Anarchist4000
[H]ard|Gawd
- Joined
- Jun 10, 2001
- Messages
- 1,659
Being able to play these 720P clips without hammering the CPU would be rather nice. Just imagine low end cards doing that. 
![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
New R600 info 320-stream processors
- Thread starter trek554
- Start date
Being able to play these 720P clips without hammering the CPU would be rather nice. Just imagine low end cards doing that.
why , what else is your cpu doing when you are watching movies on the pc?
i wouldn't care if the video decoding used 100% of 2 cores as long as the damn thing was smooth and looked good
Anarchist4000
[H]ard|Gawd
- Joined
- Jun 10, 2001
- Messages
- 1,659
If any video playback is using 100% then it won't be smooth. Anything that could momentarily spike CPU usage could cause stuttering on playback. What if you were trying to watch the videos and your CPU was a Celeron or Sempron.
- Joined
- Mar 8, 2007
- Messages
- 4,098
I don't understand why, with 320 stream processors, it isn't performing a lot better than the 8800GTX. Isn't that twice the processors?
You live in a world of fantasy bro. Don't believe everything you read posted by random forum douchebags.
Anarchist4000
[H]ard|Gawd
- Joined
- Jun 10, 2001
- Messages
- 1,659
The numbers are different but you still need to take clockspeed into consideration. The R600 processors should be running 750-800MHz while the G80 processors run about 1350MHz. Some synthetic tests have shown R600 to have substantially more math power under best case scenarios, 500FLOPS vs 350FLOPs.
As for DX10 there has been some information that has leaked out that R600 is substantially faster at Geometry shaders than G80. I wouldn't see 70% as unrealistic under certain high resolution scenarios.
As for DX10 there has been some information that has leaked out that R600 is substantially faster at Geometry shaders than G80. I wouldn't see 70% as unrealistic under certain high resolution scenarios.
For this reason a lot of potential buyers are thinking it will be awesome for F@H, seeing what the current ATI cards can do in that regard.
peacetilence
2[H]4U
- Joined
- Sep 23, 2004
- Messages
- 2,343
CGI movies are made with several thousand AA so I do believe there would be a distinguishable difference. Perhaps not with modern games but as geometry levels increase the extra AA will made games look much more impressive.
I use 1920x1200 as a typical gaming resolution and I see a big difference between 2x and 4x. 4x looks incredible with certain games (Trackmania Unlimited is one of them) but there are still jagged edges about. To see no jagged edges in games at all would be just great.
I use 1920x1200 as a typical gaming resolution and I see a big difference between 2x and 4x. 4x looks incredible with certain games (Trackmania Unlimited is one of them) but there are still jagged edges about. To see no jagged edges in games at all would be just great.
Simple. AMD/ATI's version of stream processors, are not as efficient as NVIDIA's SPs in G80. And btw, it's more than twice. Twice would be 256.
And where is that info, about R600 equivalent to SPs, running @ 750-800 Mhz ?
That's nothing short of wishful thinking on your part, mainly because we haven't seen neither G80 or R600, in action in a DX10 game. I mean, as a benchmark, since G80 has been used many times, to show off DX10 capabilities, in Crysis videos, for example and it seems to run quite well. I believe Brent saw it live.
Ummm, how do you know? For someone claiming to be neutral your posts definitely seem a bit "skewed".
Another way of looking at it would be that G80GTX's sp's are running at 1350MHz, whereas ATI's are supposedly at about 800MHz. Clockspeed counts too you know.
Skewed ? It's simple math isn't it ? 320 SPs vs 128 SPs and they have similar performance. What does that say to you ?
Yes, of course clock frequency counts. But I asked for that piece of info. Where is the "rumor", that says R600 version of SPs, runs at 750-800 Mhz ?
So far we have only unconfirmed R600 bench information. Even if similar performance how do you know that it is the sp's in each architecture that are performance defining (ie bottlenecking)? You are assuming that each architecture is fully utilising its shading resources, but you have zero proof (as do the rest of us). There is alot more to gpu's then just their shading componentry.
