Slashdot Mirror
GeForce FX Architecture Explained
Brian writes "3DCenter has published one of the most in-depth articles on the internals of a 3D graphics chip (the NV30/GeForce FX in this case) that I've ever seen. The author has based his results on a patent NVIDIA filed last year and he has turned up some very interesting relevations regarding the GeForce FX that go a long way to explain why its performance is so different from the recent Radeons. Apparently, optimal shader code for the NV30 is substantially different from what is generated by the standard DX9 HLSL compiler. A new compiler may help to some extent, but other performance issues will likely need to be resolved by NVIDIA in the driver itself."
Is that the politically correct way of saying "performance sucks"?
Nothing beats my 9500 to 9700 card. Its a simple driver hack. Now my lowly 130$ budget card can whoop any GeforceFX garbage. Plus the overclockability after its a 9700. You just dont get any sweeter.
"he has turned up some very interesting relevations regarding the GeForce FX".
"he has turned up some very interesting rasing or lifting up regarding the GeForce FX" ?
probly revelations would be better.
someone programmed the shaders to work with glide... i can't help hoping 3Dfx will perform some voodoo and ressurrect from nVidia's ashes. excuse me now, i must go stroke my voodoo5
peace,
-Grokent
Did cowboyneal even try to load the URL?x _e.php
http://www.3dcenter.org/artikel/cinefx/inde
www.3dcenter.org does not resolve and whois shows no registration for 3dcenter.org. googling for 3dcenter shows no entry that looks like the right site.
Experience:
GeForce FX is really noisy
Explanation:
It sucks in large amounts of air to keep it cool. This is one of two ways a GeForce FX sucks. The other way is beyond the scope of this post.
man, 3 nvidia releated articles posted in the same day? they must be "donating" some money to slashdot or something... pretty soon its going to say "Nvidia: The way its ment to be seen" under the Slashdot logo instead of news for nerds stuff that matters.
Bah!
Not interested in anything NVidia do or say until they strile some agreement with the people who'se IP they license for their drivers & Open them...
2 Years since I bought my Geforce & I still cant have 3D accelleration, tv out and framebuffer all working at once.
I looked around on some other news sites and can't find any mention of it, is this a scoop for slashdot?
:-(
I am so sad if this is true. He was my favourite author.
Weird timing. I'm currently writing code for a class on microcontrollers. Most electrical engineering students would at some time come across an advanced digital course on microprocessors where one learns about different machine architectures and how to write assembly code for them. Are there any /.ers who have systematically studies GPU chips as part of a class, like say on graphic algorithms or DSP?
Stay sentient. Don't drink bad milk.
Does the FX architecture involve cheating on benchmarks? :)
no comment
What I meant to say was that my ATI card really fills my "AGP Slot". Any of you who are Mac users can touch my "Video card".
Hugs & Kisses! XOXOXOXO
NVidia has much better Linux drivers then ATI. Support 'em.
--
est modus in rebus
From the article - "Because of the length of the pipeline and the latencies of sampling textures it is possible that the pipeline is full before the first quad reaches its end. In this case the Gatekeeper has to wait as long as is takes the quad to reach the end. Every clock cycle that passes means wasted performance then. An increased number of quads in the pipeline lowers the risk of such pipeline stalls."
I understand that the article writers are trying to come up with reasons that the Nvidia part is wasting performance, but this doesn't make sense. No architect in this right mind would ever design a pipeline that becomes full before the first instruction can exit. The means that you are fetching much faster than you are retiring instructions. That means you will always have a pipeline stall at the frontend and you will always be wasting cycles. I think the designers would have checked something like that. You can't afford pipeline stalls to happen regularly.
I'd would just like to add that Matrox cards have by far the best image qaulity in 2d-land. Maybe 3d-land as well. They are slower, but still snazzy.
Caffeine Good
Owens @ UC Davis
Akeley and Hanrahan @ Stanford
ATI 9x owners rejoice, indeed! Even the budget 9200 smokes the 5600 Ultra!
Wow, a driver hack that magically transforms older 3.x ns memory to 3.0 or faster memory AND can still overclock. The number of visual artifacts must be astounding and a sight to see.
In the Windows(argh) world I really couldn't care less about what card to use.
ATI or NVIDIA, it's just a matter of taste and/or faith.
But in the Linux world NVIDIA still rules.
And it's not that NVIDIA's cards are better, but they at least have a descent Linux driver.
The bottom line is: "If you use Linux, the best choice still is a NVIDIA card!"
