According to
some awful infographic from experts-exchange.com, the 1.78MHz NES can do 8 million floating point operations per second.
Nice, that's 4 floating point operations per cycle! Also I have absolutely no clue how one gets a single floating point computation out of the PPU.
43110 wrote:
Nice, that's 4 floating point operations per cycle!
The Genesis apparently also does... (
30 million instructions per second)
"experts"exchange.com credibility = 0
If I'm reading
this correctly, even with both the Mega CD and 32X attached it doesn't reach 30 MIPS, never mind MFLOPS as the chart says...
I'm not entirely sure how they calculated that the Apollo Guidance Computer was twice as powerful as the NES's 6502, unless they're going by ram which is a terrible thing to base it on. It was apparently capable of 16 bit operations, including multiplication and division, but the clock frequency was also apparently divided by 2 for the thing to actually use for only 1.024 Mhz. I cannot find how many cycles each instruction uses, but so far it seems like both processors are about tied, but they are of course better suited for what they are intended to be used in.
93143 wrote:
If I'm reading
this correctly, even with both the Mega CD and 32X attached it doesn't reach 30 MIPS, never mind MFLOPS as the chart says...
Technically 32X alone would reach 30 MIPS (since SH2 instructions are single cycle and it has two of them at around 23MHz).
But yeah, the amount of bullshit can't be understated.
Espozo wrote:
I'm not entirely sure how they calculated that the Apollo Guidance Computer was twice as powerful as the NES's 6502, unless they're going by ram which is a terrible thing to base it on.
2MHz vs 1MHz. Apparently (because NES is actually closer to 2MHz than 1MHz).
Basically completely bogus numbers and such. Bonus points for claiming at the beginning that they actually account for this stuff!
EDIT: find a font that renders tildes as something other than a dash =|
Sik wrote:
Technically 32X alone would reach 30 MIPS (since SH2 instructions are single cycle and it has two of them at around 23MHz).
Replace "32X" with "SVP chip", then.
What is the difference between "issue cycles" and "latency cycles" on that datasheet? I took a pessimistic approach and added them...
Sik wrote:
2MHz vs 1MHz. Apparently (because NES is actually closer to 2MHz than 1MHz).
They were trying to claim that the Apollo Guidance Computer was twice as powerful as the NES, not that the NES was twice as powerful.
Sik wrote:
Basically completely bogus numbers and such.
Pretty much.
Sik wrote:
Bonus points for claiming at the beginning that they actually account for this stuff!
"experts"echange.com credibility = -100
I just noticed something: on the graph measuring cpu speed, the N64 is only a hair above the playstation. Isn't the N64's cpu about 3x as powerful, seeing that they are both MIPS processors and the N64's is clocked about 3x as fast, not including that it can do 64 bit operations?
Espozo wrote:
Sik wrote:
2MHz vs 1MHz. Apparently (because NES is actually closer to 2MHz than 1MHz).
They were trying to claim that the Apollo Guidance Computer was twice as powerful as the NES, not that the NES was twice as powerful.
No, they claimed Apollo was 2MHz and NES was 1MHz... they couldn't even get those numbers right =P (since the NES clock speed is almost 2MHz)
The graphic clearly says 1.8 MHz, not 1.
http://i.imgur.com/O0KRHUO.jpgI think they're probably just thinking that the speed is equivalent but the RAM is doubled, and that these two factors are somehow multiplicative w.r.t. processing power.
Edit: You know what, they're probably not that dumb. I think Speed and RAM are just the only numerical statistics they could find to stick on the chart, and its probably not representing their entire methodology. Apollo as roughly 2x an NES seems like a decent estimate after thinking more about it.
Holy god, this is some wacky bullshit. But with a coprocessor in the cartridge, it would be possible for the NES to do 8 million floating points per second
EDIT : Or maybe they meant 8 thousand? That would give 223 cycles per floating point operation, probably below the truth but reasonable if they're optimized for speed somehow.