As for the frequency I quoted, all indications so far have R600 running without a separate shader frequency domain, thus at core speed. And given latest reports are claiming ~750-800MHz for the R600 core then.......
Xer0_P0INT_Xer0
Weaksauce
- Joined
- Jan 15, 2007
- Messages
- 126
just wondering, but what is the midrange version of the X2XXX series?
Of course, it's all unconfirmed. This is a rumor thread after all, isn't it ? And what do we base our opinions in these kind of threads ? Rumors. Until everything is confirmed, we do nothing more than speculate.
I don't understand why everyone gets so touchy when someone "thinks" that R600 will not do so well in this or that, or that it will not be as efficient as G80 in something else, especially when it's so late in the game.
I find it funny that the "R600 has 320 Stream processors" rumors, makes some of you "WOW, that's awesome" and it's almost considered as a fact, but when someone says "Well, that can only mean their 320 are not as efficient as NVIDIA's 128", then that means that person is a !!!!!! and that he is only basing his opinion on a rumor.
Anyway, I don't even want to go there, because it's way off topic and I'm only here to share my opinion on the "rumors".
Back on topic. As you know or might not know, probably the most important features, in G80's architecture, are the Stream Processors. The fact that the "unconfirmed" R600 320 Stream Processors, are more or less evenly matched to the G80's 128, in the "unconfirmed" becnhamrks, can only make me say what I did. But of course, this is only an opinion, a guess, until actual facts are put on the table. And that unfortunately, will only happen, when AMD/ATI decides to release R600 and reputable sites such as [H] or Tech-Report, put their reviews online.
We don't even know if it's really 320 SP's the way Nvidia defines them. AMD could just be countering Nvidia's marketing. What if R600 was 64 vec5 processors similiar to Xenos. AMD could just multiply 64*5 to get 320. But those processors would not be similiar to Nvidia's at all - e.g they would process all rgba components of a pixel in one shot whereas G80 would do it sequentially. So until we get more info a simplistic view of 320 being more than 2x 128 will tell you nothing about performance.
They should not have the first X anymore. Now they are HD 2xxx.
And the mid-range cards of R600, should follow the usual numbering scheme of HD 2600 XT and HD 2600 Pro, followed by the HD 2650 XT and the HD 2650 Pro.
As far as specs go, everything is unconfirmed, as are the full specs, of R600 high-end models.
- Joined
- Dec 9, 2000
- Messages
- 16,337
i wonder if the XT and the XTX have different amounts of sps(if they even call them SPs on the ATI card)
that would make things interesting
that would make things interesting
You said it all. It's simplistic, only because we have nothing else than these rumors, to base our opinions on. And I say it again, this is a rumor thread. Everything we say here, may or may not be confirmed.
Yes, assuming that the applications being benched are extremely shader heavy, the sp's may be the bottleneck. My point is that even with equal score in say 3D06, that could have absolutely nothing to do with the shaders; TMUs or ROPs etc may be the bottlenecking component, and as such an equal score in a certain bench may give absolutely zero information on sp efficiencies. Only once the architectures are actually known will the comparable strengths and weaknesses be ascertained.
To some up, at this point in time rumored benchmark scores give virtually no real indication of actual shading abilities. To draw any other conclusion at this early stage is fairly wrong.
You do understand you just said exactly what I did, in my previous reply to you, but in other words ?
And no actual conclusion is drawn from a rumor thread. I thought that was clear from the start...
from the info thats available, the r600 has 64 pixel/vertex shaders or stream processors , whatever you want to call it,
how did you figure the 64 pixel/vertex shaders are less efficient than the 128 pixel/vertex shaders in 8800gtx? the r600 has half the shaders 8800gtx does, and it outperforms it,
i think it's quite clear which one is more efficient, the 320 is just a theoretical number.
