements in the scheduling department over the past year, which, in the latter case, I suspect might be ps4-related ; )Ah k. Thanks.
Projectjustice
Banned
I love this thread. Keep up the good work guys. Very educational.
Except the benchmark I mentioned actually exists, that's a fact, they were actually done on this forum.. Facts you don't like the sound of shouldnt be a reason to stay out of a thread. Unless you're intent on ignorance in which case staying away would be best.
I mean your trying to say we should ignore a benchmark which directly compares two CPUs. Yet your suggestion is to try to indirectly compare them based on benchmarks done on different CPUs (the old ppc750 is quite different to the ppc750cl). Its completely baffling logic.
phosphor112
Banned
Wii U isn't that much smaller than a PS4. Biggest difference is the width of both, by 4 inches.
(W x L x H)
Wii U: 6.8" x 10.6" x 1.8"
PS4: 10.8" x 12" x 2"
That's nearly half the volume for Wii U versus PS4. It's significantly smaller.
ArchangelWest
Member
I'm not getting one any time soon, but I agree with the MS exec who said that the name is disrespectful. It just opens the door to way too much disparagement. They reversed many of the policies that people were upset about. Now hopefully the product can just stand on its own and people won't go into it with pre-made opinions as much based on the nickname.
phosphor112
Banned
sörine;95059564 said:
Yeah, it's twice the volume, but that doesn't mean much since both are small anyway. Excluding the fact that the PS4 has a hard drive and active cooling, the PSU is inside the PS4 while the brick is external on the Wii U.
EDIT: Beaten.
prag16
Banned
xbox one -> x b 1 -> x b one -> xbone
It's not "Disrespectful." It's nearly an exact transcription of its abbreviation.
EDIT additionally any level of implied disrespect is pretty tame compared to what went on after the "Wii" name was announced...
Sure, it was just weird to say those dimensions weren't much smaller/different when they are pretty significantly. One is half of the other.
Wow! I knew that the Nintendo console and handheld divisions were being consolidated, but where did Iwata say that, that's brilliant.
ARM64 is just about the only candidate I can think of that might serve both purposes. Nintendo could use custom versions of the ARM64 APUs AMD is developing for a new generation of powerful low-wattage devices. Nintendo have been using ARM processors from GBA to 3DS, so adoption would be swift. Developer support would be assured because the handheld games could be easily scaled up to higher console spec, much like PC games. This sort of cross-compatibility, to me, is much more compelling than the backwards variety.
IBM processors are best suited to servers, and while power consumption is coming down for x86, it might not come down to where Nintendo needs it for a handheld.
That said, if they can pull it off, a PowerPC handheld would be very Nintendo.
phosphor112
Banned
sörine;95113129 said:
Well, when the majority of the difference is the HDD and the internal PSU, I think it's worth mentioning. =P
Hero of Legend
Member
May I recommend that the OP should be updated with a summary of the discoveries and comparisons of the CPU. Because I don't know if I want to dig through every page to get an idea what kind of capabilities were found and I'm curious of comparisons between PS360 and PS4/XBONE.
And make it easy to read because not everyone is a tech savvy.
And make it easy to read because not everyone is a tech savvy.
It isn't actually, adding the volume of the Wii U power brick and a standard 2.5" HDD nets under 30% additional to Wii U's dimensions. But even if it did double Wii U's volume that wasn't what you said anyway.
phosphor112
Banned
sörine;95156221 said:
Extra PCB and I/O for HDD? Active cooling which is required for internal PSU's? Yeah, it would easily make up for most of the difference.
Internal PSUs don't need any more cooling. The Xbox 360 and One power bricks both have a fan because of the wattage.
If you think a small circuit board and port would take up as much space as 70% of Wii U's volume then more power to you. And not all internal power sources demand active cooling.
phosphor112
Banned
When you put a heat source next to other heat sources... then yes you do need a fan. How is this debatable? We are talking about games consoles not cable boxes or DVD players.
Probably, wouldn't be suprised if Espresso still has an edge in average IPC with Jaguar pulling ahead in SIMD heavy calculations.
No single benchmark is a definitive comparison. But the fact is what I posted is the closest comparison we have between Espresso and Bobcat and its a benchmark which shouldn't really favour Espresso either according to conventional wisdom. Using a CPU which is very close to a downclocked WiiU core in a direct comparion to an actual Bobcat core. Its head and shoulders a better comparison than what you suggested. So I don't see why my post was worthy of such outrage or sarcasm.
Would be interesting for someone to perform Blu's test on an actual Jaguar CPU, they are out there now in APU's right?
freezamite
Banned
So what you're saying is that you've entered this thread without reading even the first post and yet you've had enough time to publish insulting messages to other people that actually cares about this.
Look, no one is going to answer to this bs you've posted here because no one wants to lose time explaining what's been explained dozens of times.
If you want to participate on this thread, do it with a logical approach, but this is no place for console wars like the one you're desperately trying to start.
It's probably not all that interesting, but after Marcan confirmed that the Espresso bus is called 60Xe, I googled a bit, and I believe those two patents are related to the Wii U:
VERIFYING DATA RECEIVED OUT-OF-ORDER FROM A BUS
VARIABLE LATENCY MEMORY DELAY IMPLEMENTATION
One of the patents specifically mentions the 60Xe bus, which is apparently only used by Espresso, both patents were filed around the same time, and at least one of the engineers responsible for both inventions, Victor Acuña, definitely worked on Espresso.
PowerPC 750CL (1GHz): 138.7
PowerPC 750GX (1GHz): 173.6
Quite a difference, no? But wait, it gets better: the 750CL is newer than the GX, and actually faster in many benchmarks, yet it's quite a bit slower in this particular benchmark. Why? Because the GX had more cache, slightly faster RAM and a faster bus, which can make a huge difference in overall performance. So even though the CPU core itself is slower, it still scores higher thanks to other factors. This demonstrates two things: a) Not all 750s are equal, and b) a lot of things factor into benchmark scores.
VERIFYING DATA RECEIVED OUT-OF-ORDER FROM A BUS
VARIABLE LATENCY MEMORY DELAY IMPLEMENTATION
One of the patents specifically mentions the 60Xe bus, which is apparently only used by Espresso, both patents were filed around the same time, and at least one of the engineers responsible for both inventions, Victor Acuña, definitely worked on Espresso.
As already mentioned, the PPC750 in this benchmark was one of the first revisions. There are no SPEC scores for new cores, so allow me to use certified EEMBC DENmark scores to illustrate something:
PowerPC 750CL (1GHz): 138.7
PowerPC 750GX (1GHz): 173.6
Quite a difference, no? But wait, it gets better: the 750CL is newer than the GX, and actually faster in many benchmarks, yet it's quite a bit slower in this particular benchmark. Why? Because the GX had more cache, slightly faster RAM and a faster bus, which can make a huge difference in overall performance. So even though the CPU core itself is slower, it still scores higher thanks to other factors. This demonstrates two things: a) Not all 750s are equal, and b) a lot of things factor into benchmark scores.