Remember when SemiAccurate said Nvidia shuffled their Tegra lineup to insert a stopgap A9 part? Remember when SemiAccurate said it would delay the real T40 based Tegra 4? Remember when SemiAccurate said Nvidia would put out a reference phone? Guess what happened?
While we hate to dredge up old news, it seems that some people didn’t actually believe us when we laid out the tale of woe that was T30, T33, T35, and T40, the attendant delays, and the dysfunctional management moves that made it so. Well, today Nvidia announced their Tegra 4i SoC and a reference phone design called Grey to go with it. It all sounds so eerily familiar, almost like SemiAccurate nailed every point about it, and the semi-official Nvidia backchannel denials were as honest as their official denials.
OK, enough patting ourselves on the back, lets take a look at what is going on here. If you read the write-ups of the Grey/Tegra 4i announcements like this one at Ars, there are a few things that become clear. First is Nvidia is up to their old ways when quoting speeds, the chip is called a 2.3GHz part even though that is just the max turbo frequency of a single core with the others cores off. Sustained frequency or that of more than a single core is far lower and undisclosed.
The process Grey is made on was also not disclosed, but it almost assuredly a 28nm chip. A 33% top clock rate bump without raising TDPs to an unacceptable level is pretty much impossible without a shrink. Also, the GPUs are said to be the same as the T40 version of Tegra 4, ie the A15 cored new one. That is a 28nm part and is again extremely unlikely that those units would be backported to an older 40nm process just for this chip. Last is the anecdotal evidence of the relative die sizes Nvidia was showing on the launch slides. While the die shots themselves are almost assuredly faked, the scaled die is a familiar PR move by the company.
Related to the GPU side of things is what SemiAccurate said in October of 2011 when we exclusively told you about the T35/Tegra 4i/Grey project. We said that it was a stopgap hack that would put the A9 cores back in to the T40, pushing out that chip and ending up with a Frankenstein hack. Today we got exactly that, A9 cores in a T40 uncore, or at least most of the uncore, some of the shaders fell by the wayside to make room for the Icera modem.
Overall, the chip is a mixed bag of good and bad. At this point the cores are painfully outdated, the A15 was last years news, the A9 in the T4i made headlines in 2009. For CPU power this part is DOA. On the GPU side though, depending on how well the new shaders do, it may not embarrass itself there, but the Tegra 4i won’t win many graphics crowns due to it’s lack of programmable shaders. If the memory bandwidth from the uncore is a big step up from that in the Tegra 3/T30/T33, it will do just fine. If not, Tegra 4/T40 is in deep trouble. The modem, if the bugs that crippled the Asus Transformers late last year were fixed, will be quite solid. SemiAccurate thinks it will be just fine by the release date.
Technical items aside, there is one big problem with the unveiling of Tegra 4i/T35/Grey and more importantly of the Phoenix phone reference design. As we said two days ago, it points to some very fundamental problems that the entire Tegra line has, and the grim outlook surrounding it’s future. Worse yet is the name, Tegra 4i is a big flashing red flag for anyone in the financial world, pay very close attention to why Nvidia chose this naming convention.
Note: The following is analysis for professional level subscribers only.
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)
- You can now buy bare Snapdragon SoCs with 410E and 600E - Sep 28, 2016
- Spin Transfer Technologies talks about their ST-MRAMs - Sep 27, 2016
- ARM adds CMN-600 interconnect and DMC-620 memory controllers - Sep 27, 2016
- Globalfoundries 7nm process isn’t even close to the name - Sep 26, 2016
- ARM upgrades realtime offerings to v8-R and adds Cortex-R52 - Sep 21, 2016