Console that fared best vs PC's available at launch

[Mintmaster]

I think the original xbox faired much better than the 360.

When xbox launched it had a GPU which was arguably superior to the fastest PC GPU of the time (GF3 Ti500). But when the 360 launched it would have had to compete with dual 7800GTX 512MB's.

Considering a single 7800GTX 512MB has a fairly significant leg up on RSX which itself competes well with Xenos, it seems to me that Xenos was seriously overshadowed by the best that was available for the PC at the time.

As I noted with my previous post, it's not all about hardware. You're also assuming console workloads are the same as on the PC, but they're not. Finally, we don't know how well RSX keeps up with Xenos, as there are definately some areas where Xenos excels by more than a factor of two. The key to cross platform equality is avoiding/limiting those situations, and we'll never know whether that results in a poorer looking game or not.

A year later and the X1950XTX went well beyond Xenos performance without resorting to dual GPUs. I'm assuming Xenos's performance in PC terms is around HD2600XT + 25% level.

For pixel shader limite workloads only, the former is true. There are other areas where Xenos would fare much better. The latter is a rather silly comparison, as we've discussed before I think. RV630, for whatever reason, doesn't perform as well as its specs suggest, and the architectures are too different anyway.

I doubt Xenon is even remotely near the performance of a modern quad core. Competitive with the slower dual cores certainly but quad cores? I find that extremely unlikely.

When you look at the outlandish claims of PS3 performance being a factor of 2-5 times faster than quad cores, a claim of parity for 360 isn't that unreasonable, especially when you're talking about code tuned to a console.

In any case, the CPU isn't as important for gaming. I'm not saying that you can stick a PDA's ARM processor with Xenos, but it's far, far easier to spot a 2x speed difference in the GPU than the CPU when it comes to games.

 

[pjbliverpool]

You're also assuming console workloads are the same as on the PC, but they're not.

When running a game they are. In what way is the workload of CoD4 different on the PC to a console. Both the graphics and the gameplay (and the rendering process for that matter) are almost identical.

Finally, we don't know how well RSX keeps up with Xenos, as there are definately some areas where Xenos excels by more than a factor of two.

We know it keeps up well enough to make the majorty of games indistinguishable between the two platforms. And with twice the local memory bandwidth, twice the ROPs, twice the actual VRAM and a 10% boost in clock speed it would obviously fare even better.

For pixel shader limite workloads only, the former is true.

Only? What about fill rate, multisampled fill rate, texturing throughput, VRAM size, texture memory bandwidth, setup limit (not too sure on that one)...

Seems to me that the R580 is faster in pretty muich every way. Sure, Xenos wins when it comes to vertex processing but how many games actually stress the vertex processing capabilities of R580? I'm guessing pixel shader and texture limited situations are far more common. If not, then the R580 would have been a very unbalanced design, not to mention the fact that we would be seeing chips like the HD2600 outrunning it on a regular basis (when in fact they almost universally fail to keep up with lesser R5xx based GPU's).

There is of course the edram. There's 4x the bandwidth within the edram compared to R580 but there's only 1/2 to the daughter die itself. And of course there's the lack of compression which scews things.

There are other areas where Xenos would fare much better.

As stated above i'm only seeing vertex processing limited situations and possibly some situations that rely on the edrams bandwidth. Although such situations would also place Xenos above an 9800GX2 which is a good demonstration of why its not simply a matter of comparing GB/s.

The latter is a rather silly comparison, as we've discussed before I think. RV630, for whatever reason, doesn't perform as well as its specs suggest,

Were as Xenos does? I'm sorry but the only measurable evidence I have ever seen of Xenos's performance is that of its direct decendants, the R6xx family. The architectures aren't completely divergent. One is based on the other and any differences between them should obviously be in R6xx's favour. Logically if R6xx performs below what we would expect based on its specs then so should Xenos. Unless of course they changed something about R6xx which drastically reduced the performance of the architecture. Which would be quite absurd.

For example, on paper RV630 has 80% the texturing capability of Xenos. Is there any reason to believe its texture units are weaker? It has 90% the shader power, and we know that the R6xx shader design is more efficient.

When you look at the outlandish claims of PS3 performance being a factor of 2-5 times faster than quad cores, a claim of parity for 360 isn't that unreasonable, especially when you're talking about code tuned to a console.

