• Hey Guest. Check out your NeoGAF Wrapped 2025 results here!

Panajev - your thoughts on Playstation3

is Playstation3 what you thought it would be from

6 years ago

4 years ago (after Cell announced)
2 years ago
1 year ago




also, what is 35 times 35 :D
 
I too wonder what our tech guru thinks of the PS3 now that the specs are offically announced
 
http://www.gamedreamz.com/index.cgi?m=stories&s=read&id=3642

I'd like his comments on this.

"Here are the facts: Nvidia uses separate Vertex and Pixel Shaders, ATI uses a Unified Shader which does both Vertex and Pixel Shading using the same Shader. When Nvidia calculates performance, it totals both Vertex and Pixel Shaders running at peak efficiency. When ATI calculates performance, it totals the performance of the single unified shader (composed of both Pixel and Vertex shaders) at peak efficiency. So this means that in order to compare ATI's performance to Nvidia's, you would double the rated performance.

PS3:

CPU=0.25 TFlops

GPU=1.80 TFlops for both Pixel and Vertex Shaders

XBOX360:

CPU=0.18 TFlops

GPU=0.88 TFlops for Unified Shaders=1.75 TFlops for combined Pixel and Vertex Shaders

So the PS3 has a total theoretical of about 2 TFlops and XBox 360 has about 1.9 TFlops (counting Pixel and Vertex shaders separately, as thr PS3 does)."

Make sense, but I'd like some expert validation.
 
Gek54 said:
Doubling the number for both shaders on the 360 GPU is totally wrong.

I think the point was, that's what Sony is doing in their stat. So to make them comparable, you need to fudge the the 360 one, or divide sony's by two.

Again - I have no idea. But I'd like a technical explanation in English as to why it's right or wrong.
 
"Here are the facts: Nvidia uses separate Vertex and Pixel Shaders, ATI uses a Unified Shader which does both Vertex and Pixel Shading using the same Shader. When Nvidia calculates performance, it totals both Vertex and Pixel Shaders running at peak efficiency. When ATI calculates performance, it totals the performance of the single unified shader (composed of both Pixel and Vertex shaders) at peak efficiency. So this means that in order to compare ATI's performance to Nvidia's, you would double the rated performance.

PS3:

CPU=0.25 TFlops

GPU=1.80 TFlops for both Pixel and Vertex Shaders

XBOX360:

CPU=0.18 TFlops

GPU=0.88 TFlops for Unified Shaders=1.75 TFlops for combined Pixel and Vertex Shaders

So the PS3 has a total theoretical of about 2 TFlops and XBox 360 has about 1.9 TFlops (counting Pixel and Vertex shaders separately, as thr PS3 does)."

That was quite amusing.
 
GhaleonEB said:
http://www.gamedreamz.com/index.cgi?m=stories&s=read&id=3642

I'd like his comments on this.



Make sense, but I'd like some expert validation.

I am about halfway done with the CPU stuff and there is already 3.25 pages (tons of fluff as well as hopefyllu good stuff :lol) worth of stuff.

I want to do something good so I already asked some support from some friends of mine in terms of fact checking and errors in descriptions and explanations.

I will try to do more than one revision of the document as a complaint that often arises is that my threads go into techie stuff without much introduction to each piece so after finishing one piece I will try to re-read it and add to it more than just a couple of times.

I am not trying to hype my piece, just telling you that I will try to do something hopefully worth your guys' time.
 
Panajev2001a said:
I am about halfway done with the CPU stuff and there is already 3.25 pages (tons of fluff as well as hopefyllu good stuff :lol) worth of stuff.

I want to do something good so I already asked some support from some friends of mine in terms of fact checking and errors in descriptions and explanations.

I will try to do more than one revision of the document as a complaint that often arises is that my threads go into techie stuff without much introduction to each piece so after finishing one piece I will try to re-read it and add to it more than just a couple of times.

I am not trying to hype my piece, just telling you that I will try to do something hopefully worth your guys' time.

Awesome - look forward to it.

BTW - will you cover the 360 in similar fashion?
 
GhaleonEB said:
http://www.gamedreamz.com/index.cgi?m=stories&s=read&id=3642

I'd like his comments on this.



Make sense, but I'd like some expert validation.

One little thing: as far as Shading power the R500 is close to 1 TFLOPS as RSX is close to 2 TFLOPS... not within a freaking mile ;).

