One of these days is none of these days - English proverb
|
|
|
Intel's G35 failure needs a fix
TO SAY INTEL has botched graphics over the past few years is being kind. Now, the G35 is shaping up to fail in all the same ways as its predecessor, the G965.
If you remember, the G965 was promised to have features such as T&L in hardware and other shader model levels that it simply did not deliver. They were coming, it was said, just wait for the next revision. That revision was delayed, sometimes to make it compatible with the Broken OS [Charlie means Vista here folks, Ed.], sometimes to prioritise video, sometimes because the moon wasn't in phase with Venus.
In the end, the G965 has yet to deliver the promises made at launch. That launch was at Computex 2006, June. As of this writing, it is Q2 2008, and the launch features are still not implemented. What a shame.
G965 was heralded as the first of a new programmable graphics architecture, flexible, fast and power sipping. Intel didn't add 'works' to the list, and it is badly missed. The first spin, called G965 was supposed to do DX9 with all the Shader Model (SM) goodies in hardware, and as soon as MS got the Broken OS out the door, DX10 would soon follow.
If you are still waiting for the G965 DX10 drivers, well don't. Due to unfixed bugs in the architecture, it is never going to happen, but at least you got your DX9.0x. Newer variants of the 965, specifically the 965GM and it's relatives did have the bugs patched, and will have DX10. Someday. Maybe. We hope. Next driver release. Unless...
A year or so after the G965's release, it was supplanted by the G35, which other than changing the naming scheme, offered a slightly wider and faster GPU core. Ironically, it came out before the G965 had it's drivers in the pseudo-working state they are in. Yes, G965 never worked right for it's entire lifespan, and has only been mostly posthumously patched.
Back to the G35. That part was a second generation of the architecture, fixed bugs, faster and wider. It came out months after the Broken OS, so DX10 should have been there out of the gate, right? Should have been, but wasn't. DX9 with all the SM goodness as well, right? Well, hell no. They just got much of that working a month or so ago. DX10 is promised in the next driver revision - quite soon from what we are told.
There are two problems with this. First, is that they have been promising the next revision panacea for two years, and there are more confirmed checks in the 'will never work, sorry' column than the 'works' column. Call me fatally skeptical here. Second is that the G45, the successor to the G35 will be out in within weeks of the 'working' driver.
Yup, you read that right, the G35 is going to be the second Intel GPU to utterly fail, it has really good odds of being supplanted before it ever works. Will it work? I really think this one will, but I have only been wrong eight or nine times before when thinking that. Intel has the sort of track record that rivals AMD's Barcelona promises, just twice as long.
G45 is set to be released in about a few months, in the same quarter as the 'working' G35 drivers. It is in effect a slightly wider and faster G35, and Intel has promised about 1.7x G35 performance. If you consider that G35 gets about 650 on 3DMark06 with 'well tuned' drivers, that would put G45 at about 1100. If the drivers function.
If you recall, I said that I was indeed hopeful that the G45 would work more or less right out of the box. Why do I make such obviously marginally sane claims? Because the G45 is a direct descendant of the G965, and almost all of the driver work can carry over.
If Intel only started work on the drivers from the day of the chipset's public release, it has had two years to fix things. Likely as it looks, moles deep in the Intel GPU division tell me that this is not the case, they were working on drivers before the G965's release. Gasp.
The same holds true for the upcoming G43 and G41, instead of putting them on an older i945 based cores, the budget chips are actually on the newer programmable G965 cores. So, if the G45 works, so should the 43 and 41. If it doesn't, well, there is the next revision, and Windows 7 is right around the corner in Intel driver years.
In the end, it all comes down to the fact that Intel can't get a driver out the door to save their lives. The GPU cores are pretty OK, they do the job, they aren't Nvidia level laughable on power, and as long as you don't look at 3D, they are mostly functional.
Intel was partially hobbled by an architecture that didn't work out; programmable is fine if you have enough horsepower to to the job. But sources tell me the major problem with the early iterations of the architecture was that there was simply not enough horsepower. This bodes well for the G45 working, but casts a long shadow over G43 and G41.
Scrambling to make up for it with driver tweaks was a losing game, so things never worked. Distracting people from making the promised release functions to add Broken OS support and video tweaks was simply stupid. If you are thinking gross management failure along with technical failures, you may be right there too.
So, short story, G965x never worked before it was supplanted. If G35 works before it is supplanted, it will be by weeks. G45 has a shot, but we have heard that before. What a mess. µ
Tags: Intel
Share this:
Share
A 65 watt Deneb @ 3.0 may not equal Nehalem, but it will be a godsend on a 780G board for casual gamers. Both ATI and Nvidia have decent IGP's in every generation. Were it not for Intel CPU's on Intel motherboards, they wouldn't get such a large share of the integrated market.
Not looking promising:

http://en.expreview.com/2008/04/09/eaglelake-a-3-raise-g4345-graphic-clock-to-800mhz/

The Eaglelake developing is still on track, and the release date is unchanged, according to Intel. But the in initial launch the G45/G43 is in A2 status, graphic core running at 667MHz, and G45 do not support VC-1 hardware decoding. 