Cell may be faster in very specific workloads but that hardly translates into being generally 2-5x faster than a quad core or even a dual core. In fact i'm sure there would be plenty of situations were it would be much slower than even a single core.

Also, I don't see how Cells achievements compared to a PC CPU have any baring on Xenon? Cells performance comes from the SPU's which Xenon lacks. Its not as if the PPE in Cell is heralded and a performance monster.

Architecturaly a PPE has nothing on a Conroe core, its completely outclassed in almost every way; issue width, cache, number of execution units, OoOE etc... So comparing 4 Conroe cores (or Penryn for that matter) to 3 PPE's seems like a no brainer to me in terms of which one would come out on top.

 

[Megadrive1988]

Definitally not Xbox 360 or PS3.



It's tough to pick between PlayStation1 (1994) and Dreamcast (1998). Both consoles were ahead of the most expensive, highest-end PCs and PC graphics cards of the time.

I don't believe the PC had any game that could compete with Ridge Racer graphically at the time of the PS1's launch in Japan.

True.

 

[Mintmaster]

When running a game they are. In what way is the workload of CoD4 different on the PC to a console. Both the graphics and the gameplay (and the rendering process for that matter) are almost identical.

COD4 didn't show G71 in the best light, from what I remember.
http://www.firingsquad.com/hardware/call_of_duty_4_demo_performance/page3.asp
http://www.gamespot.com/features/6183967/p-4.html

There's a good chance that COD4 has a more console-like workload than PC games in the heyday of G71. It looks like the unified GPUs have a notable leg up on the older ones (if you recall R600 often has marginal gains over R580, as expected from the specs, but not here). You'll also notice the 7900GTX 512MB performing slower than the 2900XT 256MB, in case you still really believe your analogy.

Just because the PS3 and 360 versions of COD4 are both running at 60 fps doesn't mean one GPU isn't outperforming the other.

We know it keeps up well enough to make the majorty of games indistinguishable between the two platforms. And with twice the local memory bandwidth, twice the ROPs, twice the actual VRAM and a 10% boost in clock speed it would obviously fare even better.

Being indistinguishable in games doesn't really imply equality. It implies designing for good playability on the slower platform. If you could put G80 in any of these consoles, cross platform games would probably still run the same because that's how the game was designed.

Only? What about fill rate, multisampled fill rate, texturing throughput, VRAM size, texture memory bandwidth, setup limit (not too sure on that one)...

A lot of those things can actually be faster on Xenos, believe it or not, in real-world situations. Not always, of course.

Sure, Xenos wins when it comes to vertex processing but how many games actually stress the vertex processing capabilities of R580?

On the PC, I'd agree. For games primarily intended to be sold on the consoles, I bet plenty do. Your COD4 example, for one.

There is of course the edram. There's 4x the bandwidth within the edram compared to R580 but there's only 1/2 to the daughter die itself. And of course there's the lack of compression which scews things.

The 1/2 to the daughter die is less of a problem than you think, as it's a narrow point in the data stream. Sure, for pixels with no AA and no blending R580 is faster, but often BW isn't a problem anyway. Enable those and add non-zero texture bandwidth and it's very possible to see Xenos on top.

Were as Xenos does?

When I say RV630 isn't performing up to par, I mean in comparison to R600, not to paper. The COD4 benchmark looks okay, but I haven't seen many recent 2600XT benches and only remember the launch. It was a bigger step down from the 2900XT than the 8600GT was from the 8800GTS, even though specs suggested the opposite would be true. There was something flawed about RV630.

My point was that RV630 is not a good basis for you to judge ATI's architecture, even if you believe that R600 is (which it isn't).

One is based on the other and any differences between them should obviously be in R6xx's favour. Logically if R6xx performs below what we would expect based on its specs then so should Xenos. Unless of course they changed something about R6xx which drastically reduced the performance of the architecture. Which would be quite absurd.

A) You are comparing to RV630, not R600/RV670. The former is not half the speed of the latter per clock, despite specs suggesting so.
B) It's not absurd, because problems happen all the time especially when you're late. NV30 was less efficient than NV25 in almost every way. The R6xx architecture was also designed to scale downward, resulting in a different organization of the ALUs and TMUs.
C) Console workloads are different, and not just in terms of vertex/pixel load. There are things in Xenos that are probably not there in R6xx to address this.

