NVIDIA they are getting to big for their own good just read this

[skywalker[TSG]]

READ THE PDF BEFORE READING THE TEXT BELOW
http://www.geocities.com/johnrreynol...ia_on_kyro.pdf

Page 1

"Kyro2 – What You Should Know"

What NVIDIA will have you know - or - What NVIDIA would like you to believe?



Page 2 – "Kyro2 is Old – Very Old"

"Kyro2 is ‘beefed up’ Kyro designed in 1997"

Who says KYRO was designed in 1997?

However, is not the GeForce MX basically a cut down version of GeForce256 – released in 1999, probably designed in 1998/99?

"Only AGP2X"

Even KYRO 1 was AGP4X. VDO documentation states AGP4X For KYROII as well: http://www.videologic.com/Factsheets/VL-60481.pdf

Even then, has AGP4X made that much of a difference?

"no TwinView dual TV out option"

So what? Do Gamers want this? Few games support the dual output option that Matrox pioneered with their G400 in ’99.

It’s a low cost part. How many MX boards out there have it implemented? How much extra do these boards cost? How many games and apps really use it?

"designed by committee with unproven record" "Videologic design team (best effort PowerVR)"

Wrong, a PowerVR design team; PowerVR Technologies being a Division of Imagination Technologies.

PowerVR designed PCX1/2 (branded Apocalypse 3D) which went head-to-head with 3dfx in the early days (while NVIDIA were bungling with Quadratic textures and NV1 – at least PowerVR picked the right primitive!) and also have console experience (the market NVIDIA is ‘unproven’ in) in the form of the Dreamcast PowerVR designed graphics core, a system that can still put a number of PS2 titles in an interesting light!

PowerVR also designed ‘NAOMI2’, using multiple PowerVR Series 2 chips (Dreamcast) and separate T&L (capable of supporting up to 6 hardware lights, spot, directional, global no matter what type, with no performance degradation – something GeForce certainly has trouble with!) for Arcade systems – SEGA is currently making Virtua Fighter 4 for this system. Also of note is the fact that this system was delivered to SEGA at the same time as GeForce256 was sampling so you can see that PowerVR have had just as much experience of T&L systems as NVIDIA, albeit in different markets.

The PowerVR designed KYRO has been met with reasonable success in the PC space last year, and recent reviews show it clearly outperforming NVIDIA’s similarly priced MX series.

More recently PowerVR has signed a deal with ARM to supply low powered 3D graphics solutions for integration into PDA devices, a market NVIDIA had also recently signaled intent to enter prior to the PowerVR deal being announced.



"SGS Thompson manufacturing Poor record with complex design (6x86, Transputer)"

Nice NVIDIA, trashing your previous semiconductor supplier, real classy…

Anyway, both of the products mentioned above were brought to the market on time. Does everyone have a good manufacturing record when a process is new? Are there no manufacturing problems with NVIDIA latest complex design? Rumors of a sixth revision to ‘improve yields’ would suggest the contrary.

"Guillemot board design • Never made their own boards before"

Refer to next page…



Page 3 – "Guillemot has no experience designing graphics boards!"

Not content with muck spreading on only(!!) PowerVR and ST, NVIDIA now proceeds against one of their own board vendors…

"Guillemot had to hack an NVIDIA design to get Kyro2 to work!"

Did they ‘have’ to, or was it merely a quick and easy route to getting boards operable to make them available to the press…?

"All NVIDIA based Guillemot boards are designed by NVIDIA and released to them for distribution."

Does that mean Guillemot is incapable of producing their own boards, just because they choose to use NVIDIA’s reference design? They have their own in-house PCB design and research facility, and have altered designs on NVIDIA’s reference board for ones with PSU’s should NVIDIA have produced another card that saps too much power.

Even if they did have PCB design issues, they have Videologic’s board production knowledge to lean on if required – they have a proven track record in board design and manufacturing.

So, let that be a lesson to those other board vendors out there – this is the type of treatment you receive from NVIDIA should you dare to run your business as you see fit, and go with what’s best for your needs, not NVIDIA’s.



