AnandTech's KyroII Review

wascrash · Mar 13, 2001

I will be getting one of these.When are they to be on the shelves?

UKtaxman · Mar 13, 2001

<< Try instead. "We never realized people wanted T&L so we never designed it. Yes it is important but we wont put it in because that would mean pulling delaying it hence pulling a Bitboys >>

I think you'll find they already have the technology aswell as SLI technology working in the NAOMI arcade board manufactured by Sega. I think you'll also find that the T&L unit is far superior to the dross that ATI and Nvidia have in their current cards and that includes the GF3. It looks like they will be announcing a new card at CEBIT to be released around September(if the rumours are correct), could this be the real GF3 killer?
Also it appears that due to it's tile based rendering, it may well be far, far easier to sli the Kyro chips ala 3dfx. With each chip processing alternate tiles, in fact I wonder if this may also make it possible to have a large number of chips running in parallel processing a seperate tile each? This was difficult for 3dfx due to their rendering technique, but the way Kyro renders should make this far easier, or am I just dreaming?
STM have been working on TBR and HSR/defered rendering for at least 4 years now, so they have a huge march on the competition in this field. Anyone remember the original Matrox M3D? Just how far ahead are they? Also as this appears to be just a die shrink, what have their engineers been up to since the launch of the original Kyro?

powervr2 · Mar 13, 2001

hum...
I already said this in other topic ... but you might miss it so here it is :

"And if the developers start to optimize their games for kyro 2 ?

i.e. start not DOING the removal of some of the overdraw via software? what is the use to do that if the card do that in hardware ???? (waste cpu cycles that is)

I think not do something is easy to do ... don't you think ???

we will get even better numbers FOR SURE !!!
let's hope that kyro II will not get forgotten like the rest of the others players...

maybe this is really a cpu cycle saving feature even better than T&L .
Don´t you think ?"

friedchickenbutt · Mar 13, 2001

How are the Win2k drivers?

TrentReznorNIN · Mar 13, 2001

I have to wait for the review on this one. I am running WIN 2K and if this card's WIN 2k drivers are as bad as ATI's then it won't make sense to buy it. Enough of us are running WIN 2K now that maybe they should do some benchmarks for it. Also, the new Windows release will be based on the 2K kernel so WIN 2K benchmarks are becoming increasingily more important. Didn't NVIDIA aquire some rights ot tile based rendering when they purchased 3Dfx? If that is true they could be poised in a position to crucify ATI. Well, it is only a matter of time until all these questions are answered.

powervr2 · Mar 13, 2001

Lack of features???
lol
it has everything that geforce 2 have less the T&L

but it has some features of geforce3 "dot3 bump mapping":

I took this from the st site:

" Best image quality
Full screen anti-aliasing
Internal True Color?
Environmental and Dot3 bump mapping
Exclusive features
8-layer multi-texturing
Z32? comparisons per clock
Stencil buffer always available
High precision Z-buffer always available"

and this more:
"PowerVR? Series 3 arcade performance 3D

8 layer Multi-texturing
Twin high performance texturing Pipelines
Full triangle setup (hidden surface removal, shading and texturing)
RGB gouraud shading and specular highlights
Bilinear, trilinear and anisotropic texture Filtering
Color key and Alpha blended textures
Table and vertex fog
Texture compression
Full scene anti-aliasing
128-bit GUI accelerator

3 operand ROPs
Hardware clipping
Color expansion
Transparent and stretch BitBLT
AGP bus master

DMA bus mastering for minimum CPU load
3.3V PCI support
Digital Video Output

12-bit Multiplexed digital interface
Glueless connection to standard LCD interface devices and digital TV encoders
24-bit LCD modes up to 1280x1024 60Hz

Video playback & MPEG2 decode

Acceleration
Motion compensation
4:2:0 Overlay support
Sub picture blending
X, Y interpolated scaling
Color Keying
SGRAM/SDRAM 128-bit interface

Single memory for frame buffer, video and texture memory
2.4 GB/s bandwidth
16 to 64 MBytes support
Video port

Video port for video capture, TV Tuner, videoconferencing
VBI data capture for Intercast, Closed Caption and Teletext
Integrated palette DAC and clock synthesizer

32x32 hardware cursor
1920x1280 true color at 75Hz refresh "

does the geforce 2 do dot3 bump mapping ???
It does lack however that programable thing of geforce 3...let's hope that kro 3 will be a programable chip... but it has hidden surface removal, shading and texturing ...
IMHO that is better than T&L don't you think ??

