GF3 may not compete w/Kyro III but what about Radeon II

[ThExorcist]

For all the Radeon fans out there I felt this leaked news may be of interest to you...
Radeon II

6 pipelines
350mhz core
Truform Rendering Technology
HyperZ

Sounds interesting... Seems ATI is always trying to increase graphics quality as well as increase performance, or at least maintain a balance between the two. If they could release drivers more frequently that may help them more in this next round of video card battles.

Any assumptions on what kind of power you think we are looking at with Radeon II's six pipelines.

 

[Dark4ng3l]

Hmmmm I doubt that 6 pipelines and a 350MHz core would do anything.

 

[ThExorcist]

I am sure memory bandwidth will still be a problem for this card, but not as bad as the GF3. The geforce 3 can only match a Radeon now, as far as reducing overdraw goes, Link and I am sure the Radeon's HSR has improved but still should be a problem.

By the way 6 pipelines and a minimum of 3 textures per pipeline(they haven't said if they were increasing textures but at a minimum it should match previous Radeon's) is an incredible texture rate, theoretically.

 

[Ponyboy25]

Well, with what we've seen so far, the Kyro II has been able to meet it's "theoritical" fill rate. That's what makes it so good, they don't throw some huge phony numbers at you that your card will never achieve in gameplay.

 

[ThExorcist]

"Well, with what we've seen so far, the kyro II has been able to meet it's "theoritical" fill rate. That's what makes it so good, they don't throw some huge phony numbers at you that your card will never achieve in gameplay."

True, true but as I mentioned in another post if even a GF MX400, or GF GTS had no memory bandwidth problem or were 100% efficient they would be extremely powerful. But since they aren't Kyro rules that arena.

I still think Radeon 2 will have a better percentage of its theoretical fillrate covered than GF3 making it the stronger performer. This may end up being a battle between Radeon 2 and Kyro 3 if Nvidia's next card doesn't do better with its HSR.

Fillrates from Anandtech / Theoretical Fillrates (All in Mpixels) / Percentage
GF3 = 495.28 / 800 = 61.9%
GF2 Ultra = 374.95 / 1000 = 37.5%
Kyro II = 352.89 / 350 = 100%
GF2 Pro = 329.94 / 800 = 41.2%
Radeon DDR= 280.33 / 366 = 76.6%
GF2 GTS = 257.22 / 800 = 32.2%
Radeon SDR= 204.89 / 332 = 61.7%
Voodoo 5 = 168.44 / 667 = 25.3%
GF2 MX = 146.01 / 350 = 41.7%

This is what Radeon 2 and Kyro 3 would be if they hold true to the percentages given above...
(based on 350mhz core for Radeon, naturally fillrate would be less if they lower core speed)
(Kyro 3 is based on 250mhz core and 4 pipelines w/overdraw factor of 3 which Anandtech said is average today)

Radeon 2 = 1608.6 / 2100 = 76.6%
Kyro 3 = 1000 / 1000 = 100% w/ overdraw of 3 would be = to 3000 Mpixel card

Boy this is getting scary:Q
Now you can see why other review sites have been saying Nvidia's next card may not compete with Kyro 3 and you can see Nvidia has a long way to go concerning fillrate efficiency against Kyro 3...

 

[BFG10K]

I am sure memory bandwidth will still be a problem for this card, but not as bad as the GF3.

Huh? The GF3 is *very* efficient considering it still uses a brute force approach to rendering. Despite having the exact same RAM as the GF2 Ultra, it's usually 25%-50% faster in fillrate limited situations.

 

[ThExorcist]

"Huh? The GF3 is *very* efficient considering it still uses a brute force approach to rendering. Despite having the exact same RAM as the GF2 Ultra, it's usually 25%-50% faster in fillrate limited situations."

BFG10K, I wasn't saying GF3 wasn't more efficient then an GF2 Ultra as it obviously is... What I was saying is that the Radeon is more efficient than a GF3. The only thing holding the Radeon back is its 366 Mpixel fillrate. Now if Radeon 2 maintains its current specifications as well as improves its HSR (HyperZ) then Radeon 2 will be quite the card.

Fillrates from Anandtech / Theoretical Fillrates (All in Mpixels) / Percentage
GF3 = 495.28 / 800 = 61.9%
GF2 Ultra = 374.95 / 1000 = 37.5%
Radeon DDR= 280.33 / 366 = 76.6%

Lets assume GF3 Ultra comes out and has similar specifications to Radeon 2...

6 pipelines
350mhz core
(may not be too unreasonable since GF3 has 200mhz core and 4 pipelines right now)
This also would yield same theoretical fillrate as Radeon 2 but lower actual fillrate. They would have to improve their HSR much much more because we all know Nvidia will not be content to just match the performance of the Radeon 2 if they are currently the performance leaders.