Page 4 – "The SharkyExtreme Review"

"*Note: We have had some difficulty in getting the 3D Prophet 4500 to work properly on our Intel-based platforms. However, we are working with Hercules to resolve this issue, and will be updating the preview when this problem is resolved."

Such as what?

KYROI appears to work fine on Intel platforms out there. Remember, this was also a pre-release, pre-production board (based on a GTS board as NV are so keen to point out!), did/does GeForce3 work 100% at this stage…?

"We did have an issue with Evolva, where the scene would render under Windows 2000, but no textures were applied to the models. After modifying all of the available compatibility options, we determined this must be a driver bug."

Wow, a driver bug? What, on a pre-release, pre-production board? And in Windows2000? Gasp! Are NVIDIA telling us that GF3 has none of these little niggles, ever? Given the pushed back release schedule and the patchy reviews of GeForce3… Anyway, Evolva seems to run fine on KYROI under Win9X/ME, just waiting for Computer Artworks to supply that Dot3 patch.



Page 5 – "The SharkyExtreme Review…more"

"Although Hercules claims that the 3D Prophet 4500 supports DXTC, it was interesting to note that both 3D Mark 2001 and 3D WinBench were unable to take advantage of these features. If they are supported, there is apparently some work to be done still."

KYROI/II supports DXT1 texture compression format, but not DXT2-5. PowerVR felt the low compression ratio (1:4, instead of 1:6) of these formats did not warrant the extra complexity support for these formats introduced, especially because of the fact that PowerVR cards are not that bandwidth limited in the first place. In Quake 3 engined games, for instance, this is not an issue as only DXT1 texture compression format is used for all textures. 3Dmark2001 and 3D Winbench, however, check all DXCaps texture formats and assumes that if one format isn’t available none are – these applications could use DXT1 fine on KYRO, while resorting to non-compressed textures on the other formats.

You may also like to note that KYRO’s implementation of DXT1 texture format doesn’t display the horrible banding issues that has thus far been evident on GeForce256 / 2 GTS / 2 Ultra / 2 MX (and Go and 3?)!

"There are caveats to buying such a card…"

Are there not caveats to buying any card? If you bought an MX over KYRO1 you’d be looking at worse performance, ugly textures (DXT1) and dithered, 16bit output in many, many games. Not to mention no EMBM…

"For one, the feature set is relatively dated."

Dot3 is dated? And NVIDIA has only begun to support EMBM with their latest $400+ chip. And 8 Layer multitexturing, even NVIDIA can only support 4 in a single pass with their $500 behemoth.

"While STMicroelectronics claims that T&L is ready to be implemented…"

Refer to NAOMI2 earlier.

"the KYRO II does not sport it, so future applications that make use of gratuitous polygons may be subject to slowdown as the increased number of computations could possibly bog down the tile-based system."

Interesting – NVIDIA make no mention of CPU performance, but opt to attack the tile system. However, the tile system will alleviate much of the work that immediate mode renderers will needlessly perform in such games. It will be a long time before the geometry detail really begins to bog down the sorting and culling system, at which point several other features can still be implemented within tile based systems to alleviate this issue. And, surely, when these geometrically complex games are available you wouldn’t want to be stuck with an MX either.

NVIDIA also introduced the word ‘may’ into that sentence, because, basically, they don’t know – they can’t actually state that this will happen; KYRO may not have hardware T&L support, but it may not necessarily be the case that all non-T&L cards will slow down in these upcoming games. Its interesting to note that another of NVIDIA’s ‘showcase’ GF3 games (as the big NVIDIA logo on the developers website bears evidence to: http://www.multiplayer.it/dronez/), Dronez, runs fine on KYROII with an average of roughly 50FPS at 1600x1200x32 (1.2 GHz Athlon with 256MB, KYRO2 clocked at 175) – is that so bad for one of NVIDIA’s next gen., fully fledged T&L titles…?

Of course, NVIDIA also conveniently overlooked the fact that as the scenes in games get more geometrically complex, PowerVR’s tiling architecture may also help out more, as more overdrawn polygons will be getting culled, over which immediate mode renderers will be needlessly slogging away at.

"Similarly, a lack of DirectX 8 features means upcoming applications will slow down significantly."