powervr2 · Mar 13, 2001

Gigapixel was way behind the kyro team when they were acquired by 3dfx...
:Q
maybe in 3 years we will see a nvidia product tile based .. with lots of royalties paid to the kyro team
:Q

richleader · Mar 13, 2001

OK, which program was used to demonstrate the 4X FSAA? Serious Sam, which happened to be the only program at which it really excelled. Quake III might not be the perfect test bed for video cards, but using Serious Sam as an indicator of the total package? Come on. I'm not blaming anand, it was convenient for him, but he's not the one saying this is the next best thing like some of you. This is just a better voodoo 5500 (OK, a WAY better 5500

), a great deal for the price, but so are many of the geforce 2's that HAVE reached a similar pricepoint, and they are much more scalable in the future, having both T&L and developer support. If you desperately need a vid card right now, the Kryo II is a highly viable option, but to ditch your geforce 2 gts or pro, or Radeon for one? That's idiotic. If you want to throw money away, at least do it on a geforce 3 or windows XP!

powervr2 · Mar 14, 2001

(answer to richleader)

what ?
are you telling that geforce 2 is more scalable because of T & L ?
lol

fact:
why the hell have the geforce 2 to process all that triangles and vertices and shxt if they will not appear on the final image? to use the T & L feature ? (those obscured by others in front)
lol

fact: if the games are getting more complex i.e. more polygones,triangles, etc... the overdraw will increase... thus this kyro will get "more" fill rate and have to process "LESS" TRIANGLES than the "others" don't you think ?

conclusion:
I think that this kyro will get even better if they decide to support T&L (more triagles in games)... of course they have to support T& L via software also

there are good alternative ways to do T & L via software...
by the way "everybody" knows that the geforce's 2 have a mediocre performer T & L ... geforce 3 is great because "THEY" can program the T & L via software lol like a cpu ... to bad that it's only 600 US$

frown;

yap I am those guys that like to endorce and follow the underdogs
VIVA AMD,DDR and kyro 2!!!
lol

powervr2 · Mar 14, 2001

ATTENTION !!!
the MBTR runs ok with the kyro II card ...
I saw on a beyond3d forum topic...

the reviewer forgot to enable external z-buffer on the drives ...
is that simple

darth maul · Mar 14, 2001

Is anandtech still using 1.17 for quake3? And thus demo001? Or did they go to 1.27?

Ummmmm, every competition orientated group (OGL, CPL, etc.) has gone to the 1.27 patch and thus demo127 would be much more relevent to what fps these cards are pushing then demo001 on 1.17. As a player that takes his Q3 playing very seriously, I can only hope anand has or will soon, only give us benchmarks for demo127. And I would love to see them bench with a "professionals" Q3 config file, but thats prolly asking way to much.

PeAK · Mar 14, 2001

Ben/Luke, regarding your comment:

<< You forgot the by far worse offender of this, ATi. They pushed their RagePro "Turbo" drivers as boosting performance ~30%-40% even though it was only in WinBench98 and overall the drivers were slower in actual games(where they were ~20% slower in some cases) >>

The skinny on all of this was that Winbench disabled waiting on VSYNC during its tests and that games (at that time...3 years ago) defaulted it on as recommended by MS. The problem...the driver waited for two successive VSYNCS before updating the screen contents.
So, a 20% hit for such a horrendous coding error somehow gets hailed as a landmark article titled "3D Winbench 98 - Only a Misleading Benchmark or the Best Target for Cheating ? ". Funny how uninformed outspokeness in the internet fast food world gets readily digested. In the end, Winbench was vindicated with its 30-40% speed improvement prediction when the driver was updated to remove this coding error and the above hardware site went on to become the McDonald's of hardware sites...all courtesy of the stroke(s) of a keyboard somewhere deep in the dungeons of ATI, 2 graphic generations ago

With the recent release of 3DMark2001 and some of the warped effects that "weighting" had on 3DMark2000, I think people will begin to see the value of dissected benchmarks making a come back in relation to DX8. Winbench 2001 anyone?

I like to see the KyroII in an updated Dreamcast Box or perhaps a X-box clone.

wascrash · Mar 14, 2001

When is it hitting the shelves? My Ford profit sharing money is burning a hole in my pocket.I have the Kyro now and love it.Very nice card and the image is top notch.