IMO, undergrads are barely taught how simple general purpose RISC chips work. Things like real-world I/O and priviledged modes aren't really covered that much. This is somewhat understandable as it is very unlikly one would be working as a CPU/DSP/GPU designer without an MS or PhD.
I bought nVidia 5 times in a row, so really I find it quite cool to be back to ATI. I was worried the competition in the market was over.
The key thing ATI did besides great cost/performance was get drivers out the door that didnt totally suck. For the first time in memory I have the original video driver and am not forced to download a patch!
Good job ATI!
Had to suggest it.
The original GeForce FX cards were noisy because the fan cowling was misdesigned and the fan blades rubbed upon it. The later versions have fixed this problem, and thus no longer sound like a leaf blower.
...found out the hard way that just because something has a more recent model number, in no way makes it a better product. After suffering through the miserable peformance of the 5200fx, and even the 5600fx, we came to the conclusion that NVidia should have just skipped the NV stuff and applied that effort to something more worthwhile. I ended up with a TI4200, and my friend ended up with a Radeon card, and as far as I'm concerned, most of the fx line just plain sucks.
Wow, I didnt expect my Anonymous Coward to get modded up. That really warms my heart that slashdot crowd listened to me for once. I might have to sign up for a real account now.
:)Check out ocfaq.com/softmod for more info.
As for you haters out there. It has nothing to do with the memory speeds, memory can be overclocked independently of the core. And no my Infineon 3.3 does not overclock to much. As for the hack itself, it involves opening up all 8 pipelines, as opposed to the 4 default in the 9500. Core can be overclocked trough the roof
I'd like to see that list.
so, you payed full price for a 9700?
From the article:
Er, oh wait, it's in German as well...
Well you can make the GeForce a nVidia Quadro and gain additional OpenGL functionality you generally won't need :-) Although, speed will stay at the same level.
ATI is getting serious with its Linux support from what I saw.
Ant(Dude) @ Quality Foraged Links (AQFL.net) & The Ant Farm (antfarm.ma.cx / antfarm.home.dhs.org).
Fuck you and your hacks.
This article seems to reiterate what everyone has been saying (Carmack, Valve, everyone). The GeforceFX architecture can only be made competitive for 3d engines using modern shaders with herculean effort. This is to be competitive, not dominantly superior.
Honestly, I thought nVidia learned their lesson with the NV1 - don't make weird hardware.
Now, what has to be making GeforceFX owners worried is Gabe Newell's warning that the new Detonator drivers might be making illegitimate 'optimizations' and, furthermore, covering them up by rendering high quality screen captures.
Yeah, they disected an old ATI and found out how they were cheating on Q3 numbers so they figured they'd better join the race and start cheating.
I must say, they are doing an Excellent Job
This absolutely has to be one of the best examples of how the graphic card companies are using the ignorant "tech" sites to spread false stats.
Those guys got an offer from nVidia to do the Benchmarking along with "A new and unreleased nVidia driver" (yeah right !).
And even when the nVidia card smoked the Radeon card by a difference of 70% at a resolution of 1600x1200, these guys didn't even have a single clue about what was happening..
echo '[q]sa[ln0=aln80~Psnlbx]16isb572CCB9AE9DB03273snlbxq' |dc
no text
Do you at all realise you were very lucky?
The 9700 was meant to have an R300 with 8 PS pipelines. (The Pro with faster clockspeeds, both with 256-bit memory bus.)
The 9500 was meant to have a "half-broken R300", with just 4 functional PS pipelines. (The PS pipes take up more silicon area than anything else in there, so a fabbing flaw is statistically likely to appear there -- ATI anticipated that.) (The Pro with faster clockspeeds and 256-bit memory bus, the non-Pro with a 128-bit memory bus.)
They didn't get enough half-broken chips from the fab to satisfy the 9500 demand, so some times they had to insert fully functional R300 chips in the 9500 cards. Exactly those are the ones that can be software converted to 9700 cards. The other 9500 cards just can't be software/hardware converted.
I'd say you were more lucky than 1337 there... supposing you didn't start with the non-Pro 9500, in which case the poor memory bus cripples your card regardless.
Of course, I'm also slightly jealous.
Aarrgh! YABA! And a FLA, no less. And it's not even clear-cut -- is the Free there Gratis or Libre? Shouldn't it be GLOSS, GOSS, or LOSS, hmm?
Your point, however, was good.
They need work they still use a defucted interface just because if they changed they would have to be opensource. Instead of DRI they use GLX.