You are twisting the rumor mill, to your own will. It says right on the thread title "R600 has 320 stream processors" and that's what's being discussed here.
im not twisting anything,
i can guarantee you that r600 doesn't have 320 physical stream processors,
more like 64 which are capable of processing 5 pixels/shaders per cycle, which would give you the 320 number, which again, is just a theoretical number and can not be directly compared to the 128 physical stream processors of the 8800gtx
Thx, for the link, shi**y there's no 'one-time' fee for payment options..I could see this becoming very useful in the months/years to come.
Looks as though I got some serious thinking to do in the wattage dept. when considering my dx10 upgrade. The rumors of the r600 look nice on a forum, but I'm still holding out for some real comparisons of what the cards are made for. After all, i won't be upgrading for dx9, as it is -- I can do that just fine.
I knew I was close to my max wattage, but 743 out of 750 is pretty damn close. I still have my awesome amperage with the Silverstone, so I think I'm in the clear a bit longer; may need to get an ext. enclosure for some of my RAID drives; [H]Writers, Kyle, can you guys drum up some cheap and easy NAS case reviews for the new year?
Unsure if you'll know this answer or what your relationship to the project is but..
So, is that calc considering load times, idle times, ambient room temp, climate zone or anything else detrimental to power capabilites/capacity?
Anarchist4000
[H]ard|Gawd
- Joined
- Jun 10, 2001
- Messages
- 1,659
The 320 number came from Dave Orton I believe. Yeah it's not 320 separate SPs but Nvidia isn't exactly using 128 either. They're actually using 8xVec16(or 16x8, can't remember) versus 64xVec4+1.
As for processing power we have "confirmed" numbers that show R600 is almost 50% faster in raw math processing, 512GFLOPs vs 345GFLOPs. Now considering it's highly likely they're using the Vec4+1(Xenos) setup, and Xenos is said to be 95% efficient when running shaders, according to an old Beyond3D article, I wouldn't expect a very big difference between the cards.
The clockspeed, although mentioned on many sites, can even be calculated to around 750-800MHz based on the Teraflop demo/benchmark that ATI ran. So 2 cards with >1TFLOP means each should be about 500TFLOPs. Also 800MHz*64*5=~512GFLOPs.
As for the DX10 performance we have quotes like, "R600 is much more performant with geometry shaders" coming from the lead developer of FSX. Now GS being the only major hardware addition that comes to mind with DX10 I would consider that to be R600 running DX10 better. And in the context that the "much more performant" was used I don't read it as 15-20%, more like 30+%.
And even with the current generation you can come up with scenarios where one cards is 70% faster than the other. Granted you're typically looking at sub 20 framerates but it's not exactly unheard of.
As for processing power we have "confirmed" numbers that show R600 is almost 50% faster in raw math processing, 512GFLOPs vs 345GFLOPs. Now considering it's highly likely they're using the Vec4+1(Xenos) setup, and Xenos is said to be 95% efficient when running shaders, according to an old Beyond3D article, I wouldn't expect a very big difference between the cards.
The clockspeed, although mentioned on many sites, can even be calculated to around 750-800MHz based on the Teraflop demo/benchmark that ATI ran. So 2 cards with >1TFLOP means each should be about 500TFLOPs. Also 800MHz*64*5=~512GFLOPs.
As for the DX10 performance we have quotes like, "R600 is much more performant with geometry shaders" coming from the lead developer of FSX. Now GS being the only major hardware addition that comes to mind with DX10 I would consider that to be R600 running DX10 better. And in the context that the "much more performant" was used I don't read it as 15-20%, more like 30+%.
And even with the current generation you can come up with scenarios where one cards is 70% faster than the other. Granted you're typically looking at sub 20 framerates but it's not exactly unheard of.
G80's SIMD setup is 16x8. But it is not Vec8 the same way Xenos is Vec4. Vec4 there implies four components of the same pixel/vertex. Vec8 in G80 refers to the same component of 8 different pixels/vertices. And I believe the quoted Xenos utilization numbers were per ALU. That would not have captured intra-ALU inefficiencies where a vector operation did not apply to all 5 slots.