Finsti · Mar 14, 2001

On the Hercules site it said Mid-April I think.

oldfart · Mar 14, 2001

darth maul, I've asked about this in previous articles. They are still using Q3 1.17 and UT 4.32. 1.27 and 4.36 have been out for some time. Also, where is the UT 1600 x 1200 x 32 test?

BlvdKing · Mar 14, 2001

OVERCLOCKING! I want to see if this card overclocks since it came with 5ns 200mhz RAM! I bet this sample card could do 200mhz+ and production cards from Hercules would be good for at least 200mhz with 5.5ns RAM. I think overclocking would put this card over the top (arm wrestling anyone?)

RoboTECH · Mar 14, 2001

the reason they maintain 1.17 is so that their old results will be compatible, and they won't have to redo all their benchmarks

if you're a "serious" competition Quake3 player, then their results are irrelevant to you anyway.

Get a GeF SDR and run @ 512, since they all run the same speed anyway @ that res/detail level

as far as UT, well, using it as a benchmark indicative of DX7 is silly, always has been. They should use Evolva.

BenSkywalker · Mar 14, 2001

Robo-

At first I was thinking I shouldn't respond to some of your comments cuz that would be thread crapping..... then I recalled I started the d@mn thread so I can crap on it

"MBTR aside (is it REALLY that stressful of an engine, or is it just coded poorly?)"

~60FPS on a GF2 Ultra @1024x768 32bit color sounds reasonable given the quality of the graphics. I'll say this, it runs a he!l of a lot faster then Giants

"I really wish they would've done some tests in UT with the high-res textures. Can the Kyro-II manage that?"

Yes.

"also, Ben, I noticed something funny....these new benchmarks you've been rallying about (MBTR and Serious Sam)....I noticed that the 5500 seems to do pretty well in comparison to the MX, don't you?"

Yep, I noticed that

What was slugging it out with the GTS and Radeon in 1999 games is now fighting with the budget brothers in 2000/01 games. Hmmmmmm...... future proof anyone

"I really hate the fact that the tester doesn't have the sense to disable T&L for the cards that don't have T&L units in MDK2. that automatically drops your framerate significantly, who the hell plays like that?"

It looks a decent amount better and is playable, why wouldn't you? 100+FPS doesn't do a ton of good in MDK2, for actual gameplay even if I had a V5<shudder

> I would definately enable hard T&L. This does bring up another point though, we can now look forward to */compromised benchmarking being a thing of the past. Why doesn't AnandTech use Q3 HQ settings? Because the V5 can't do it. Why did 3DMark2K use 16bit as default? Because 3dfx's at the time latest and greatest couldn't handle 32bit. Why does the V5 have an asterik next to it in MBTR? Because it couldn't run the test like all the other cards could. Then there is the MDK2 T&L situation as another example. The Rampage would have rectified this situation, but speaking for myself I'm sick of benches having to dance around the shortcomings of one companies boards. When the V5 is dropped off the charts we can get away from having 3dfx friendly or asterik included benches

Edit- Teach me to have eight different IE windows and OE open doing multiple things at once, you replied while I was doing something else

"as far as UT, well, using it as a benchmark indicative of DX7 is silly, always has been. They should use Evolva."

D@mn right they should

powerVR2-

I can understand your enthusiasm, but you have a few misconceptions

"it has everything that geforce 2 have less the T&L"

Doesn't have register combiners either.

"but it has some features of geforce3 "dot3 bump mapping":"

The GeForce1 SDR had Dot3. On top of that, right now Dot3 still isn't working properly on the Kyro boards(driver issue it appears). It does..... something, but it doesn't look like either the reference rasterizer or the GeForce/Radeon(which both look like the reference rasterizer as they are supposed to).

"what ?
are you telling that geforce 2 is more scalable because of T & L ?
lol"

I'll tell you this, the GeForce2 handles high poly loads quite a bit better then the Kyro does(traditional versus "tilers" overall for that matter). Every tiler to date has issues when dealing with high enough geometry loads. This is one of the reason many people were quite intrigued with the Gigapixel technology, they had reportedly worked around the issues that have been part of the technology to date.

"fact: if the games are getting more complex i.e. more polygones,triangles, etc... the overdraw will increase... thus this kyro will get "more" fill rate and have to process "LESS" TRIANGLES than the "others" don't you think ?"