KYROI/II’s support for single pass 8 layer texturing has also proven to be a considerable boon in some newer titles, enabling KYROII to outperform NVIDIA’s high end / high cost Ultra boards (along with the tiling nature of the chip) – something that will no doubt be utilized more in DX8 titles.

Lest we not forget NVIDIA’s own supplied benchmarking tools also show that DX8 games using Vertex Shaders mean that those who have opted for GeForce256/GTS/MX’s T&L will be left using CPU power for their T&L needs as well. Seeing as NVIDIA is so keen to quote Anandtech in this document, let’s take a little excerpt from Anands GeForce3 review:
http://www2.anandtech.com/showdoc.html?i=1442&p=11

"It isn't a surprise that the GeForce3 comes out on top again, what is interesting is that the three GeForce2 cards and the Kyro II are all capable of the same real world T&L power. Keep in mind that the Kyro II has no hardware T&L, meaning that all four of the cards are relying on the host CPU for the T&L processing.

If this is an indication of what can be expected from future titles, are GeForce2 owners left in the lurch with a hard-wired T&L unit that will yield no tangible performance improvements in future games? If developers all move to support programmable T&L like that on the GeForce3, which they most likely will, will the T&L units on the GeForce2 series of cards be rendered completely useless?"

Oh, and where’s NVIDIA’s DX8 offering in this market segment? NVIDIA currently want you to shell out $400+ to have those features…



Page 6 – "Kyro2 - Unproven technology"

First it was old, now it’s unproven!

"Kyro2 still has serious flaws"

Based on what?

"Z problems and rendering errors"

It appears they are only basing this off the result of the Mercedes Benz Truck Racing tests, which will be explored later.

"Alpha in game = slow performance"

Alpha in game = still better performance than a traditional renderer!

It’s true to say that KYRO’s architecture is not able to perform its visibility culling of the textures used have alpha channels, however it still perform better than a traditional architecture because of the use of the on chip tile – all the alpha blending can take place on the tile, which is extremely fast; with traditional chips the blending of alpha textures introduces an extra frame buffer read/write which is far more costly in terms of performance than doing it on the tile.

"Dynamic textures (shadows) perform horribly"

Hrm – sorry? Would you care to explain that a little please NVIDIA?

Presumably they are (again) talking about MBTR’s shadowing issue, which all comes down to the same Z-Buffer issue.

"Bad compatibility on all games"

This has to be the most lunatic statement that has been in this document so far (and that is saying something!). They appear to be basing this comment off the results of one game, MBTR, which I think it’s fair to say is hardly typical of the games everyone is playing right now. However, many of you have read the reviews. . .other than MBTR how many reviewers have actually complained about game compatibility? Sure there’s been a few issues, but nothing that’s beyond any other chips – including NVIDIA’s.

"Kyro2 will look and perform terrible on new games • Giants, Black & White, Doom3"

What? How on earth do they fathom this? This is just a blatant, outright lie (in fact the document appears to be getting more desperate the further you go through).

Giants runs fine one KYROI, let alone KYROII, and in fact a patch has just been released by Interplay to enable Dot3 Bumpmapping in the game – meaning it looks much like a GeForce based boards, I assume NVIDIA thinks that these look horrible as well? That doesn’t speak much for the developer support contract NVIDIA has with Interplay!

As for Doom3, John Carmack has already stated that even GeForce3 will only be turning about 30FPS at 800x600 with all options on, so you certainly wouldn’t want to be stuck with an MX with that game. However, John has also stated that Doom3 will be using 8 texture layers – seeing as NVIDIA could only manage to implement 4 textures in one pass on GeForce3 in one pass it will need to do twice the passes KYROI/II does, which will eat up a whole bunch of performance.



Page 7, Page 8 – Another NVIDIA T&L Games List.

The list of T&L titles on these pages has been discussed many times before, and most, if not all, hardly scratch the surface of ‘requiring’ hardware T&L, as opposed to merely supporting it. Diablo 2? Puh-lease!



