The GeforceFX(NV30) will have less memory bandwidth than a Radeon 9700!!!

[Overclocker456]

Boy what a shock this is... Clock speed isn't everything folks...

"The GeForce FX will be using 2ns chips. With DDR2 at 1GHz on a 128bit memory bus, the raw memory bandwidth comes out to 16GB/sec. The raw bandwidth of the Radeon 9700 Pro, with its 256bit memory bus and DDR running at 620MHz, is 19.8GB/sec."

That says it all.. So all the idiots that say that 1GHz DDR2 memory is going to destory the Radeon 9700, think again...

[Cullam3n]

Can you say... overclocK?

-PC

[quegyboe]

Yeah, I fail to see how the NV30 is going to smash the Radeon 9700, when the Radeon has more memory bandwidth. Should be interesting to see the outcome of this battle of the video giants...

[Tipycol]

I can't wait for the NV30 to come out. And the reason is, the competition should lower prices :D Hopefully when it comes out, the price for the 9700 can be a little more affordable. I mean I still would like to see it perform, but the price/performance competition with ATI is more of a reason to see it come out (at least to me) :)

[Cullam3n]

Doesn't history like to repeat itself? Does anyone else see the resemblance between this and the Geforce 3 Ti200/Radeon 8500 war?

-PC

[CrashOveride]

if it OC's massivly and with-out much cooling, it will be worth it to me otherwise.... poop... the ATi cards run Moby dock (OSX bar emulator type thing) like crap but... for beter gaming... i might take tha scrifice:( :D :eh?:

[ShadowFolder]

Originally posted by Overclocker456
Boy what a shock this is... Clock speed isn't everything folks...

"The GeForce FX will be using 2ns chips. With DDR2 at 1GHz on a 128bit memory bus, the raw memory bandwidth comes out to 16GB/sec. The raw bandwidth of the Radeon 9700 Pro, with its 256bit memory bus and DDR running at 620MHz, is 19.8GB/sec."

That says it all.. So all the idiots that say that 1GHz DDR2 memory is going to destory the Radeon 9700, think again...

i don't think that's not a real problem 4 gf-fx...
whatever, we'll see benchs, soon!

[snyper1982]

you guys are funny, bashing on the nv30 before you have even seen a single benchmark, and i really think the idiot comment is taking it a bit far there overclocker456. and just so you know bandwidth isnt everything either, the memory interface is just as important as the bandwidth. look at the parhelia, twice the bandwidth of the 4600, but the 4600 smokes it. so we will just have to see how it turns out.

[Cullam3n]

Originally posted by Cullam3n
Doesn't history like to repeat itself? Does anyone else see the resemblance between this and the Geforce 3 Ti200/Radeon 8500 war?

-PC

-PC

[funnyperson1]

Originally posted by Cullam3n
Doesn't history like to repeat itself? Does anyone else see the resemblance between this and the Geforce 3 Ti200/Radeon 8500 war?

-PC

it could play out like that....but the only reason the 8500 lost initially was because of crappy drivers, and the 9700Pro has a lot better drivers at release than the 8500 had after 3 months into release lol.....

[donny_paycheck]

16GB/s x 4x compression = virtual 48GB/s, which > 20GB/s.

We'll see how this compression holds up, but the NV30 has a 6+ month lead on the 9700 and it will cost more.

It'll be faster. Duh.

[HotKoala]

Originally posted by snyper1982
look at the parhelia, twice the bandwidth of the 4600, but the 4600 smokes it. so we will just have to see how it turns out.

Got link? I've seen nothing but Parhelia beating 4600 regularly with AA and/or AF on. Nevermind in multimonitor.

As for the raw memory bandwidth, I'm pretty sure the NV30 has that LMAIII(whatever they call it) tech going for it. Remember, the GF2Ultra had more bandwidth than the GF3 but the GF3 still beat it due to bandwidth saving features.

[donny_paycheck]

Link (http://www.hothardware.com/hh_files/S&V/nv30_preview(2).shtml).