No, the Kyro must process every single poly and it is possible it will have to handle a great deal of them several times. Simplified- The way the PVR chips determine visibility is by using the geometry data before applying textures/rasterizing(by "shooting a beam" that stops when it hits something). Without handling every triangle first, the PVR wouldn't have any way to know what is visible. The geometry data is placed in "buckets"(to borrow Dave's term

), it all has to be handled. It is possible to require the same geometry data multiple times as the Kyro breaks a scene down into tiles based on pixel size/screen area, not geometric amounts.

"by the way "everybody" knows that the geforce's 2 have a mediocre performer T & L ... geforce 3 is great because "THEY" can program the T & L via software lol like a cpu"

In terms of strictly performance, a static T&L engine(GeForce2) is faster then a flexible(GeForce3) one. You think the 3DMark2K scores that show geometry throughput are a joke? Now, things change considerably when you add other factors in the picture, but in terms of raw performance the GF2 is certainly not mediocre by any stretch of the imagination. In fact, its' only competitor in that area is the Radeon which is absolutely destroyed in terms of T&L when facing a GF2, still gets beaten quite badly by the GeForce1 for that matter.

The Kyro2 is definately a very interesting product, the performance looks extremely impressive and the price is real hard to ignore.

PeAK-

All sounds nice in theory, I still own, and use, mine(paid $320 for my 8MB All In Wonder RagePro) and that performance boost never materialized in games. In fact, the fastest drivers I have seen to date for gaming are the ones that shipped on the included driver CD

RoboTECH · Mar 14, 2001

Ben:

regarding MBTR - very cool stuff there, I like that. that is one of the games that makes me want to grab a GTS-U, that game REALLY needs anisotropic, wish the Ultra could do better than 8-tap, which is just a tease!

also gotta figure out what anand was talking about with the 24-bit thing, I was getting ~30 fps @ 1024x768x32 w/24-bit textures checked, and a few other things checked beyond default (can't remember off-hand)

I really wonder which driver they selected in the driver area of MBTR, that made a pretty good difference, IIRC

regarding Giants - does it REALLY run that slow??? gadzooks! enable all the options in MBTR, and it'll slow EVERYTHING to a crawl, and Giants is SLOWER??? yikes....

does the Kyro do OGL or D3d in UT with the compressed textures? I've heard OGL looks better than D3d w/the compressed textures, and lemme tell ya, that looks DAMN GOOD!

as far as future proof is concerned, never said it was more future proof than the GTS, I don't think anyone claimed that. I was just a bit taken aback from when we were arguing about the MX being more future-proof than the 5500.

as far as HW lighting in MDK2, it does look a bit better, but I didn't think it was anything spectacular. In MDK2, I found 1024 w/2xFSAA looked much better than anything I could get on the GTS Pro I had.

As far as compromised benchmarking and 3dMark2000, well, you have your opinion as to why they used 16-bit as a default. I have a different one.

Notice how much closer the scores are @ 32-bit. the GTS is still ahead, but the difference is MUCH MUCH less. I've always bitched about HW T&L using too great of the memory throughput of a card that is already throughput limited. Remember that discussion?

Soo...how can you get past that? Why, enable 16-bit, of course.

as far as Q3 HQ, I hear what you're saying, but the lodbias slider makes a far greater visual impact than enabling trilinear filtering in Q3, by a long shot, so that doesn't worry me too much, although it would certainly make benchmarking a pain, and I agree, it's stupid that trilinear wasn't an option in multitexturing mode.

and as far as having an asterisk, how about not registering at all? (MX @ 1600x1200x32) I just noticed that. <G>

Now then, if SOMEONE could PLEASE explain the FSAA results of the MX in Serious Sam vs. the 5500, I'd be very interested.

We all know the MX sucks in FSAA. Seeing the regular benchmarks indicate that the 5500 is slightly faster (Very slightly) in SS, yet the MX whups up on the 5500 in FSAA??
PUHHHHHLEEZE!

talk about BS. sounds to me like nvidia's infamous "loosely defined FSAA" took over yet again.

BenSkywalker · Mar 14, 2001

Robo-

"regarding Giants - does it REALLY run that slow??? gadzooks! enable all the options in MBTR, and it'll slow EVERYTHING to a crawl, and Giants is SLOWER??? yikes...."

About half as fast, no joke(~35FPS average on a GF2U with GHZ TBird). Why do you think I make such a big deal out of the 50% boost hard T&L gives you

The animation in the game though makes 30FPS look real good, models move more like an actual living being would then the "moving pixels" in Quake3 or UT.