Cell may be faster in very specific workloads but that hardly translates into being generally 2-5x faster than a quad core or even a dual core. In fact i'm sure there would be plenty of situations were it would be much slower than even a single core.

Of course. I'm just saying that if someone claims one thing then one can claim the other. In terms of my own opinion, I agree that a quad core from Intel is substantially faster than Xenos.

 

[mrcorbo]

I think this is a bit of an odd question. If you're going to confine your comparisons to the PC in terms of the classic Wintel definition you could probably go back as far as the NES and find few consoles that weren't superior graphically to what was available on PC at the time of their release. Widen it to include all "personal computers" and yeah, you're probably looking at the PS1. They just totally nailed their product design pushing both CD-ROM and 3D at exactly the right time with the exact right product. They executed so unexpectedly well, that I bet that it caught a lot of the industry flat-footed and forced them to scrap a lot of their existing plans in order to play catch-up.

 

[Heinrich4]

My guess is Dreamcast with cpu SH-4 with 1.4Gflops and gpu PowerVR2/DC with compatability with DX 6.1 end 1998

(in this time pcs run games like unreal and hal life 1 with pII 400MHz + Voodoo 2/3Dfx sli)

 

[AlphaWolf]

I think this is a bit of an odd question. If you're going to confine your comparisons to the PC in terms of the classic Wintel definition you could probably go back as far as the NES and find few consoles that weren't superior graphically to what was available on PC at the time of their release. Widen it to include all "personal computers" and yeah, you're probably looking at the PS1. They just totally nailed their product design pushing both CD-ROM and 3D at exactly the right time with the exact right product. They executed so unexpectedly well, that I bet that it caught a lot of the industry flat-footed and forced them to scrap a lot of their existing plans in order to play catch-up.

You could actually go back further to platforms like intellivision when the pc's on the market were still text only.

 

[pjbliverpool]

COD4 didn't show G71 in the best light, from what I remember.
http://www.firingsquad.com/hardware/call_of_duty_4_demo_performance/page3.asp
http://www.gamespot.com/features/6183967/p-4.html

True, G71 doesn't look all that great in the game but it should be remembered that its running at something like 1024x600 on the consoles so its a drastically lower resolution that what those benchmarks are taken at. I also don't fully accept that its a solid 60fps on the console versions as there is too much variation when playing the game to allow that (i.e. an 8800 may average 80fps but it will still regularly drop into the 40's). I expect 60fps is more of a "most of the time" kinda thing. With most of the time equating to >50%. I can't prove it of course but it certainly wouldn't be the first console game to claim to run at 60 or 30fps but clearly run slower. One things for sure, to maintain a solid 60fps all the time the game would have to average about 100-120fps

There's a good chance that COD4 has a more console-like workload than PC games in the heyday of G71. It looks like the unified GPUs have a notable leg up on the older ones (if you recall R600 often has marginal gains over R580, as expected from the specs, but not here). You'll also notice the 7900GTX 512MB performing slower than the 2900XT 256MB, in case you still really believe your analogy.

Thats a good catch, in fact its one of the very few examples I have seen were the 2600XT does come out on top (and amazingly even the 8600GT). Its unlikely to be down to unified shaders though based on R580's performance. Perhaps something to do with branching capability?

Being indistinguishable in games doesn't really imply equality. It implies designing for good playability on the slower platform. If you could put G80 in any of these consoles, cross platform games would probably still run the same because that's how the game was designed.

Perhaps. However in light of any onscreen evidence showing either platform to be performing significantly better, and given how on paper they are fairly even GPU wise, there doesn't seem to be much reason to assume Xenos is significantly superior.

Granted, I would pick Xenos over RSX if it was a one or the other choice of which is better but I don't think its a cut and dry win. And of course, something like the GTX 512 is faster than RSX again.

A lot of those things can actually be faster on Xenos, believe it or not, in real-world situations. Not always, of course.

There's my problem. On paper, R580 is faster. On screen evidence, i.e. cross platforms games, tend to corroborate this and at the time of launch, even ATI said it was more powerful. I just can't swallow the idea of an older, theoetically weaker GPU being faster in the real world when all the evidence I see points to the opposite.