No matter how high the core is clocked, if the GPU isn't fed enough data, performance will suffer. A GPU this powerful needs a ton of memory bandwidth to operate at or near its top speed. NVIDIA's solution to this problem is three fold. NVIDIA has implemented a third generation version of their very efficient "Lightspeed" Memory Controller, which operates with four independent 32-Bit memory controllers, for an effective 128 Bits. In addition, GeForce FX boards will be populated with ultra fast DDRII-type memory with an effective clock speeds hovering around 1GHz! The combination of this high-speed memory and NVIDIA's memory controller offer about 16GB/s of bandwidth when operating at 500MHz. The final boost comes from a proprietary 4:1 lossless color compression scheme that effectively raises theoretical max bandwidth to 48GB/s.

[Maxvla]

its not how fast the memory is... its how efficient the architecture is.

LMAIII is supposed to be alot more efficient than Hyper Z

[DaddyB]

Originally posted by donny_paycheck
16GB/s x 4x compression = virtual 48GB/s, which > 20GB/s.


16x4=64 ... 16x3=48

Obviously Nvidia knows that they wont get 4x the bandwidth all the time from their 4:1 compression, so they advertise 3x (48GB/s). Different situations will allow for different levels of compression, its not a finite amount.

As I have said in other threads the 20GB/s the 9700 gets doesnt take ATi's compression into account (Hyper Z III) so comparing those 2 numbers is useless, the 9700 probably gets at least 40GB/s of "effective" bandwidth from compression but since nividia has just invented that "spec" no other company has given "effective bandwidth" (or whatever you want to call it) specs before.

So, 9700 has 20GB/s, GFFX has 16GB/s of real bandwidth

and 9700 has up to ??GB/s, GFFX has up to 48GB/s of compressed/effective bandwidth. No conclusion as to which is better can be drawn from that, I think only the benches will show one to be better then the other.

[Overclocker456]

Originally posted by DaddyB


16x4=64 ... 16x3=48

Obviously Nvidia knows that they wont get 4x the bandwidth all the time from their 4:1 compression, so they advertise 3x (48GB/s). Different situations will allow for different levels of compression, its not a finite amount.

As I have said in other threads the 20GB/s the 9700 gets doesnt take ATi's compression into account (Hyper Z III) so comparing those 2 numbers is useless, the 9700 probably gets at least 40GB/s of "effective" bandwidth from compression but since nividia has just invented that "spec" no other company has given "effective bandwidth" (or whatever you want to call it) specs before.

So, 9700 has 20GB/s, GFFX has 16GB/s of real bandwidth

and 9700 has up to ??GB/s, GFFX has up to 48GB/s of compressed/effective bandwidth. No conclusion as to which is better can be drawn from that, I think only the benches will show one to be better then the other.

Well said my friend.. It seems Nvidia is "MILKING" the specs a bit much eh??

[Rav]

Good point DaddyB. If Nvidia is claiming bandwidth based on compression, what would the 9700's 'effective' bandwidth be? 2x, 3x of the raw? Anyone know?

-Rav

[donny_paycheck]

Hmm, this bandwidth compression stuff is annoying. We'll see how it ends up when we get out hands on them. It may be calculated 3:1 to give leniency for some stuff that can't be compressed well. Average compression in real world usage will probably fall between 3:1 and 4:1. The 48GB/s figure might be a little conservative, even.

[Dragon_Engineer2]

Is the GeforceFX supposed to have better IQ then past Geforce cards?

[OC Noob]

The final boost comes from a proprietary 4:1 lossless color compression scheme that effectively raises theoretical max bandwidth to 48GB/s.


Theoretical max, not a conservative estimate. Maybe the reason it isn't 16 x 4 is because it isn't used to compress everything. Hence the "color compression" discription.

ps I have no basis for this, but I wonder if this compression is why Gf cards IQ isn't as good as Radeons. Did the Gf2 use an earlier version of this compression?

[Kinerry]

ok this is why you guys dont design cards and they do ;)

memory bandwidth isnt that important, thats why id go amd over intel anyday

[CrystalMethod]

Why doesn't everyone wait till the card comes out before slamming it? Overclocker456, defianately seems to be obsessed with it's performance, and what it may or may not be capable of. I have no doubt that Nvidia will push production back until they manage to beat the Radeon 9700. GPU vs GPU, the NV30 is better. How the OEM's implement it is a completely different story. It's up to the OEM's to do something with it, based on a REFERENCE DESIGN.