Espozo wrote:
I'm not entirely sure how they calculated that the Apollo Guidance Computer was twice as powerful as the NES's 6502, unless they're going by ram which is a terrible thing to base it on.
rainwarrior wrote:
I think they're probably just thinking that the speed is equivalent but the RAM is doubled, and that these two factors are somehow multiplicative w.r.t. processing power.
I really don't see how an old computer from 1966 that wasn't really a super computer but probably still high end would be twice as powerful than a consumer one from 1983. (Famicom) The AGC doesn't even run at 2MHz, it gets cut in half... (I'm assuming it would be like if you said the SNES's 65816 was running at 21MHz, because it hasn't gone through the divider yet like the 2MHz AGC.)
I still think the PS being that close to the N64 is a bunch of BS. (See what I did there?)
From what I've seen of the website, it looks weird... I looked at the "technical" thing and the categories where like: Microsoft Excel, Windows XP, JavaScript. I bet half of the people have never seen any sort of assembly language. (They couldn't have if they think a 1.8 MHz processor is capable of 8 million instructions...)
Espozo wrote:
(They couldn't have if they think a 1.8 MHz processor is capable of 8 million instructions...)
A modern superscalar one could do it. Of course the 6502 is not one of those... and the claim wasn't IPS; it was FLOPS, which is much worse...
Espozo wrote:
"experts"exchange.com credibility = 0
Ever since Stack Overflow and Super User came out, there has been little need for Expert S-ex Change.
How many flops came out of LJN and Active?
Espozo wrote:
I really don't see how an old computer from 1966 that wasn't really a super computer but probably still high end would be twice as powerful than a consumer one from 1983. (Famicom) The AGC doesn't even run at 2MHz, it gets cut in half...
The first thirty or forty years of computers spent almost all of Moore's law on becoming usable (smaller, lower power), not faster.
Anyway, since people are reacting to people's reactions, here's the relevant section of graphic, with a little bit of annotation:
Attachment:
mflooooops.png [ 46.38 KiB | Viewed 2123 times ]
More than anything else, what I find confusing is that, in reality the 2600 is either 67% the speed of the NES (in the unlikely event that you're just talking about ability to do math, and ignore drawing the screen), or somewhere closer to 20% the speed of the NES (2600: computation during vblank for ~70 scanlines at 76cy per scanline; NES: computation during active screen for ~240 scanlines at ~114cy per scanline) ... either way, not the very clear 50% they depict.
tepples wrote:
Ever since Stack Overflow and Super User came out, there has been little need for Expert S-ex Change.
I think this person would beg to differ:
Your link is broken (403 Forbidden).
thefox wrote:
Still broken.
Grr... I was just trying to make a joke on how tepples said Expert S-ex Change. Well, that didn't work., it was just meant to be something stupid. It was a picture of before and after people had undergone a sex change operation.
I feel stupid.
Works fine here. The second one, that is. The first one just says "Image".
93143 wrote:
Works fine here. The second one, that is. The first one just says "Image".
Yeah, when I checked earlier the 2nd one also displayed as "Image", but it worked later on for some reason. Maybe slow/choppy server.
Wait, are they trying to compare processor power by speed and RAM?
Uh. Sounds like a terrible idea.
Like, from what I've seen, the Z80 is good with 16-bit calculation because its 8-bit registers can be combined to create 16 bits. The 6502 doesn't seem to have this capability, yet can probably do the 16-bit operations at the same speed.
Due to the fact that they're basing it off overall speed, however, there would be a complete bias and no actual analysis of what processor can do what.
From a high level, they're trying to demonstrate the growth of computing power over the years. Even though their methodology is awful on a detailed level, the error is of a small enough order that it doesn't overpower the overall trend, which is the whole point of the article anyway.