The 9500-9700 hack is for the very early 9500 cards only. As both where using the refence card they where basiclly the same, but with 4 pipeplines closed.
Ati have now forced the makers to use a different card design, making the 9500-9700 hack impossible on newer cards. However the 9500-9500pro overclock still works, As does the 9700-9700pro overclock.. but the days of a $200 card upgrade for a simple hack are gone.
Oh, but you are wrong my friend. :) Newegg has these old style cards in stock. I bought one last week. I have all 8 pipelines after the soft mod.c atalog= 48&DEPA=1&submit=property&mfrcode=0&propertycode=& propertycodevalue=4396,3668
Notice in the picture the arrangement of the memory chips AROUND the core.
http://www.newegg.com/app/ViewProduct.asp?
That link doesn't work, and a search for 9500 at newegg doesn't reveal such a card today.
'nuff said.
I still won't buy ATI. Sure it's faster, but given their driver quality track record, it's like swapping the engine from a Viper into a Yugo. Wicked fast until you crash.
Yes, I'm an *extremely* unhappy former ATI customer. I will NEVER buy one of their cards again.
retrorocket.o not found, launch anyway?
The only drivers for ATI that allow you to play modern games are their binary-only Catalyst drivers. (One acronym: S3TC) Which blow chunks compared to NV's binary-only drivers.
I believe S3TC is one of the major factors in why BOTH ATI and NV are binary-only. I know it's the reason given for ATI's open-source drivers not being able to run UT2K3. Sadly, there aren't really any acceptable alternatives to S3TC.
retrorocket.o not found, launch anyway?
No it doesn't.
Due to intellectual property issues, there are no open-source drivers that support S3TC.
"working perfectly" implies that it can run a modern game like UT2K3 - Which the open-source drivers can't.
Your only option for UT2K3 (And likely Doom3 when it comes out) are either NV's or ATI's closed-source drivers. And NV's Linux drivers are FAR better.
retrorocket.o not found, launch anyway?
The proof is in the pudding.
The headline under the first diagram is "Die CineFX Pipeline"
DX9 aside, how does the Nvida card compare, to the ATI card, when used in a purely OpenGL based game?
Jumpstart the tartan drive.
Nothing Beats my DIAMOND VIPER!
I did a project a while back using TIGA and the associated chips. You might want to take a look at , FPGA and User Interface Guide.
s try/ARB/vertex_program.txt">ARB_vertex_program and "a href="http://oss.sgi.com/projects/ogl-sample/regis try/ARB/fragment_program.txt">ARB_fragment_program .
It's all obsolete and legacy now. But it gives you a good idea about how a current day graphics card is designed. Back then, the various components had to be implemented on separate chips (eg. RAMDAC's, clock oscillators, memory decoding, graphics).
TI also had the TMS34082 vector processor. You could have up to four of those in a slave/master configuration (a bit like the PS2 VU0 processor). The TMS34020 supported 1/2/4/8/16/24/32 bit pixel sizes and had a parallogram rendering instruction (Two of those allow you to render a triangle). If they had kept the product range going and allowed Moore's law to keep going, they would probably have been able to keep up with 3Dfx.
Intel also has the i860 which combined the floating point and graphics processing onto a single chip. The Intel XEON chip still supports this instruction set.
If you can access the IEEE and ACM archives, you'll find out about dozens more such processors.
Presently, you should have a look at the OpenGL extension a href="http://oss.sgi.com/projects/ogl-sample/regi
Any Google search on these topics will provide an almost infinite list of topics.
It's german for
"The CineFX Pipeline, The."
I'd say the Linux drivers for my ATI card get it done.
Oh, that's certainly some good news! I'd say I've waited for years for it to finally happen!
This stupid fucker didn't deserve to live anyways, and it's always a good thing to see another american idiot die.
I was wondering why VisionTek stopped making NVidia products. Perhaps they had sensed that the company had finally strayed too far from its 3Dfx roots. It puzzled me at the time because FX was getting such publicity. VisionTek is known for leading quality manufacturing, so it makes sense that they would want to support the chipset that allows them to make the most compatible and blazing fast video cards.
Of course pipelines become full before the first one can exit. Ideally, pipelines are full all the time. That's why you have them, so you can have more operations in flight at the same time and more fully utilize your available operational units.
I think the text is poorly trying to refer to the latency produced by a pipeline and is saying that the long pipe causes problems. In essence, it'd be better to do fewer operations a bit faster, even if it hurts overall throughput.