[OC Noob]

Tell that to me using PC133 with your AMD:p

[EyelessFace]

The way I was think of it is that it may have less memory bandwith, however, it will use what it has with great effiecency*, which could put it far ahead of the Raedon.

*sorry, cant spell today

[OC Noob]

I would wait, but its fun to BS about vid cards. Much better then Funniest Animal Videos:)


EDIT: I don't doubt it will be very fast, but I do think they made a mistake by not going 128 bit instead of 256. They went way out of their way to move to a .13 micron process and use DDR-II, but skimped on the bus? Why?

[funnyperson1]

Originally posted by OC Noob
I would wait, but its fun to BS about vid cards. Much better then Funniest Animal Videos:)


EDIT: I don't doubt it will be very fast, but I do think they made a mistake by not going 128 bit instead of 256. They went way out of their way to move to a .13 micron process and use DDR-II, but skimped on the bus? Why?

easy.... 1GHZ>600mhz...thats what most consumers will see....try to explain bus widths and theyll just stare at you blinking lol

[runsalone]

personally, I am just going to wait until february 2004 when Doom III is released and then pick out a video card. I'm sure there'll be some nice stuff out by then!

[gravitywell]

how about we just wait til its out

[snyper1982]

ok, check out the bechies there:
http://www.tomshardware.com/graphic/02q2/020625/index.html

You will sometimes see the parhelia come out on top, but for the most part it gets smoked.

[DaddyB]

Yes it should have better IQ due to 128bit FP precision.

Now first let me say that I highly doubt the GFFX will attain a full 3x bandwidth boost in all situations, in fact even the qutoes from nvidia say it will achieve "up to" 48GB/s. According to Anand's preview (http://www.anandtech.com/video/showdoc.html?i=1749&p=5) the reason 4:1 compession translates into 3x the bandwidth is:

The lossless compression algorithm can obtain up to a 4:1 compression ratio which explains where NVIDIA gets their 48GB/s of memory bandwidth from. Remember that the 500MHz DDR2 memory on the GeForce FX provides 16GB/s of bandwidth on the 128-bit memory bus, but multiply that by 4 (don't forget to subtract out the original 16GB/s of data) and you'll get the 48GB/s of memory bandwidth NVIDIA is claiming.

On to the real reason for this post... I have just done a lot of searching and reading to find the answer to this whole compression issue.

From what I have read it seems HyperZ (ATi's first implementation of compression/culling on the radeon DDR card) was able to "increase" the bandwidth by up to 40% (none of these compression methods increase bandwidth, they just cut down on how much bandwidth is needed so it has the effect of increasing it), that comes from an article at
Anandtech (http://www.anandtech.com/showdoc.html?i=1435).

According to this article at guru3d.com (http://www.guru3d.com/review/ati/radeon-showdown/index2.shtml) HyperZ increases the bandwidth by a mere 20% yet Anand claimed 40% (above).

HyperZ III (now we are talking 9700) is said to increase the bandwidth by "over 50%" according to hothardware.com (http://www.hothardware.com/hh_files/S&V/radeon_9700_9500_preview.shtml) and digit-life.com (http://www.digit-life.com/articles2/radeon/ati-r300.html#p4).

EDIT: Sorry, it doesnt increase bandwidth by 50%, it "reduces memory bandwidth consumption by over 50%", that doubles the bandwidth (if the 9700 were using all 20GB/s and you reduce that by 50%, then its only using 10GB/s which is half). This is more consistant with ATi's claim of "4:1 compression" and "minimum of 2:1 compression".

And interestingly enough in this article at beyond3d.com (http://www.beyond3d.com/reviews/ati/radeon9700pro/index.php?page=page4.inc#hyper) they say:

ATI claim a minimum of a 2:1 compression ratio and a best case of 4:1 during normal rendering. However, when FSAA is enabled the compression ratios will nearly scale linearly with the FSAA sample number, so, for instance, with 6X FSAA applied the best compression ratio achieved will be 24:1.