GF3 Ultra = 1299.9 / 2100 = 61.9%
Radeon 2 = 1608.6 / 2100 = 76.6% --- These fiqures may be better if they improve HyperZ.

 

[kazeakuma]

bit OT here but is anyone else annoyed at what companies are calling their technologies. I mean cmon, Charisma Engine? Pixel Tapestry Architecture? HyperPipelined tech? It get's a bit stupid after a while.

While the HSR debate is going to continue I don't think it's fair to compare the HSR on Radeon to GF3. Mainly because the Radeon has a MUCH lower theoretical fill rate so it's less likely to be limited by the RAM anyway. Where as a GF3 would need blinding RAM to acheive it's theoretical Fill rate. I'm not saying that it's bad or anything but the lower the Fillrate the easier it is to achieve.

 

[Ponyboy25]

ThExorcist, what you say about "Nvidia's next card may not compete with Kyro 3 and you can see Nvidia has a long way to go concerning fillrate efficiency", has me thinking about something I read around here in the forums. I remember seeing a post stating that 3Dfx had been working on something similar to the technology used on the Kyro (sorry, I don't remember the 'technical' name). Could it be possible that NVidia has continued the work now that they own the rights, and will release that on an upcoming card rather than their 'brute force' technique? Maybe they won't have such a problem in the future...

 

[ThExorcist]

"While the HSR debate is going to continue I don't think it's fair to compare the HSR on Radeon to GF3. Mainly because the Radeon has a MUCH lower theoretical fill rate so it's less likely to be limited by the RAM anyway. Where as a GF3 would need blinding RAM to acheive it's theoretical Fill rate. I'm not saying that it's bad or anything but the lower the Fillrate the easier it is to achieve."

Not necessarily. It depends on the implementation (whether it be Tile based or HSR). Kyro only has about a 16 Mpixel fillrate less than Radeon and is about 240 Mpixels away from GF3 w/ 100% efficiency using SDRAM. Radeon has a 76.6% efficiency rate with only 183mhz DDRAM and its actual fillrate is only about 220 Mpixels away from GF3 as well. GF3 uses 250Mhz DDRAM and is pulling 61% which is alot better than previous GF models but if Radeon or Kyro increased their fillrates, they have the option of using much faster memory then what they use now. GF3 is at the ceiling and could only go up to 300Mhz DDRAM which was just recently announced and will be more costly. They must optimize there HSR process or implement a new process. (embedded memory or gigapixel technology perhaps)

Should be interesting.

 

[ThExorcist]

"ThExorcist, what you say about "Nvidia's next card may not compete with Kyro 3 and you can see Nvidia has a long way to go concerning fillrate efficiency", has me thinking about something I read around here in the forums. I remember seeing a post stating that 3Dfx had been working on something similar to the technology used on the Kyro (sorry, I don't remember the 'technical' name). Could it be possible that NVidia has continued the work now that they own the rights, and will release that on an upcoming card rather than their 'brute force' technique? Maybe they won't have such a problem in the future... "

Yes, it is the technology they got from gigapixel which 3dfx owned. I hope they have continued the work because from my post above they are pretty much going to run into a ceiling if they can't optimize their HSR process more. The question is will the technology be implemented in GF3 Ultra or GF4. If you asked me now I would say most likely GF4 since the technology would be radically different then GF3 (as far as rendering goes) and they would probably want to market it differently to associate it with a totally new process and leave their successful brute-force method behind them. Is it possible that GF3 Ultra may get a few preliminary features in it from GF4 for Nvidia engineers to mess around with in drivers and what not? Very possible also...

 

[DevilsAdvocate]

Even still, it does take a few generations for companies to optimize their products for new features. It could be fairly argued that TnL did not mature until the GF3 was released, for instance.

TBR took a while to develop as well. Powervr had to nurse that product along through several renditions to get 99% of the bugs worked out of it. The Kyro2, as a result, is a very efficient, solid card.

Just because nVidia purchased 3dfx does not mean that their version of tbr will be flawless out of the box. It will likely take a while before all of the bugs get worked out of it. The same is true for ATi... Radeon's Hyper-Z (an HSR, not TBR methos I realize) took a little while to get the bugs ironed out of it.

My point is that Kyro3, because of ST's history with it, should be a more mature product with regards to its implementation of TBR. Kyro3's TnL unit may be another story. It may take a while to get programmable T&L to turn out right.

 

[BFG10K]

ThExorcist, the predicted scores you gave for the Radeon II are only valid if the memory speed scales in the same ratio as the core speed does. If only the core speed increases, the efficiency will go down. In otherwords the prediction you gave is extremely bold.

ATi are going to have to do a lot more than improve Hyper-Z to compete with nVidia's Z-occlusion culling and the crossbar memory controller. You're also assuming that nVidia's next card will not have faster memory. After all, if they leave the core speed the same and increase the memory speed, the efficiency increases.