After A-2, Intel will keep improve the Eaglelake and release A-3 version in the future. In this revision, not only some small bugs will be fixed in G45/G43, but also the clock of the graphic core will be raised to 800MHz, and VC-1 hardware decoding will be enabled for G45.

Though Intel have already sent out their chipset sample, A-3 sample still TBD. Intel will begin mass production of P45/P43, G45/G43 in the week 16, and the graphic driver release for partner will be a week later. The developing of the graphic driver is not satisfying, because in the launch day the driver will only support OpenGL 1.5, OpenGL 2.0 still unsupported. you have to wait till Q3 for a new driver release.
And don't forget that Intel can be almost exclusively blamed for the failing of OpenGL. 

Chicken-and-the-egg, you can't use shaders in casual games if the gamers don't have shaders. And what GPU is in most casual gamers systems?
I've got an Intel integrated graphics thingy in my laptop and when I test some simple DirectX on it, I find it sometimes plots polygons only if all of the corners are off the screen, it's like some clipping check that is reversed and horribly wrong. I thought it was my programming, but having read about this mess it looks possible it's a driver error. Works fine identical program on a desktop.
Forget speed, DirectX, how about some trivial functionality. I can't even get my laptop to drive the ubercommon 16:10 widescreen LCD monitor, the drivers are restricted to the standard 4:3 5:4 resolutions. I boot into GNU/*/Linux and it works automagically. How many people are working on the intel windows video driver? I bet it's many more than in the Xorg division. Pathetic.
Making GPUs is tough, tough work. They are actually more complex and have more transistors than CPUs do, which surprises people who think that "the CPU is the brain of the computer" or some other such nonsense.

So Intel has to eat some humble pie because of their pitiful GPU efforts. Good; it may keep them from getting too arrogant.
USB 1.0 on cocaine?
I have a better idea bunny! would you like to analyse how the AMD 10% cut will save AMD? I look forward to that story chua lee. 

Flush
and the problem is that, this lousy Chips takes almost 40% of the graphics market share!!!
shame on intel they got to make a working integrated chip..
or may be easier to let the professionals do it.. why not take a license from AMD or nvidia to one of their great integrated chips. but hey are you nuts, it's intel baby!! the best chip company in the world!!!
Didn't the i740 win loads of (synthetic) Benchmarks when it first came out? I remember almost buying one since it seemed quite a bit faster than the other cards at the time..well, at least until it was benchmarked using games.

Interesting link, btw:

The i740: Intel's Lead Architect Talks
http://www.gamasutra.com/view/feature/3264/the_i740_intels_lead_architect_.php

Certainly funny to read it in retrospect :-)

Then again: What does Intel need to care: Aren't they market leader as far as integrated graphics go?
not sure why everyone is talking about the i740 wasn't that a fully featured add in graphics card and not a chipset bound super wimpy peice of crud. Different categories entirely

P.S. When I say fully featured add in card I mean add in board bound super wimpy peice of crud.
Quote: "G35 is going to be the second Intel GPU to utterly fail".

*Third*, surely? Or has everyone forgotten the i740?

(Broken hardware, check. Endless streams of patched drivers that never properly worked, check. Finally dumped the whole thing in ignominy, check. Yup, it all seems to fit.)
So with all these problems, how on earth will Intel successfully deliver competitive graphics to compete with Nvidia and ATI with Laterbee? A few things you could add to the story is the G915 Vista debacle and yes, the famed i740 joke back in the 1990s.
If you look at Intel's own motherboards they don't even offer a G35-based board, at least in the US! That is telling.

AT adds: Perhaps a visit to the optician is in order.
I like Chalie Demerjian's writing way better than his big brother Charlie. Please fire the one for the other.
Intel has an unbelievably rotten track record with graphics hardware. It's unbelievable, because many small startup graphics companies have done a better job with their hardware than Intel has. Intel simply has no excuse to be producing products that have such lousy drivers. I'm an ISV developing 3-D applications, and the users that have Intel graphics are the ones with the most ongoing problems. It has been going on for years now... Anyone that believes that Intel's going to be a threat to AMD or NVIDIA with Larabee would be well served to take look at Intel's track record with GPUs to date. When I hear the marketing speak saying that Larabee is going to be more popular with developers because it uses x86 instructions, I have to wonder if the Intel people have any clue at all what really matters for graphics. I don't know about anyone else, but I have zero interest in writing x86 assembly language. We've been using DX and OpenGL for decades now, Intel should show us that Larabee will have actual functioning drivers that work correctly for a change. Failing that, Larabee will be doomed to the same fate as all of the prior Intel graphics products. They really must get with the program on this one....