They are not 50 GFLOPS, but they are not like 900 GFLOPS either as far as programmable Shading power is concerned (not coutning the FP ops coming from FP blending, swizzling, texture sampling and filtering, etc...).
 
GhaleonEB said:
I think the point was, that's what Sony is doing in their stat. So to make them comparable, you need to fudge the the 360 one, or divide sony's by two.

Again - I have no idea. But I'd like a technical explanation in English as to why it's right or wrong.

Sony's spec is more correct, the .88 Tflops is total theretical performance for the 360 GPU.
 
GhaleonEB said:
Awesome - look forward to it.

BTW - will you cover the 360 in similar fashion?

I cover both: this is not focused only on CELL.

Actually I am starting from the basic core that originated both Xbox 360's CPU and the Broadband Engine's PPE core then delving into the X360 CPU and then mentioning what is different on the PPE.
 
Gek54 said:
Sony's spec is more correct, the .88 Tflops is total theretical performance for the 360 GPU.

48 Shader ALU's * ( 8 FP ops/cycle + 2 FP ops/cycle) * 500 MHz = 240 GFLOPS

There is more to the GPU than Shader ALU's (there is also a certain programmable bit people will like a lot ;)), but it will get counted... do not worry.
 
Thanks for the update, I look foward to reading :D
 
Panajev2001a said:
48 Shader ALU's * ( 8 FP ops/cycle + 2 FP ops/cycle) * 500 MHz = 240 GFLOPS

There is more to the GPU than Shader ALU's (there is also a certain programmable bit people will like a lot ;)), but it will get counted... do not worry.

Thanks - nothing like a steaming pile to get your attention. :)

Is it safe to say then, that NVIDIA kicked ATI's ass pretty soundly this time around?
 
GhaleonEB said:
Thanks - nothing like a steaming pile to get your attention. :)

Is it safe to say then, that NVIDIA kicked ATI's ass pretty soundly this time around?

ATI's part is launching earlier. Given time AND RSX's transistor budget I'd say they'd do a very good job.

OTOH, Nvidia's RSX is a shading MONSTER.
 
GhaleonEB said:
*so....confused.....*

Okay, I'll just wait patiently for Panjev's formal paper. Good to know it's coming. :lol :lol

He illustrated the gap between theoretical "total" power of a GPU and "real" programmable power of a GPU using ATi's part, but the same applies to any chip including RSX. Whether RSX will scale down proportionally, we'll have to wait and see (there's not enough detail out there on it yet, I don't think).

And indeed, I look forward to Pana's thoughts too!
 
GhaleonEB, The estimates and thats all they are at the moment because we don't know all the features each GPU will have indicate that in Shader preformance 360 is at least 2x greater than todays best GPU'S and RSX is at least 2.5x greater. This is my estimate anyway, maybe Pani will say this is rubbish but hey I've tried my best. Aslo its seems to me that CELL once understood will be more flexible thant the tri-core and once the use of unified shaders are understood by developers better the R500 will be more flexible. I think overall the 360 and PS3 are within the same technical time frame (clockspeeds suggest this) with PS3 with the advantage of the extra 3-6 months time frame. What this means in terms of visuals or whatever I don't know. The fanboys can argue over that.
 
Panajev2001a said:
I am about halfway done with the CPU stuff and there is already 3.25 pages (tons of fluff as well as hopefyllu good stuff :lol) worth of stuff.

I want to do something good so I already asked some support from some friends of mine in terms of fact checking and errors in descriptions and explanations.

I will try to do more than one revision of the document as a complaint that often arises is that my threads go into techie stuff without much introduction to each piece so after finishing one piece I will try to re-read it and add to it more than just a couple of times.

I am not trying to hype my piece, just telling you that I will try to do something hopefully worth your guys' time.

Pana, I expect a 5 page report on what you think about me. STAT!
 
Pug said:
GhaleonEB, The estimates and thats all they are at the moment because we don't know all the features each GPU will have indicate that in Shader preformance 360 is at least 2x greater than todays best GPU'S and RSX is at least 2.5x greater. This is my estimate anyway, maybe Pani will say this is rubbish but hey I've tried my best. Aslo its seems to me that CELL once understood will be more flexible thant the tri-core and once the use of unified shaders are understood by developers better the R500 will be more flexible. I think overall the 360 and PS3 are within the same technical time frame (clockspeeds suggest this) with PS3 with the advantage of the extra 3-6 months time frame. What this means in terms of visuals or whatever I don't know. The fanboys can argue over that.

