NVIDIA IS TRYING to make Ion2, its next generation ‘chipset’ seem like something it is not, but the specs say otherwise. It is just a warmed over G218 integrated graphics chip with a few ports added on.
There are two parts that are due soon for Pinetrail, the GT218-ION-A3 and the GT218-ILV-A3. Production started in November according to some fab moles we ran into recently. For the code name averse, the GT218 is a DX10.1 GPU marketed as the G210 on the desktop and has since been renamed to the G310 to cynically snow consumers. Now it is called Ion2, presumably because the snow wasn’t deep enough.
The -ION- version will connect to Pinetrails aimed at desktops, the -ILV- is aimed at widgets and devices, but both support the Atom D510, D410, and N450. They have an astoundingly low number of cores, 16, running at 535MHz core and 1230MHz memory. The earlier desktop variants run at 589MHz and 1402MHz, so these parts will be about 10 percent slower for core bound functions than their already anemic desktop brethren.
For outs, the ION variant has HDMI, DVI, DP or VGA, pick two. ILV has LVDS and, pick one, HDMI, DVI, SL, DP or VGA. Claimed performance is P900+ in 3DMark Vantage, quite an interesting number considering the desktop version is clocked higher, has much more memory bandwidth, and scored lower. Is this the classic Nvidia honesty shining through again?
They both support DDR2 or DDR3, so the memory will be cheap enough. Nvidia claims that the GPUs have a TDP of 12W with a sub-5W idle power draw, excluding framebuffer memory, so add 4-5W for that. This is down from the 18W, Nvidia’s claim, of the MCP7A when coupled to an Atom, and that only scored P450 or so. Then again, that didn’t need a dedicated framebuffer, so the real world gain is only 1-2W. Call me overly skeptical, but it looks like Nvidia is claiming wins for Intel’s core advances. How unethical if true.
Back to the story, Nvidia is trying to sell a 12-17W part to the handheld and crotchtop market. While it may be down on the draw from MCP7A, it also does notably less, like not having the memory controller on board. While the numbers look good in the way that they are presented, given how much functionality Pinetrail pulled on-die, and given how much Ion2 does not have to do any more, that 12+W is pretty lame. Expect Nvidia to lose all the battery gains that Intel worked so hard to engineer in, and then some.
Pinetrail, Tigerpoint and G218 Ion2
Now comes the funny part. Nvidia is claiming that Ion2 is a chipset, not just a G218 with a few bits added to it. If you look at the part name, it is GT218-xxx-A3, not MCPxx like Nvidia GPUs. On top of that, wait for it, it is connected to the Intel Tigerpoint chipset, not to the CPU. To make matters more laughable, it is connected over a 1x PCIe2 link.
When was the last time a chipset was defined as a GPU connected over PCIe to a real chipset again? Do you sense that some company is taking liberties with the truth here? Me too. Furthermore, that PCIe2 1x link has ‘Nvidia bandwidth optimizations’, Nvidia’s term for overclocking the link. No, really, it is claiming that jacking up the PCIe2 link clock beyond its recommended specs is a technology. Then again, Nvidia still tries to claim driver blacklisting to be advanced ‘SLI technology’, and this ‘technology’ isn’t nearly that egregious. But whatever it wants to label this, overclocking will burn more power.
So, in the end, Ion2 is G218, also known as G210 or G310 on the desktop, nothing more. It uses a clock-jacked PCIe2 1x link and has some pretty curious power claims on top of that. I guess the surprises Nvidia keeps talking about will be seen on purchasers’ faces when they realize what happened to their battery life.S|A
Latest posts by Charlie Demerjian (see all)
- Gigabyte’s ARM servers represent a sea change in the market - Mar 30, 2015
- VESA shows off working USB-C Alt Mode with Displayport - Mar 24, 2015
- UzBrain’s Rail Gun turns a toy weapon into an FPS controller - Mar 18, 2015
- Fotonation uses computational imaging to focus faces - Mar 17, 2015
- HSA foundation releases v1.0 of their namesake spec - Mar 16, 2015