"does the Kyro do OGL or D3d in UT with the compressed textures? I've heard OGL looks better than D3d w/the compressed textures, and lemme tell ya, that looks DAMN GOOD!"

Both. Yes, it does look better(and plays hella better) under OpenGL then D3D, but nothing like moving from non S3TC to having it.

"as far as future proof is concerned, never said it was more future proof than the GTS, I don't think anyone claimed that. I was just a bit taken aback from when we were arguing about the MX being more future-proof than the 5500."

Let's bring this up again in another six months

The benches used, while much closer to actual current games, still aren't anything like Giants. The MX is now closing in on the V5 as features and poly move onwards, the GF2MX also has a T&L engine that is basicly twiddling its' thumbs even with the most demanding games out right now. The gap has now closed, my guess is the current momentum keeps moving in the same direction. Right now, just for gaming, I would pick the GF2MX over the V5 even if 3dfx were still around. The games I am playing the most, Sacrifice and Giants(and soon to be B&W) benefit quite a bit using hardware T&L.

On the 3DMark2K subject, I know you are too smart(and old enough

) to fall in for that MadOnion/nVidia conspiracy BS. Back in the beginning of 2000 3dfx wouldn't have been able to have any boards present if MO went with 32bit default. I have been saying over and over to run the d@mn bench in 32bit and screw the default and those that did have seen what happens(all of the sudden the playing field evens out, even in 3DMark2K

).

"and as far as having an asterisk, how about not registering at all? (MX @ 1600x1200x32) I just noticed that. <G>"

32MB board, notice they don't even report UT scores for 16x12x32bit

"Now then, if SOMEONE could PLEASE explain the FSAA results of the MX in Serious Sam vs. the 5500, I'd be very interested."

OK- AGP texturing

Figure out the framebuffer needed for that setting let alone the RAM needed when combined with the huge textures of SS- it makes sense to me(no I'm not absolutely certain). You have seen the Quake3 scores with the V5 without TC, you know what happens. SS supports TC but it also uses a lot more and better quality textures then Quake3. Also, don't know if you have run the bench but there are a he!l of a lot more textures on screen at once then any Quake3 bench likely could ever show. The MX also chokes hard, only 3FPS faster(though that is a huge percentage boost

), to me that could indicate some texture swapping going on.

RoboTECH · Mar 14, 2001

Why do you think I make such a big deal out of the 50% boost hard T&L gives you

50% boost?

Ben, PLEEEEEEASE tell me you're not talking about the comparison on Rev's page, athe one with Giants, are you?

aside from the -2 lodbias on the 5500 which enables a huge hit on the 5500, (-1.5 is far faster and almost as good looking)

is Giants an OGL game? That would explain the framerate "hit", not the T&L (tho the dot3 looks cool as heck)

Let's bring this up again in another six months

bah! 6 months from now, that's several lifetimes for me! I'll have a next-gen part. Both cards will be stupid slow. I'll have a GTS-U, at least, by then.

the GF2MX also has a T&L engine that is basicly twiddling its' thumbs even with the most demanding games out right now.

and it will keep on twiddling it's thumbs, because the GF2MX doesn't have anywhere near the memory throughput to handle the high detail with worthwhile resolution.

I have been saying over and over to run the d@mn bench in 32bit and screw the default and those that did have seen what happens(all of the sudden the playing field evens out, even in 3DMark2K

agreed there. I've ranted on that subject several times myself. Sometimes, we see eye to eye, especially when you're correct.

and as far as Serious Sam is concerned, NO WAY IN HELL would the MX outperform the 5500 w/equal settings in FSAA in ANY game, especially not one where their performance is relatively similar. Even the GTS takes a greater %age framerate hit with FSAA enabled, how is something with far less memory throughput going to do BETTER?

nope. something was amiss there, no way, no how, nope nope nope...

darth maul · Mar 14, 2001

RoboTECH LIES!!! I get about 150fps with my config at 800x600 (on demo127), even if I droped to 640x480, my 32meg GTS will still be beating a GFSDR with plenty of room to spare.