So the 9700 is already capable of 4:1 (60GB/s "effective" bandwidth) compresion and at minimum does 2:1 (40GB/s)? Stay with me here I'm almost done... If thats true then the GFFX's 48GB/s of "effective" bandwidth means nothing, to further fuel that fire according to Nvidia.com (http://www.nvidia.com/view.asp?IO=feature_lmaii) the LMA II used on the GF4 cards:

Reduces Z-buffer traffic—one of the largest consumers of memory bandwidth in a graphics subsystem—by a factor of four, without any reduction in image quality or precision.

And according to the writting about LMA II on the side of the GF4 box (which you can read at tiger direct (http://www.tigerdirect.com/applications/SearchTools/item-details.asp?sku=P450-7240T)):

Lightspeed Memory Architecture II
- LMA II boosts effective memory bandwidth by up to 300%. Radical new technologies?including Z-occlusion culling, fast Z-clear, and auto pre-charge?effectively multiply the memory bandwidth to ensure fluid frame rates for the latest 3D and 2D games and applications.

So even the GF4 is capable of tripling (increasing by 300% means multiplied by 3) its bandwidth thanks to compression and according to that bit about the 9700 it is capable of doing the same. If this is true then the 9700 will have more bandwidth then the GFFX but when I look at all of the percentages that I found its really hard to say which numbers best represent the actual "increase" in bandwidth by any card. I expect that nvidia's implementation of bandwidth compression would be better then ATi's but now I am less sure then ever that this will make up for the difference in actual bandwidth between the GFFX and 9700.

So I believe it is possible that the GFFX's bandwidth will be its bottleneck and wont be able to keep up with the 9700 (the bandwidth, not the card) but I will stick with my original opinion on the matter... Only the benchmarks will tell us for sure.

[Overclocker456]

DaddyB, another excellent post... :0)

[OC Noob]

Yeah, thanks for doing the dirty work that most of us were too lazy to do:)

[Wangster]

So here's a question.

When are there going to be any games that will utilize this kind of processing power?

I hope to whatever gods are listening that Doom 3 and other games that will utilize the engine are going to come out in 2H03 or earlier.

Speaking of which, does anyone know when Doom 3 is slated to come out?

W

[ShadowFolder]

why we want that chip?
tons of user says: "don't u see doom3 on E3, i need that chip!"
there's a another absurd "id software & nvidia partnership" :D

i think that mandness'll continue till 500$ limit,
and
we'll say: "i can't pay 500$ a graphic card, f*** new chips! "

i think that's just madness of chipmakers...

[donny_paycheck]

Remember Doom III is an immature engine as it's been shown so far. So the v1.0 released engine will probably run a lot smoother on slower cards.

[Morbid]

Personally, I think you're all comparing apples to oranges here. The NV30 (unfortunately named the GeForce FX) is totally rebuilt from the ground up and resembles past GeForce's in no way. The "LMA III" is complete reworked and will be called something else... the name escapes me now.

Basically, no one can make a conclusion on which card will be better until we see some actual benchmarks from reliable sources. People can speculate and nothing is wrong with that... but when people start taking it seriously, then you have a problem.

Just keep this in mind: GeForce FX is not out yet. GeForce FX utilizes new technology. Radeon 9700 uses known and previously used technology.

Then again, that's just my opinion, I could be wrong.

[Darkseid]

Originally posted by whereismy386
why we want that chip?
tons of user says: "don't u see doom3 on E3, i need that chip!"
there's a another absurd "id software & nvidia partnership" :D

i think that mandness'll continue till 500$ limit,
and
we'll say: "i can't pay 500$ a graphic card, f*** new chips! "

i think that's just madness of chipmakers...

wasn`t that demo on e3 played on a 9700?

and yes daddyb great article, for me theortethical bandiwth means s*** because more ore less ATI and Nvidia both have goos bandwith savings, so in the end it all translates into the same REAL bandwith plus something, nvidia could have a little advantage in lma iii but that will make up for the 4gb/s less of real bandwith? don`t think so

[PreservedSwine]

Originally posted by Cullam3n
Can you say... overclocK?

-PC

Can you say....Overheat?:beer:

[donny_paycheck]

Originally posted by Morbid
Just keep this in mind: GeForce FX is not out yet. GeForce FX utilizes new technology. Radeon 9700 uses known and previously used technology.