On the PC, I'd agree. For games primarily intended to be sold on the consoles, I bet plenty do. Your COD4 example, for one.

But thats just it. How many console games come to PC and won't run as good or better on an X1950XTX than the console from which they came? If there were plenty of examples then I could agree with this but i'm not seeing any. COD4 for exmaple, is averaging 41fps at 1920x1200 and 16xAF. Its would be a stretch to say thats not performing as well or better than the console version which runs at well below 720p (assuming identical levels of details are used, HDR format springs to mind).

The 1/2 to the daughter die is less of a problem than you think, as it's a narrow point in the data stream. Sure, for pixels with no AA and no blending R580 is faster, but often BW isn't a problem anyway. Enable those and add non-zero texture bandwidth and it's very possible to see Xenos on top.

Yes i'm sure there are definatly situations when the edram will help achieve things that wouldn't be possible on R580 but what i'm saying is its not just a straight 64GB/s < 256GB/s. There is the interface between the dies to be considered as well as the lack of compression on the daughter die. Isn't that something like 8:1 on the R580? Thats gotta have a major impact.

I'm sure each memory architecture has its benefits over the other, i'm just saying edram doesn't equal a clear and sweeping memory bandwidth win for Xenos.

When I say RV630 isn't performing up to par, I mean in comparison to R600, not to paper. The COD4 benchmark looks okay, but I haven't seen many recent 2600XT benches and only remember the launch. It was a bigger step down from the 2900XT than the 8600GT was from the 8800GTS, even though specs suggested the opposite would be true. There was something flawed about RV630.

I would have said it performs pretty much in line with its specs. 1/4 the ROPs, 1/2 the texture units, ~1/3 the shaders, ~1/4 the memory bandwidth.... if anything its amazing that it performs as well as it does

My point was that RV630 is not a good basis for you to judge ATI's architecture, even if you believe that R600 is (which it isn't).

For arguments sake then lets put RV630 aside and instead focus on R600. You say its not a good judge of Xenos and I agree its far from perfect but its also quite easily the closest architecture that you are going to get and its definatly in the ball park. R600 is a direct decendant of Xenos. ATI had ample oppotunity to study Xenos's performance in the real world and improve upon it. There's little reason to expect R600 to perform worse for a given number of execution units than Xenos and in fact I think its pretty generous to simply assume parity between them in terms of efficienty. In all likelyhood for a set amount of resources R6xx will be faster.

But anyway, on paper R600 averages around 1.5-2x more powerful than Xenos (ignoring efficiency improvements).

Is it really that unrealistic to assume that at Xenos would be around 50-66% the real word power of R600? What better indicator is there? Unified shaders don't help Xenos in this comparison and in comparison to R600 in particular, neither really does the edram.

A) You are comparing to RV630, not R600/RV670. The former is not half the speed of the latter per clock, despite specs suggesting so.

As mentioned above, RV630 is actually much less than half the spec of R600.

B) It's not absurd, because problems happen all the time especially when you're late. NV30 was less efficient than NV25 in almost every way. The R6xx architecture was also designed to scale downward, resulting in a different organization of the ALUs and TMUs.

Well NV30 was a brand new architecture which bore little relation to NV25. Thats completely different to R600 which is basically an extension and enhancement of Xenos's architecture (the second generation unified shader architecture in ATI's words). ATI engineers would have to be pretty incompetent to actually re-design it to go slower

Anyway, I don't really think NV30 was inefficient compared to NV25. It was afterall much faster while also supporting DX9 (however poorley). Sure it was crap compared to R300 but it really did wipe the floor with NV25 for the most part.

C) Console workloads are different, and not just in terms of vertex/pixel load. There are things in Xenos that are probably not there in R6xx to address this.

But again, how are they different? A game is a game is a game whether you playing it on PS3, Xbox 360 or PC. All may have different ways of getting that game onto your screen but the task and the end goal are identical in each case. Pixel/vertex loads for example are no different in the PC version of COD4 than the PS3/360 versions. Nore is any other aspect of the game as far as I can tell.

 

[archie4oz]

You could actually go back further to platforms like intellivision when the pc's on the market were still text only.

