IT LOOKS LIKE Tweakers.net has spilled the beans of the upcoming Evergreen mobile parts from ATI. While that is the best recent leak, it is far from the only news.
According to a post by The Source, there are nine separate mobile parts inbound, three GPUs with three variants each. The family is called Manhattan, the chips are Broadway, Madison, and Park. Each has high end XT, mid range Pro, and low end LP variant, but some of those names may change before you see them. Since they are Evergreen parts, they obviously are all 40nm, DX11 and should be notably faster than their mobile M9x predecessors.
Power is said to be 45-60W for the GDDR5 Broadway XT, dropping to 30-40W for the Pro, and the GDDR3 based LP takes only 29W. Madison uses GDDR5 for the 20-30W XT, either GDDR3 or 5 for the 20-25W Pro, and 15-20W for the GDDR3 based LP. Park goes down from there, 12-15W for the GDDR5 XT, 10-12W and sub-8W for the Pro and LP respectively, both of which use GDDR3. As with the 7xx parts, the memory controller will allow vanilla DDR3 in place of GDDR3 as well.
The Source claims that Broadway and Madison will be released in August, Park in November. Since they are notebook parts, that means you probably won’t see either until the back to school notebook refresh a month or so later. Then again, since they are said to be pin compatible with M9x, you might just see one or two on sale before the desktop Evergreens. Stranger things have happened.
To change gears quite a bit, you may recall we said that the Nvidia GT215 had gone *poof* from the roadmaps a while ago. According to Expreview, the chips are now slated to come out in December, woefully late. Our sources back this up as well. The parts are likely to be called GT230, and complement the ‘why did they bother?’ GT210 and GT220 that were recently snuck out the back door.
The problem is that if Nvidia can’t make a barely over 100mm^2, low clocked, low power chip on TSMC’s 40nm process, what chance do they have of making a 400mm^2+, high clocked, high power one? If you know anything about semiconductor yields and bin splits, you know the answer to this is more or less zero chance.
That brings us to the last point, the GT300 has still not taped out. Nvidia missed their internally promised mid-July date, and for now, nothing is on the horizon. Will this part make it at all? Will it still be relevant when (if) it does? If not, can Nvidia survive?
No matter what happens, at this point, there isn’t enough time to go from tapeout to anything more than a few hand-built publicity stunt parts before 2010. I wonder if anyone saw this coming? At this point, given that the tape out has been delayed by 6 months or so with no end in sight, any respins will add about 8 weeks to the tally. If it tapes out tomorrow, and needs only one respin, don’t looks for parts on sale before March. Given NV’s execution of late, a second spin is almost a given, so we are now well into Q2/2010.
So, we reiterate our earlier statements, GT300 is the wrong architecture, done for the wrong reasons, in the wrong way. At this point, we would be very surprised if Nvidia was at all competitive against Evergreen or Manhattan until 2H/2010, but even that is optimistic.S|A
Latest posts by Charlie Demerjian (see all)
- ARM upgrades realtime offerings to v8-R and adds Cortex-R52 - Sep 21, 2016
- Everspin and Globalfoundries team up for embedded ST-MRAM - Sep 15, 2016
- Intel’s Xpoint is pretty much broken - Sep 12, 2016
- ARM adds 2048-bit vectors to v8A with SVE - Sep 7, 2016
- AMD releases Bristol Ridge 7th Generation APU - Sep 5, 2016