These types of comparisons just end up confusing people as it's not that straightforward. Hence why AMD could choose to promote 320 SP's - it's just easier for people to understand.
400 dollars is tempting for a card like this..... again, most of us will look at reviews before we buy.
I'm waiting on the 1 gig of ram version, because my girlfriend will freak if I buy a video card in a month and want another one two months later. (we live together).
I'm waiting on the 1 gig of ram version, because my girlfriend will freak if I buy a video card in a month and want another one two months later. (we live together).
Commander_HK47
Weaksauce
- Joined
- Jan 13, 2007
- Messages
- 73
Thank you for posting information about both cards SPs(or at lest expected numbers). I had asked a few days back how AMD was gonna combat the G80 GTX 128* SPs with only 64 SPs(witch 64xVec4+1=320* makes sense) and was basically told to go eat my foot
0.5 TFlops and not just with MAD instructions... well thats easy with G80 and "future driver release" (hopefully coming soon)
G80@575MHz + 97.92 series drivers
-------------------------------------------------
gpudip 18: cel shading - 23 70 1 512 768 10000 - 682,3311 msec - 403,3983 GFlop/sec
G80@575MHz + "future driver release"
-------------------------------------------------
gpudip 18: cel shading - 23 70 1 512 768 10000 - 584,9589 msec - 470,5480 GFlop/sec
G80@702MHz + "future driver release"
-------------------------------------------------
gpudip 18: cel shading - 23 70 1 512 768 10000 - 496,1952 msec - 554,7236 GFlop/sec
G80@575MHz + 97.92 series drivers
-------------------------------------------------
gpudip 18: cel shading - 23 70 1 512 768 10000 - 682,3311 msec - 403,3983 GFlop/sec
G80@575MHz + "future driver release"
-------------------------------------------------
gpudip 18: cel shading - 23 70 1 512 768 10000 - 584,9589 msec - 470,5480 GFlop/sec
G80@702MHz + "future driver release"
-------------------------------------------------
gpudip 18: cel shading - 23 70 1 512 768 10000 - 496,1952 msec - 554,7236 GFlop/sec
Uuuhh.. maybeWhy 702Mhz....you expecting an SKU at that speed?
its strange you care more about the name then the theoretical performance
because:
a) theoretical performance means squadoosh to me, and
b) when someone asks me what kind of video card I'm running I don't like having to say x fourteen times
well. you can always rename the card if u don;t like the name,
name it puffy or bobby or something "i have a bobby 512"
..that sounds good
what!!!!!!!!!!!!~!!$#@%$#%$% NOOOOOOOOOOOO i loved the X before the number, changing the X to the HD was a bad move.
- Joined
- Dec 9, 2000
- Messages
- 16,337
Who cares how many Xs my card has? as long as it's better then what you hvae
JK seriously though why does it matter, it's a component in your box , do you care that they components of your car? no if it really matters to you that much, you should be buying your pc parts from best buy
Great! So you can guarantee something that not even AMD/ATI has confirmed...
Which btw, is something I don't understand. Why doesn't AMD release the damn full specs already ? Didn't they say they could release R600 "today", a few weeks ago ? So what exactly is the problem of revealing the specs ? IMHO, that was the ultimate marketting BS, concerning R600.
That can be explained away for competitive reasons. Things aren't looking too hot in the market for AMD this year. They have major losses to contend with in addition to mounting shareholder pressure. After their rather hasty purchase of ATI, they have experienced nothing but setbacks with Intel and NVidia biting at their heels. I don't blame them one bit for remaining tight-lipped about a highly crucial product launch that could spell doom for them if they make a single misstep. It is not only about maintaining relevance in two major markets in the technology industry for AMD at this juncture, it is a matter of survival.
It can work both ways. Having said that, if R600 have enough muscle to surpass the GTX, wouldn't it be wise to revealed the specs and bench? The numbers will speak for themselves and it will certainly convinced the fans that are sitting on the fence to wait for it's release.