Indeed, glad to seem somebody mentioning the Intellivision.

However, I'd go with the Sega SG1000 since it launched (w/limited availability) with a 3.58MHz Z80 clone the same year the IBM PC launched with a 4.77MHz 8088, and was faster than pretty much any other home computer (e.g. Atari 400/800, Commodore PETs and VIC-20s, TRS-80s. Plus it had more sophisticated audio graphical capabilities at the time. The next year was the ColecoVision which was pretty much the same piece of hardware...

After that, I'd say the Dreamcast then the Playstation. Honorable mention to the Mega Drive which also launched with a fairly powerful CPU for 1988...

 

[Scott_Arm]

Super Nintendo. Definitely not this gen. The quality of PC games this gen is too similar and close to that of the new consoles. PS1 and PS2 gens didn't really interest me from any of the console makers. They catered to different styles of games, but the genres I wanted to play were better on the PC.

 

[swaaye]

Maybe N64. PCs didn't quite have the 3D hardware going yet (Voodoo1 was $300 and I didn't know anyone with one) It was very exciting for me, at least, to see N64 games. I was never impressed with PS1 and its pixels and perspective problems. Pre-rendered games and FMV were kinda neat initially. But the PC had those too. 2D games were superior in my eyes. N64 was the first console that could do capable 3D rendering, IMO.

NES and SNES never blew my mind compared to what PCs could do during their day either. People went bananas over Street Fighter 2 and MK, StarFox, etc, but the games never really amazed me much. Part of that was probably because the games weren't my thing.

 

[Scott_Arm]

Maybe N64. PCs didn't quite have the 3D hardware going yet. It was very exciting for me, at least, to see N64 games. I was never impressed with PS1 and its pixels and perspective problems. 2D games were superior in my eyes. Same with everything that game before, really. NES and SNES never blew my mind compared to what PCs could do during their day either.

I'm not sure if I was really blown away by the SNES from a technology standpoint. It's hard to remember that far back. But I do remember being blown away by some of the games. I never owned an NES and had no interest in getting one, but when I played the SNES I was blown away, and had to have one. There really were a lot of exciting titles for that system. When it started to die down. I got back into the PC world with Doom/Doom2 and the fps, rts explosion and never looked back to consoles until the current gen of consoles.

 

[swaaye]

Yeah I was perfectly happy with the nice, clean Gouraud-shaded 3D games on PC like MechWarrior2 and the Star Wars sims (because they were incredible games!). 2D games like Crusader, Diablo, Descent, Theme Hospital, C&C, X-Com, and even Doom were impressive to me too. Many of these were ported to consoles too.

Played a lot of SNES, but it was really a platform for very different entertainment than what was on PC. Unless you were into the Japanese RPGs, it was more "casual" gaming IMO. Sit down and play for a bit vs. get really involved with a game. At least for me that's how it was.

 

[Scott_Arm]

Yeah I was perfectly happy with the nice, clean Gouraud-shaded 3D games on PC like MechWarrior2 and the Star Wars sims (because they were incredible games!). 2D games like Crusader, Diablo, Descent, Theme Hospital, C&C, X-Com, and even Doom were impressive to me too. Many of these were ported to consoles too.

Played a lot of SNES, but it was really a platform for very different entertainment than what was on PC. Unless you were into the Japanese RPGs, it was more casual gaming IMO.

Don't forget Syndicate and Aces Over Europe. I played those to death.

SNES to me was really the Nintendo titles, like Super Mario World, Zelda and Star Fox. It also had Secret of Mana, which I really enjoyed. Oh, and Mortal Kombat 2.

 

[DeadlyNinja]

The N64 came out around 1996, while the Voodoo 2 didn't come out till mid or late 98. The texture resolution was obviously the biggest draw back, but well made games could still hold their own against the Voodoo 1 back then. Don't forget that once the 4MB RAM expansion came to, it also allowed the N64 to not look too dated next to a Voodoo 2.

 

[Akumajou]

Definitally not Xbox 360 or PS3.

Indeed, short term memory attention span is not a good thing.

The Playstation 1 by far made a huge impact in terms of displayed technology I was even thinking of mentioning NES, SNES and Genesis but the tech in those consoles was several years old at launch and the PC could have hosted the games that were made on those platforms but it did not and thats why those old consoles are so fondly remembered, mainly for their games, not technology.

