It looks like Nvidia has canned the GK11x line, replacing it with simple reworking of the fusings of GK10x, something that foreshadows serious internal problems. If you recall the 9800GTX+/GTS250 re-badging scams, you have a good idea of what the next 12 months are going to look like.
As SemiAccurate exclusively brought you last October, Nvidia’s GK11x not including GK110 is barely an upgrade to GK10x. For the same reasons that AMD’s Sea Islands is at best a minor tweak, GK11x is too. No process shrinks, no new overarching architectures, it is effectively a bug fix and a nod to OEMs and their marketing wishes. But that is no reason to cancel an upgrade entirely, and there were good technical reasons to go through with it too.
As Fudzilla pointed out there are “new” GeForce 7xx cards on the short-term horizon. SemiAccurate’s info lines up with his and we can add a few more specifics. The bottom of the line is the new 760Ti and it is based on the GK104-225, a new fusing of the current GK104s. Nvidia OEMs are claiming it is 20% or so faster than the current 660Ti and will launch at Computex. This is not however a new core, the -225 is just different fusing of the current parts, likely a one less cluster left on with higher clocks. At the same shader count and clocks, the “new” GeForce 760Ti is the exact same part as the old.
Luckily there is also a “new” Geforce 770 based on the same GK104 die as the current 680 but in GK104-425 guise. This one is promised for the middle of May, so slightly before Computex with ~25% higher performance than the current GeForce GTX680 and around the same price. The ASIC is once again physically the same as the current GK104/GTX680 just with slightly higher clocks and possibly a few different fuses blown. This is a new bin, not at new ASIC.
Last up is the profit sinking special, the “new” GeForce GTX780 the current Titan with a few more fuses blown. This “new” chip is called the Titan LE and will be less than 10% faster than the GT680 for a significant price premium. Two to three times the die area of a GK104 for $50 more is not a good thing for Nvidia’s margins. This comes at the end of May so the new parts go 780 in 3 or so weeks, 770 a week or two later, and then the 760Ti a week or two after that. The launch schedule is as contrived as the “new” parts are.
There are two significant problems here for Nvidia though, one extremely serious, the other just their usual sleaze in full force. The sleazy bit is like the GTS250 line that was literally a re-badge of the old 9800+ parts like they are doing with the “new” 700 boards. There are no new ASICs, no new parts, just the same old ones with a different shaped cookie cutter used on the die and a spit-shined badge sporting an artificially inflated number. It will be spun as all-new, but it isn’t. Woe betide the ad income on any site that says this though. The technical problem however points to far more serious internal issues.
Note: The following is for professional and student level subscribers.
Disclosures: Charlie Demerjian and Stone Arch Networking Services, Inc. have no consulting relationships, investment relationships, or hold any investment positions with any of the companies mentioned in this report.
Latest posts by Charlie Demerjian (see all)
- More on Intel’s 10nm process problems - Sep 17, 2018
- Intel puts out another 14nm 2020 server platform - Sep 11, 2018
- Why Can’t Intel Supply Enough 14nm Xeons? - Sep 10, 2018
- Intel can’t supply 14nm Xeons, HPE directly recommends AMD Epyc - Sep 7, 2018
- AMD reintroduces the Athlon name with two CPUs - Sep 6, 2018