Page 9 – "Anandtech Review"

"Supplied 175 MHz boards to reviewers, spec’d to run at 166MHz (source SGS Thompson)"

Does it matter what it was spec’ed to run at if the boards are shipping at the reviewed clock speed? So far two vendors have announced KYROII based boards, Hercules and Videologic, and BOTH have announced a clock speed of 175Mhz – so where is the issue? Perhaps ST were just being conservative in their initial estimate of 166Mhz, but now that the chip is in full production they have found yields are good enough for a 175Mhz clock speed.

And NVIDIA is hardly able to stand on a moral high-horse here either – many may remember TNT2 reviews at differing clock speeds then those released at retail. At least the vendors have announced at the reviewed clock speed in this case.

"concern lies with the display errors we experienced in Mercedes-Benz Truck Racing… ….it is possible that other games may be effected by the improper rendering."

Again, harping on about this one game – how many people actually play it? As for other game having the issue – how many do? None that I’ve seen so far. I’m sure all it takes is a little liaison with PowerVR developer support to circumvent these issues, and with the increased profile that KYROII has brought them I’m sure many more developers will doing so in the future.

"buying Kyro2 is a risk."

As opposed to buying what? Buying NVIDIA’s lesser performing cards for the same retail price?

"from an unproven ‘committee’"

They seem pretty proven to me.



Page 10 & 11 - "Kyro2 Picture Quality…"

All based upon MBTR (again).

Quotes:

"In this picture right, one can see tire tracks that should clearly be behind the truck."

"The second picture shows brake lights, which once again should not be visible in the rendered scene, coming through the front of the truck."

Well, it’s perfectly true to say that in KYRO’s normal rendering mode MBTR exhibits these issues – it uses the Z buffer in such a way that doesn’t agree with PowerVR’s on chip Z-Buffer; however, these issues could have been circumvented if PowerVR developer support were contacted when the game was being made. Another issue is that MBTR has been coded to lock the entire frame buffer, rather than just portions – this causes issues with KYRO, but it also performs pretty poorly on other cards anyway.

However, one thing that NVIDIA completely overlooked when continuing to rant about MBTR issues (and then claiming that all games will be like this on KYRO – which plainly isn’t the case) is that KYRO has an option in the drivers then enables an external Z buffer, i.e. a Z buffer exactly the same as on any of NVIDIA’s currently available range of cards, which completely cures all issues with MBTR. The only downside of enabling the external Z-Buffer on KYRO is that performance is degraded because the external memory is being utilised, rather then the faster on chip Z-buffer. Interestingly, though, a patch has recently been released for MBTR that improves performance somewhat, meaning KYROI/II owners can play faster with the external Z-Buffer enabled, or even faster by keeping the on chip Z-Buffer active, and just removing the shadows in the game options!
http://patchez.de/index.php3?show=game&id=3

Seeing as NVIDIA wants to be so critical of KYRO based upon this one game, let’s remind them of something shall we – again, taken from Anand’s GeForce3 review:
http://www2.anandtech.com/showdoc.html?i=1442&p=2

"The only issue we had with the GeForce3 that we couldn't work around was that Mercedes-Benz Truck Racing would not run without missing textures. This forced us to remove it from our benchmark suite."

Oh, look at that, NVIDIA’s latest and greatest has issues with MBTR as well – does that mean GeForce3 will look and perform terribly in all games? According to NVIDIA’s logic perhaps we should assume that.

I always wondered why Anand removed this benchmark from the GeForce3 review and not the KYROII review when they both display issues. Having seen exactly what is the issue with GeForce3 it’s not difficult to see why. The following is an actual shot from the MBTR demo running on GeForce3 (Hercules Prophet III no less – remember, it’s an NVIDIA reference design!) and Detonator version 11.01:

And remember, there’s no options in the GeForce3 drivers that cures all its issues in this game like there is with KYRO!

Considering all the conclusions for KYRO that NVIDIA drew from this one game, I assume they are doing the same for GeForce3 – looks like they’ve got a long way to go with development of that as yet…!



Page 12 – "Conclusion"

"GeForce 2 MX is cheaper…"

A 64MB MX from Hercules has the same MSRP as Hercules Prophet 4500 KYROII board. KYROI clearly outperforms MX in many cases, and KYROII has been seen to outperform a GTS Ultra in some, and is close if not better to a GTS in most cases!