The N64 in my opinion would have to be the undisputed choice mainly because it was Super Mario 64 that put to shame all other consoles and the PC establishment.

The Dreamcast although another good choice has a problem for me because the defining game on that platform is Shenmue, something that PCs had come close to only in 2d years before and only a home console powered by Yu Suzuki's SEGA-AM2 could pull off, thank goodness that Shenmue's spirit is still alive and kicking this time on Sony Playstation 3 Ryu Ga Gotoku Kenzan aka Yakuza 3.

 

[swaaye]

The N64 came out around 1996, while the Voodoo 2 didn't come out till mid or late 98. The texture resolution was obviously the biggest draw back, but well made games could still hold their own against the Voodoo 1 back then. Don't forget that once the 4MB RAM expansion came to, it also allowed the N64 to not look too dated next to a Voodoo 2.

Back on the 'ol Dimension3D forum that was popular during those days (B3D of the '90s), people used to debate how N64 compared to Verite V1000 and Voodoo. It wasn't really a very powerful machine (compared to Voodoo) and had lots of drawbacks going on in its hardware, but it certainly did ok. Voodoo 1 was out of its league, but in the end the difference wasn't enough to really do more than increase a game's frame rate.

Until Unreal, anyway. Unreal was a huge step forward for 3D game graphics as far as I was concerned and it ran fine on Voodoo1.

 

[function]

If the N64 had launched with a CD drive back in 1994 when Sega turned it down (not sure that was a good idea tbh) it would have been quite something - it took the larger carts to really show the N64 off.

If the M2 had appeared in 1996, that would have possibly offered a "Dreamcast" level of 3D dominance (arcade, console and PC) but it unfortunately never appeared. That would have been a nice one to have in the private collection of favourite failed consoles ...

 

[Unknown Soldier]

I am talking about the average PC requirements to play the defining PC game. In other words my midrange 1994 pc:

Midrange 1994
40MHz Intel 486DX2
8MB RAM (whatever type was used back then)
500MB hard drive
1MB SVGA-capable graphics card
3.5" floppy drive (although I'll probably cheat and add a CD-ROM drive)
Windows 3.11

was more or less based on the requirements to play Quake II at the time. Slightly subjective, I know, but come on it's not that hard to define.

Sorry Butta .. I had about the same system as you've specified above but for the 500MB HDD .. I don't think they existed then .. I think I had a 80MB HDD.

But the rest I had .. DX2-40, 8MB Ram, 1MB SVGA with SB16 Sound card and Creative CD-Rom(4x I think it was).

I think the PS1 was nice .. Grand Tourismo 1 was awesome visually although I never bought the PS.

US

 

[Rangers]

I think the original xbox faired much better than the 360.

When xbox launched it had a GPU which was arguably superior to the fastest PC GPU of the time (GF3 Ti500). But when the 360 launched it would have had to compete with dual 7800GTX 512MB's.

Considering a single 7800GTX 512MB has a fairly significant leg up on RSX which itself competes well with Xenos, it seems to me that Xenos was seriously overshadowed by the best that was available for the PC at the time.

A year later and the X1950XTX went well beyond Xenos performance without resorting to dual GPUs. I'm assuming Xenos's performance in PC terms is around HD2600XT + 25% level.



I doubt Xenon is even remotely near the performance of a modern quad core. Competitive with the slower dual cores certainly but quad cores? I find that extremely unlikely.

A single 7800GTX is actually almost identical to RSX, not vastly superior. In fact early on 7800GTX was clocked at 430mhz. It may have had more bandwidth, but when you combine PS3's two buses with the ability of programmers to use tricks in a closed box console enviroment, the 7800GTX's bandwidth advantage was probably not a factor in real terms.

I'm assuming Xenos's performance in PC terms is around HD2600XT + 25% level.

Whoa that's way wrong, lets assume it's about as capable as RSX (most games show this) then it's about as capable as a 7800GTX which puts it firmly in X1800/X1900 class.

Looking back at it 360 did seem to be about as powerful GPU wise as the top PC's at the time. PS3 was a different story, as it came out a year later. Of course not counting SLI here, which is a bit of a "cheat" and rarely works well anyway.