And my goodness, who cares about comparing to OLD benchmarks, OLD bencmarks are just that, OLD!!! If no one progressed with their benchmarks we would still only be using bytemark. Thats a lame answer, IMHO. Anand used to be cutting edge in their benchmarks, and oooh even testing cards and motherboards with overclocked processors, but alas they have fell into the way to mainstream of sites. There is a big gapping hole out there in review land, just wish someone would fill it. Again thats just my stupid lamo opinion.

RoboTECH · Mar 14, 2001

look maul, i'm sorry ben fu(ked you up so badly, obviously you are still kinda upset about that.

once you get over your anger, why not calm yourself down a bit and think....you don't spend a $hitload of $$$ to run @ 800 or 640

and an SDR can hit 150 fps easily @ 8x6 w/vertex lighting and r_picmip 5, all the visual shiznit turned off

hell, my 5500 can hit that @ 1024x768, duh?

I was getting well into the 200s with a GTS Pro, but about 150 is where the line gets drawn in q3 (mainly only 2 places that I know of)

BenSkywalker · Mar 14, 2001

Robo-

"50% boost?

Ben, PLEEEEEEASE tell me you're not talking about the comparison on Rev's page, athe one with Giants, are you?

aside from the -2 lodbias on the 5500 which enables a huge hit on the 5500, (-1.5 is far faster and almost as good looking)

is Giants an OGL game? That would explain the framerate "hit", not the T&L (tho the dot3 looks cool as heck)"

Not sure what you are talking about, honestly haven't been to the Pulpit in a while(he was talking about it on the B3D forums). Rev's comparison was a GF2U with hard T&L on and off, no V5 involved. Giants is D3D, and in the game when you turn T&L off you also disable Dot3. The game runs ~50% on his GHZ system with Dot3 and T&L on then with both off(big difference between it and Evolva for Dot3).

"bah! 6 months from now, that's several lifetimes for me! I'll have a next-gen part. Both cards will be stupid slow. I'll have a GTS-U, at least, by then."

You will, but that doesn't mean anything for those people who were looking for a card to last them a year or two who were reccomended to pick up the V5.

"and it will keep on twiddling it's thumbs, because the GF2MX doesn't have anywhere near the memory throughput to handle the high detail with worthwhile resolution."

For me or you, there are still quite a few people around here that are running V3s and TNT2s, what about those type of people holding on to both?

"and as far as Serious Sam is concerned, NO WAY IN HELL would the MX outperform the 5500 w/equal settings in FSAA in ANY game, especially not one where their performance is relatively similar. Even the GTS takes a greater %age framerate hit with FSAA enabled, how is something with far less memory throughput going to do BETTER?"

2048x1536x32bit color with a heavy texture load the GF2MX could have a significantly higher memory bandwith then the V5.

PCI texturing= 133MB/s

AGP 4X= ~1GB/s

Try running any test on the V5 with over 64MBs of textures/frame buffer combined useage and compare it to the GF2MX and see what happens. Run through Quaver at the highest res you can and disable texture compression on the V5 and see what happens, the board will choke very hard, moreso then the GF2MX.

The effective memory of the V5 is quite a bit lower at that setting then any of the other 64MB boards. If you look at the scores for the GF2Pro and the GF2Ultra there is only a ~5% difference between the two at 10x7x32x4 even though they are seperated by ~18% at 1600x1200 32bit color. But, when they exceed local memory they would both be stuck with the same bottleneck which would make sense to me.

RoboTECH · Mar 15, 2001

Not sure what you are talking about, honestly haven't been to the Pulpit in a while(he was talking about it on the B3D forums). Rev's comparison was a GF2U with hard T&L on and off, no V5 involved. Giants is D3D, and in the game when you turn T&L off you also disable Dot3. The game runs ~50% on his GHZ system with Dot3 and T&L on then with both off(big difference between it and Evolva for Dot3).

well, hot damn, I am impressed.

it'll be interesting to see if Kyro2 has a worthwhile T&L unit. Hope so.

RE: the MX and AGP texturing, I would think framerates would be even lower than they were if they had that much texturing to do. I"m not half as disappointed iwth the 5500's scores as I am surprised by the MX's scores (FSAA). I still think they're wrong.

AnandTech's KyroII Review

Junior Member

Senior member

Senior member

Member

Member

Senior member

Senior member

Golden Member

Senior member

Senior member

Platinum Member

Member

Junior Member

Junior Member

Lifer

Golden Member

Platinum Member

Diamond Member

Platinum Member

Diamond Member

Platinum Member

Platinum Member

Platinum Member

Diamond Member

Platinum Member