Now THAT I understood. Thank you. :)

The baby is finally sleeping...gonna crash now....
 
YellowAce said:
Pana, I expect a 5 page report on what you think about me. STAT!

Who are you ;) ?

You could be the Square guy who humps GCN's... you could be a pseudonym of a ghei Maylasian-Chinese bi-CMOS engineer... you could be a very hot stack...


;).
 
Izzy said:
ATI's part is lauching earlier. Given time AND RSX's transistor budget I'd say they'd do a very good job.

OTOH, Nvidia's RSX is a shading MONSTER.

I agree with this. It's not a matter of ATI sucking. It's just a matter of time and budget. Sony is willing to fab a half billion transistor monster with next-gen media format to top what Microsoft is already shipping WORLDWIDE this year. Microsoft and ATI did a tremendous job assembling an incredible piece of hardware in time for everyone to enjoy in 2005. To beat this, Kutaragi has no choice than to wait and force Sony spend more money to fab an even more expensive console. I think the advantage the NVidia GPU will have over R500 (and you can bet it will be there) will be determined just from the bigger hit Sony is willing to take (say what you will, I can't believe for a second that a chip shipping later from the likes of NVdia can be outclassed by a chip with HALF (not 15%) the transistor count). Now, if ATI had more time and MS asked them a chip equally expensive, it would be all a different matter...
 
Panajev2001a said:
Who are you ;) ?

You could be the Square guy who humps GCN's... you could be a pseudonym of a ghei Maylasian-Chinese bi-CMOS engineer... you could be a very hot stack...


;).

You could also be Mrs. Charlie or Aika from Team Nishi for all I know... in that case I humbly bow as your humbly servant :lol.
 
Panajev2001a said:
You could also be Mrs. Charlie or Aika from Team Nishi for all I know... in that case I humbly bow as your humbly servant :lol.

Nope, it's that lad who said PS3 would feature 256 MB Main RAM + 64-128 MB VRAM configuration. ;)
 
Endymion, the transister count you keep going on about is only 1 part of a big story. First up do realise that all ATI cards tend have significantly fewer transisters than there Nvidia counter parts? And we know what happened in terms of GPU output, it took Nvidia a fair while to match up to ATI's output. By the way the 150 million transister count of the R500 many think is minus the edram (becasue people are unsure if its embeded or not) if it is embeded (probable) add another 90-100 million transisters to the ATI GPU.
 
Pug said:
Endymion, the transister count you keep going on about is only 1 part of a big story. First up do realise that all ATI cards tend have significantly fewer transisters than there Nvidia counter parts? And we know what happened in terms of GPU output, it took Nvidia a fair while to match up to ATI's output. By the way the 150 million transister count of the R500 many think is minus the edram (becasue people are unsure if its embeded or not) if it is embeded (probable) add another 90-100 million transisters to the ATI GPU.

Which are for RAM and the 2x anti-aliasing MS requires from all developers, right? However, we'll know soon enough ;)

Looking forward to reading what Pana has to say, too.
 
Pug said:
And we know what happened in terms of GPU output, it took Nvidia a fair while to match up to ATI's output.

People keep saying this, but sans-NV3x, the opposite has been true IIRC. Most recently the oppostive has been true (compare 6800 Ultra to ATi's cards coming 6 months later).

Pug said:
By the way the 150 million transister count of the R500 many think is minus the edram (becasue people are unsure if its embeded or not) if it is embeded (probable) add another 90-100 million transisters to the ATI GPU.

I'm fairly sure it does exclude eDram, it must. I've heard estimates in the 50m range for 10MB of eDram, but then there may be some other logic for certain operations..another 50m on top of that, though? I'm not a transistor guru, just curious..

Still, though, that'd be ~300m compared to at most 200m for (mostly?) computational logic, ex eDram. But as Huang said himself, transistors are pretty obscure entities ;)
 
Endymion,Edarm is part of the graphics subset they are still transistors on the GPU, and remember all the transister on the RSX will not be do with shifting shaders etc. How about the Sram etc all that will counted. As I said it still give the RSX an advantage, but until we see all the details of each GPU its hard to give difinitive answers. Oh by the way the 150M Transistors on the R500 is a rummour as far as I'm aware.
 