It's yet another reason why simply looking at the efficiencies doesn't show you the whole picture. At the end of the day it's only the real world fillrate that really matters.

 

[pen^2]

everytime the competitors try to come up with some fancy hype, but at the end its nvidia that comes on top. this aint surprising, they had been a whole cycle ahead of the rest of the industry. as much as would like to see kyro3 and radeon2 doing well, i predicts its gonna be good ole nvidia delivering yet again. while kyro2 and radeon are great cards and breakthroughs for the respective companies, they are still playing a catchup game. all this speculation bores me, talk the talk when the cards are actually out. by the time those cards come out, gf3 ultra or whatever nexgen card would be right around the corner. remember when x3dfx hyped v5 to hell? nvidia quitely came out with gf2 which was supposed to be just a follow-up refresh for gf1, matched and bettered v5 in many terms. you might think i am an nvidiot but i am a happy radeon user, just having a hard time comprehending this overzealous nvidia hatred and alternative worship. sure gotta have the top dog right?

 

[ThExorcist]

"ThExorcist, the predicted scores you gave for the Radeon II are only valid if the memory speed scales in the same ratio as the core speed does. If only the core speed increases, the efficiency will go down. In otherwords the prediction you gave is extremely bold."

It is true that if the memory doesn't scale with the core than the efficiency will go down. But I would say it is safe to say Radeon 2 will use faster memory. Especially since Imagination Technologies said 250mhz is very likely in Kyro 3. Since Radeon 2 would need to stay competitive the memory will be 200mhz min. most likely 250mhz since Radeon used 183mhz initially.

"ATi are going to have to do a lot more than improve Hyper-Z to compete with nVidia's Z-occlusion culling and the crossbar memory controller. You're also assuming that nVidia's next card will not have faster memory. After all, if they leave the core speed the same and increase the memory speed, the efficiency increases"

Just remember that 300mhz DDR is the fastest memory for Nvidia right now which is only an increase of 17% from what they use in GF3. That doesn't leave them much room for upgrading memory. Radeon 2 has options between 200mhz and 300mhz as well as a more efficient HSR process than GF3 (as long as they scale the memory and core together).

Based on this new info I just read at xbitlabs
Radeon 2

it isn't too far fetched to expect Radeon 2 to have
250mhz core / 250mhz DDR
or
300Mhz core / 300mhz DDR
(or any mixed combination of the above...) - very unlikely

P.S. Looks like Radeon 2 just cut two of its pipelines which would affect above fillrates that I posted. Oh well may just be Kyro 3 only this Xmas. Still waiting on info from Nvidia's next card to see where competition is going to be.

 

[BFG10K]

Just remember that 300mhz DDR is the fastest memory for Nvidia right now which is only an increase of 17% from what they use in GF3

No, you've calculated that wrong. Currently the fastest RAM in use (GF2 Ultra and GF3) is 460 MHz (or 230 DDR) while the new RAM is 600 MHz (or 300 MHz DDR). That means that the new RAM is ~30% faster than the fastest RAM currently in use.

nVidia has room to support 500 MHz, 550 MHz or 600 MHz configs (or anything in between) and they will still get significant performance boosts.

 

[powervr2]

and significant price boosts...

 

[nam ng]

Hopefully the Radeon2 will be here soon, Radeon is about one year old already.

Perhaps the real KyroII also will show up this year, there's only a KyroI_ultra masquerading as KyroII right now.

 

[ThExorcist]

"No, you've calculated that wrong. Currently the fastest RAM in use (GF2 Ultra and GF3) is 460 MHz (or 230 DDR) while the new RAM is 600 MHz (or 300 MHz DDR). That means that the new RAM is ~30% faster than the fastest RAM currently in use."

That is what I meant. 300Mhz DDR module would equal 600mhz Rate. Should have phrased that better. Just remember that even DDR Ram is theoretically suppose to give or double your bandwidth performance but in real life situations it doesn't (more like 3/4 or less). This applies to all cards that use this memory (Kyro 3,Radeon 2,GF3). So in reality the percentage isn't that far from what I stated and it may be a little higher but still means that the most efficient HSR will in the long run prevail(bad driver support can change things). I still stand by Radeon being more efficient than GF3 using DDR RAM when applying what I just said. Don't get me wrong though I am sure Nvidia's engineers will pull more performance out of the GF3. Their engineers are very efficient "pun intended" and capable. I was just stating as it stands now Radeon is more efficient...

 

[Ironduke]

<< Hopefully the Radeon2 will be here soon, Radeon is about one year old already.

Perhaps the real KyroII also will show up this year, there's only a KyroI_ultra masquerading as KyroII right now. >>



so what if it is!

Its the best bang for your buck!

 

[BFG10K]

That is what I meant. 300Mhz DDR module would equal 600mhz Rate

ThExorcist, you still don't get it. You said it would be 17% faster. In reality it would be 30% faster.

That is what I was showing.