"…and more fully featured"

MX: Hardware T&L, Dot3 & very poor FSAA implementation; KRYOII: EMBM, Dot3, enhanced 16bit output, 8 layer multi-texturing, much faster FSAA implementation & Tile Based Deferred rendering (only render what’s onscreen). You decide. Gamers should always have a choice, but it looks like Nvidia wants to be the only player around.

"GeForce 2 has a brand name"

So what?

"GeForce 2 has proven drivers"

So what?

"GeForce 2 is available now"

So what?

"Developers working on NVIDIA GPU’s, not Kyro"

I’m sure that most developers will be using NVIDIA boards, but does that mean to say that newer games will not operate on competitor’s parts? Surely this goes against the point of standard API’s such as DirectX and OpenGL – NVIDIA spent many years bemoaning the likes of 3dfx for promoting proprietary API’s, but now apparently NVIDIA want to go one step higher and promote themselves as the proprietary hardware platform!

However, even though developers may not be ‘working’ on KYRO, it’s interesting to see new games, such as Serious Sam, having features coded that will benefit KYRO’s architecture, such as support for 8 layer multitexturing in a single pass – Serious Sam uses 5 texture layers in many places, which means most NVIDIA users have to resort to 3 passes on each polygon. This is all achieved in one pass on KYRO. Also, as mentioned earlier, even developers who have support contracts with NVIDIA are releasing patches and updates that specifically benefit KYRO, as the recent Beta Giants patch shows.

"Buying Kyro2 is a risk – and when cards and PCs get returned it damages your finances and your reputation."

And now NVIDIA goes further to insinuate that somehow any potential KYRO issues there may be (and remember, based on the evidence NVIDIA has used GeForce3 has equally, if not worse problems) will mean PC’s fail to operate correctly! Astonishing.

Really, you have to wonder why NVIDIA resorts to these type of tactics – with the breadth and depth of engineering talent and knowledge in their company I find it astonishing they have to resort to pathetic smear campaigns such as this. What would possess them to produce a document such as this? I assume it can only be the fact that KYROI/II is attacking their bread and butter market – the low end, and doing it a little more successfully then many people may have suspected.

I suspect that this document has been handed to NVIDIA’s board vendors, anyone who happens to want to stray away from NVIDIA a little to get a little diversity in their product line-up. The question is, if they do decide to look to alternative suppliers, as Hercules has, does NVIDIA take that a step further – there are noises of Hercules removing the MX’s from their product line-up; is this because they purely see PowerVR as a better choice in this segment, or now that they have taken this step is NVIDIA not supplying them with MX chips?

One last thing to ponder – if this is the type of reaction NVIDIA produces when they face competition from a relatively small company in the form of PowerVR, can you imagine the campaigns that may have been running in the background for the likes of ATi and 3dfx? It hardly bears thinking about…or does it...

[This message has been edited by skywalker[TSG] (edited 04-04-2001).]

[skywalker[TSG]]

PDF link does not work

you can acces it via the original page
http://www.geocities.com/johnrreynol.../the_truth.htm

[Barney]

I think the whole site is ****...

I don't believe nVidia made that .PDF. Do you really think they would say:"Kyro II is ****, buy nVidia"? I don't think so, it would only make people buy KyroII instead of GeForce (me anyway). Also all the facts in the pdf file are wrong. Do you really think nVidia would make that many mistakes? And I noticed Max Payne is in the list with Hardware T&L games. It won't be out for another year!

It looks to me like the person who made that website hates nVidia, maybe because they bought 3Dfx or something. He created that PDF to be able to whine and talk **** about nVidia.

Does anyone know the link to this pdf on the nVidia website to prove it isn't fake?

[easyrider]

Umm barney, sorry to break the news to you but it IS real, and Nvidia themselves admited to it
http://www.theregister.co.uk/content/3/18092.html

" The PowerPoint presentation that has been circulating the Internet was an
tool created for the NVIDIA sales team. It was created to help position our
products against competitive offerings, as well as to educate and motivate
our sales staff. It was never intended for public viewing or distribution.
Brian Burke
Senior PR Manager
NVIDIA Corp.
http://www.beyond3d.com/
http://www.3dgpu.com/

That's about all I can find on it.