It really doesn't take a rocket scientist to figure that computers have become more powerful over the years. What is more important is to see how much more powerful they have become, and in that case, it seems that they have screwed up. Maybe they should have detailed like what lidnariq said about how computers at first where mainly focused on becoming smaller and more useable instead of processing power. I just feel like on something like that that is solely based on information, they really shouldn't screw it up. That'd be like a science experiment where I was trying to see how plants would grow based on how much salt was in their water. It'd be like if the plant with the salt water failed to grow at all and the one with water grew 3 ft. tall, but I messed up the statistics and said that the plant with the saltwater grew 2 and a half feet. The outcome would have been the same in that the plant with the saltwater wouldn't have grown as high, but it's still incorrect and kind of defeats the purpose of the experiment.
My entire point was that the magnitude of error is relatively low overall, and your response was comparing it to a stupid hypothetical example with an error of 83%. There's a difference between a little bit wrong and a lot wrong.
The majority of the computers compared have published FLOPS statistics that I think are reasonably accurate, and a fair comparison. The extension down into the last range that doesn't have floating point hardware is no longer a fair comparison, but they made some attempt to compare that isn't entirely meaningless.
On the local scale, the error is high, but so is the ability to compare these things. You want to spend 200 pages discussing whether you can say that SNES was more powerful than the Genesis; this infographic is deliberately simplifying all of that, because their goal is to demonstrate the overall trend, not bury us in the details of measurement. In this graph we have a bar that says the Genesis is roughly 4x more powerful than the NES. An actual numerical comparison is unfair and impossible, but 4x is a plausible estimate.
The wider you start comparing, the smaller the magnitude of error. If you go all the way to the bottom, you're comparing 6,000,000 for the NES to 300,000,000,000,000 for the Milky Way 2. Even if the number for the NES was off by a factor of 100, the error relative to that scale would still only be something like 0.0002%. Everything on the graph is in the correct order, and roughly the correct scale. From the broad viewpoint of the article's purpose, I think it's fine.
Also, I'm going to take back what I suggested about the Apollo vs NES image earlier. They're probably not so dumb as to compare Speed and RAM in that way. I think Speed and RAM are just the only numerical statistics they could find to stick on the chart, and its probably not representing their entire methodology. Apollo as roughly 2x an NES seems like a decent estimate after thinking more about it.
rainwarrior wrote:
My entire point was that the magnitude of error is relatively low overall, and your response was comparing it to a stupid hypothetical example with an error of 83%. There's a difference between a little bit wrong and a lot wrong.
Sorry?
rainwarrior wrote:
Also, I'm going to take back what I suggested about the Apollo vs NES image earlier. They're probably not so dumb as to compare Speed and RAM in that way. I think Speed and RAM are just the only numerical statistics they could find to stick on the chart, and its probably not representing their entire methodology. Apollo as roughly 2x an NES seems like a decent estimate after thinking more about it.
I don't think that their methodology really spans beyond clockspeed and ram if they could possibly think you could do 4 instructions in one clock cycle...
I stumbled across this nice whitepaper from Intel yesterday comparing trends in computation per energy:
http://download.intel.com/pressroom/pdf ... elease.pdf
nicklausw wrote:
Like, from what I've seen, the Z80 is good with 16-bit calculation because its 8-bit registers can be combined to create 16 bits. The 6502 doesn't seem to have this capability, yet can probably do the 16-bit operations at the same speed.
The Z80 is actually pretty terrible at 16-bit operations since what you can do with them is quite limited and not that faster (sometimes even
slower than doing the same with multiple 8-bit operations), it's not like the 68000 that can do nearly everything on 32-bit operands except the most complex operations. The 16-bit support on the Z80 is better for manipulating pointers than for doing math.
Espozo wrote:
I don't think that their methodology really spans beyond clockspeed and ram if they could possibly think you could do 4 instructions in one clock cycle...
Depends on whether "GetNextPixel" is considered an instruction. You get three of those on the PPU for every ultraslow CPU cycle (and all 2A03 cycles are ultraslow).