Well put. Speculative benchmarks are fun to read but bear no weight in the real world. Early next year it'll be time for nVidia to put up or shut up.

[PreservedSwine]

Just keep this in mind: GeForce FX is not out yet. GeForce FX utilizes new technology. Radeon 9700 uses known and previously used technology.
Really? You might want to define "technology" before you just throw it out there, as an all defining tool. What about the NV30's FSAA? It hasn't changed a bit, they are still using the same ordered grid sampling pattern....And they are now using an algorithm like ATI in thier A.F. design.

Not exactly new.

Their memory controller is pretty cool, and they are using bandwidth saving techniques, and their DX9 shaders seem impressive. Of course, all the same applies to the R9700 as well....

I do agree, the GeforceFX is not out yet, and likely still won't be for another 2-3 months, at best......

By then, who knows what else will be on the scene....:burn:

[OC-Master]

Originally posted by Overclocker456
Boy what a shock this is... Clock speed isn't everything folks...

"The GeForce FX will be using 2ns chips. With DDR2 at 1GHz on a 128bit memory bus, the raw memory bandwidth comes out to 16GB/sec. The raw bandwidth of the Radeon 9700 Pro, with its 256bit memory bus and DDR running at 620MHz, is 19.8GB/sec."

That says it all.. So all the idiots that say that 1GHz DDR2 memory is going to destory the Radeon 9700, think again...

Um, GF FX runs at 500MHz GPU which will mash the Radeon up any way you look at it. ATI having 3GB/s more of memory bandwidth wont be able to make up for all this massive GPU power. Besides, everyone now will want 256MB of video ram. 128MB is a joke if you ask me, even my friends good ole GeForce3 Ti 200 had 128MB on it LOL


OC-Master

[rUfUnKy]

ok I'm not going to get into the technical mubo jumbo seeing its pretty much all been said ..But does anyone honestly think nvidia is stupid enough (business wise) to release a second rate card 7 months later then its highest competitor when they have a 9700 pro right in there hands to test it against..They surly aren't going to let NV30 hit the shelves until it is past par with the 9700 pro.

[shadow]

Speaking of which, does anyone know when Doom 3 is slated to come out?

March 31 2003

Amazon.com and various other retailers begin taking pre-orders for Doom 3. Price is listed at $49.99, and a March 31 2003 date is indicated.

NIN Hotline (http://www.theninhotline.net/features/doom/doom.html)

Doom3 @ Amazon (http://www.amazon.com/exec/obidos/tg/stores/detail/-/videogames/B00006C2HA/qid%3D1027036049/sr%3D8-3/ref%3Dsr%5F8%5F3/002-2677974-0499206)

shadow





edit: more info and links

[odin30]

Actually we are comparing ati and nvidias compression, but we should not.
Anands article clearly states that nvidias is superior because it compresses everything(all colors) before it goes to memory.
Which is why they say aa is nearly free. Ati does not compress all colors and is speculated that it compresses gamma only.
I dont know about the ati compression tech (maybe someone can clear it up), but most of the threads ive seen on anandtech,b3d,hardocp etc have all said that nvidias compression is much better, therefore we could see much higher bandwidth in the nv30. Of course this is speculative until we actually see benchies but it is interesting. You can't compare effective bandwidth and raw bandwidth. Thats like comparing a v10 dodge truck with a v10 dodge viper, which vehicle would win a race? The viper of course even though they have similar horsepower the viper uses its horsepower more efficiently when it comes to speed. Anywho i am very excited about ati's next offering and would like to see it announced soon. We really can't compare a 9700 with a nv30 when they are available almost 8 months apart. Hopefully we get a r350 to compare the nv30 with.
If ati comes out with it by may or june nvidia may be in a real pickle. Remember 3dfx and nvidia?

[PreservedSwine]

Originally posted by rUfUnKy
ok I'm not going to get into the technical mubo jumbo seeing its pretty much all been said ..But does anyone honestly think nvidia is stupid enough (business wise) to release a second rate card 7 months later then its highest competitor when they have a 9700 pro right in there hands to test it against..They surly aren't going to let NV30 hit the shelves until it is past par with the 9700 pro.

