just came across this less than stellar article about the GTX295. from The Inquirer NVIDIA IS SET to trickle out the latest batch of 55nm parts. Expreview has some pictures and tidbits about the latest 55nm GT200/GT200b here, and some GX2 info here. It looks like the on-again, off again GT200GX2 is on again, and it is called the GTX295. Yay. The 55nm parts, internally code named GT206, are finally trickling out like we said they would, with no speed increases, and no power gains. What should have been a simple optical shrink is turning into a totally botched job, with the 'real' 55nm parts unlikely to come out until late January at the earliest following yet another spin. Given the lack of gains with the B2 stepping, the GX2/GTX295 still seem unmakable in volume, but such trifling concerns have never stopped Nvidia in the past. We hear they are going to launch it even though they can't make it, along with the requisite 19 parts to Newegg so they can claim it is on sale. Real volume won't happen until (if?) they can fix the power problems. We hear that the 'launch' is likely going to happen at a shindig on the 12th of December so they can claim the win they promised before the end of the year. One has to wonder if cherry picking parts in an attempt to use tame press to snow the public is the definition of 'Whoop-ass'? I am sure they will claim a stunning victory in any case. One way you can tell how screwed up the chip is is the use of a heat spreader and a stiffener (the metal ring around the chip). If you have a big die, you need mechanical support for it, or it can crack or break bumps. A stiffening ring is usually the cheapest and most efficient way to go, but in many cases, a heat spreader will do the same job. The problem with a heat spreader is that it introduces two additional thermal barriers, the paste under the lid and the lid itself, to the cooling of the silicon. Each one makes cooling incrementally less efficient, not to mention material and assembly costs. You don't do this unless you have to. If you are wondering why every modern CPU out there has one, the answer is simple, so ham-handed monkeys like most DIY people don't crack the die when they clamp the heatsink on. Think AMD K8 here. CPU makers think the cost of a spreader, and the reduction in performance it brings, is worth the protection it gives. GPUs however come assembled. Factory robots don't break chips, so the mechanical protection is not an issue, but the costs remains. So, why did Nvidia do it on the GT200? They can't control hot spots. The lid is a heat spreader, and it helps keep chips with poor hot spot control alive and working. When you see a heat spreader on a part that comes assembled, it is a pretty sure sign something is wrong thermally, it simply is not worth the cost and performance drop otherwise. Make no mistake, the spreader and stiffener combo on the GT200b is a bad bad sign. Why is the GT200b such a clustered filesystem check? We heard the reason, and it took us a long time to actually believe it, they used the wrong DFM (Design For Manufacturing) tools for making the chip. DFM tools are basically a set of rules from a fab that tell you how to make things on a given process. These rules can be specific to a single process node, say TSMC 55nm, or they can cover a bunch of them. In this case, the rules basically said what you can or can not do at 65nm in order to have a clean optical shrink to 55nm, and given the upcoming GT216, likely 40nm as well. If you follow them, going from 65nm to 55nm is as simple as flipping a switch. Nvidia is going to be about 6 months late with flipping a switch, after three jiggles (GT200-B0, -B1 and -B2), it still isn't turning on the requested light, but given the impending 55nm 'launch', it is now at least making sparking sounds. The real question is, with all the constraints and checks in place, how the heck did Nvidia do such a boneheaded thing? Sources told us that the answer is quite simple, arrogance. Nvidia 'knew better', and no one is going to tell them differently. It seems incredulous unless you know Nvidia, then it makes a lot of sense. If it is indeed true, they will be chasing GT200 shrink bugs long after the supposed release of the 40nm/GT216. In fact, I doubt they will get it right without a full relayout, something that will not likely happen without severely impacting future product schedules. If you are thinking that this is a mess, you have the right idea. The funniest part is what is happening to the derivative parts. Normally you get a high end device, and shortly after, a mid-range variant comes out that is half of the previous part, and then a low end SKU that is 1/4 of the big boy. Anyone notice that there are all of zero GT200 spinoffs on the roadmap? The mess has now officially bled over into the humor column. ouch!! He doesn't paint a pretty picture.