Gofreak so your saying the ATI GPU has got less than 100 million transistor for logic? And the Edram numbers are a guess but taken from the numbers I've seen you are looking at 90-100 million (B3D) I have seen none at 50, they are guessing that the transistor count in R500 will be around 250M. By the way we still don't know how efficient the 48 shaders are compared to the pipes. There are loads of unknows and what I was trying to say that transistor numbers are not a lock that a higher number means greater output, hence the ATI-Nvidia comparison although the numbers in the RSX do point ot an advantage which I guessed at in my shader numbers in my post. And I did point out that it took a while for Nvidia to catch up, which Gofreak they did have to do. I think people are underplaying the R500, i will leave Pani have the final say because he understands the in and outs much better than I.
 
Pug said:
Gofreak so your saying the ATI GPU has got less than 100 million transistor for logic?

No, 200m for non-edram logic, using your numbers.

Pug said:
And I did point out that it took a while for Nvidia to catch up, which Gofreak they did have to do.

With the NV3x/R3xx, yes. More recently, with the NV4x/R4xx, NVidia took a performance and functionality lead that ATi only just responded to several months later (and only in terms of performance). Before NV3x/R3xx, NVidia almost always held the performance lead over ATi. I think it's unfair to say NVidia are playing catch-up.

Pug said:
I think people are underplaying the R500, i will leave Pani have the final say because he understands the in and outs much better than I.

Well, the problem is a lack of detail on both sides - I'm not sure what conclusions can be reached without more information. But I've seen some ridiculous claims made about RSX v R500 (unfavourable toward RSX) in the last couple of days too.
 
Panajev2001a said:
I am about halfway done with the CPU stuff and there is already 3.25 pages (tons of fluff as well as hopefyllu good stuff :lol) worth of stuff.

I want to do something good so I already asked some support from some friends of mine in terms of fact checking and errors in descriptions and explanations.

I will try to do more than one revision of the document as a complaint that often arises is that my threads go into techie stuff without much introduction to each piece so after finishing one piece I will try to re-read it and add to it more than just a couple of times.

I am not trying to hype my piece, just telling you that I will try to do something hopefully worth your guys' time.



thanks Panajev. we dont expect anything of you man, just like to hear your thoughts on PS3 and how it compares to what you wanted it to be. :)

......and soon, the long road to PS4 begins :lol
 
Someone should explain to me HOW ATI can produce a 150 Mtransistor GPU with 48 ALUs and it has currently on the market a GPU with 160 Mtransistor/s and less than half of R500 shading power.
R500 transistors count press told us is bogus, no doubts about it.
In fact you can't find transistors count data in official specs..
EDIT: 10 mb of edram need at least 85 Mtransistor (1 transistor per stored bit + sense amps and other stuff), edram is not embedded ram on R500 as it sits on another die with ROPs too.
 
Nostromo said:
Someone should explain to me HOW ATI can produce a 150 Mtransistor GPU with 48 ALUs and it has currently on the market a GPU with 160 Mtransistor/s and less than half of R500 shading power.
R500 transistors count press told us is bogus, no doubts about it.
In fact you can't find transistors count data in official specs..

you know, I agree!
 
Nostromo said:
Someone should explain to me HOW ATI can produce a 150 Mtransistor GPU with 48 ALUs and it has currently on the market a GPU with 160 Mtransistor/s and less than half of R500 shading power.

There's probably a little "fat" in PC gpus. If you're focussing on getting as many lean ALUs in there, I think it could be doable. The X850 PE is only 160m transistors, IIRC...if you took that logic, cut out any fat, and poured into ALUs, might it not be possible? Looking at the X850's vertex and pixel shaders, there's a lot going on in there that either isn't necessary or isn't wanted in R500 (e.g. 2 mini-alus in addition to the "normal" ALUs per pipe etc. - if you take it that they're half the transistor budget of a "normal" ALU, then that's 16 ALUs you could "convert" them into right there (or the equivalent of 8 of the R500 variety, I guess)) How many texture units does the R500 have? I'm thinking you may also get some transistor savings going with a pool of units versus seperate pipelines (?)

The distribution of logic is very different.
 
Izzy said:
Nope, it's that lad who said PS3 would feature 256 MB Main RAM + 64-128 MB VRAM configuration. ;)

Uhm, I get lost with the GA nick-name since most of thsoe lads do not use the same ones here... heck, YellowAce...:P.

I am kinda getting an idea now of that person might be now though, thanks :).

Oh I got it now ... it indeed is one of those names I excluded :P.
 
Top Bottom