I agree w/ you there, just remember that the NV30 AND the R300 were being developed, on paper, at the same time, both for Autumn release dates. Nvdia underestimated the amount of delays they would run into going w/ .13. After thet realized their card was going to miss an entire product cycle, it was too late to make any major chip changes. Nvidia also underestimated what the R300 chip would be capable of, especially at it's .15 design.

Suddenly, it's easy to understand how one chip, designed for a release date 6 months prior, may not be nearly as superior as you would expect.

[EluSiOn]

0.13 manfacture proccess = smaller die... runs faster.. cooler.. more efficient and also means more gpu per waffer .... which means lower manufacture cost for nVidia and higher profit margin for them

0.15 man. proccess is good but reaches its limit. I have no doubt about it ATI will go through same amount of painful proccess using 0.13 man. proccess. It just that nVidia did this proccess early and it pay its price for it. Intel and AMD is trying to move to 0.09 man. procces. ATI has some catch up game ahead of them.

I won't comment on GF FX vs Radeon 9700 until we see 3Dmarks and benchmarks. Of course... I am a loyal nVidia Fan just because their driver support for Geforce series never sucked.

[PreservedSwine]

Originally posted by EluSiOn
0.13 manfacture proccess = smaller die... runs faster.. cooler.. more efficient and also means more gpu per waffer .... which means lower manufacture cost for nVidia and higher profit margin for them

0.15 man. proccess is good but reaches its limit. I have no doubt about it ATI will go through same amount of painful proccess using 0.13 man. proccess. It just that nVidia did this proccess early and it pay its price for it. Intel and AMD is trying to move to 0.09 man. procces. ATI has some catch up game ahead of them.

I won't comment on GF FX vs Radeon 9700 until we see 3Dmarks and benchmarks. Of course... I am a loyal nVidia Fan just because their driver support for Geforce series never sucked.

Good points, though a bit incomplete.
While it's true than a smaller die will produce less heat, you lose that advantage when you ramp up core speeds to such high levels. At 500mhz, the NV30 is going to be the "hottest" card around, no kidding. Did you read about it? Apparently, the air exhuast isn't coming out warm, it's downright HOT!

Secondly, it's true you can get more chips/per waffer on a smaller die, but you must also take into account your yield %. If Nvidia is producing 20 more GPU's per wafer, but the Yields are 20% lower than that of the .15 process, they just lost money. At issue here, who eats that? TSMC or Nvidia? Asd far as Nvidia having a head start on the .13- no doubt. But sometimes it's wise to learn from someone else's mistakes, instead of your own. ATI just better hope they figured out what Nvidia did so wrong to put them so far behind schedule, otherwise, they are sure to duplicate it.

Benchmarks, I agree, nothing to really compare. We'll have to wait until any actually exist. But about Nvidia drivers never failing- that made me laugh. I swore I'd never own another Nvidia product after owning a Geforce1:) Times sure change, eh:)

[Stumpjumper5200]

I just hope the card will come out already, or this will turn into one of the biggest fanboy wars of all time :D

Just for the record.........nVidia vs ATI, Intel vs AMD, Rambus vs DDR......I don't care, I'm neutral. I just want whatever's faster :D

[ShadowFolder]

Originally posted by Stumpjumper5200
I just want whatever's faster :D

i just want whatever's faster & CHEAPER!!! :D

[PreservedSwine]

How about IQ?
Neither the R9700 or NV30 will be noticably faster than an R8500 or GF3 if you're running your games @ 800x600 w/ no AF and no FSAA.

Each one will render more rrames than you moniter can display.

The advances w/ the R9700 and NV30 will allow you to play at high res- using FSAA and AF, w/out losing too many fps. If you're soley interested in fps and not IQ, then you'd be well advised to skip these cards:)

[Stumpjumper5200]

What???????

Well yeah, that's true..........but who in their right mind buys a card like that then runs 800x600 with no AA or AF??

[CrashOveride]

i used to like ATi until i found that Nvidias 2d engine is better... or at least for some stuff... Nvidia cards run Moby dock- an OSX taskbar emulator- MUCH better than ATi.. my card and a ATi Radeaon 9700... my card runs it smother and games more stable:D :eh?: :D