We'll this is amazing I didn't think we would move from 14 nm any time soon seeing how long we where stuck on 28nm

Let's hope moving to 7nm, instead of refining 14nm, doesn't bite them in the a*s like HBM2 did.

We'll this is amazing I didn't think we would move from 14 nm any time soon seeing how long we where stuck on 28nm

It kind of makes sense, for the following reasons: * I've heard from some sources that 7nm is bordering the limits of silicon transistors. Intel has already started working on 14nm and I think 10nm. Perhaps GF and TSMC are like "do we want to keep up or do we want to win?" so they decide to spend the extra funds for 7. * In terms of GPUs, AMD needs whatever it will take to shrink their die size, since it doesn't seem they have any immediate plans to replace GCN. The downside is that doesn't prevent Nvidia from getting the same advantage. * It is theorized that the manufacturing process of Ryzen hasn't been perfected, which is supposed to explain the underwhelming overclockability. If it's actually a problem with the architecture itself, AMD could use the die shrink as an excuse until they figure out a way to improve it. * I'm sure it costs less for AMD to adjust for a die shrink (and maybe some tweaks here and there) than it does to make revisions for all of Ryzen's weaknesses. A die shrink will offer a performance and wattage improvement but ultimately the cost comes down on companies like GF rather than AMD. In other words, AMD gets a better product without working as hard and by paying less. I think they're just trying to boost revenue as much as possible. As long as revenue goes up, investors will keep coming and then they can start cracking down on improving products "the hard way".

Let's hope moving to 7nm, instead of refining 14nm, doesn't bite them in the a*s like HBM2 did.

Using HBM2 hasnt bitten them. They have 16GB of more advanced and generally better memory with proper ECC on the GPU competeing with the $5,000 P40 and P6000s.

Well this is CPU's, not GPU's. Intel has been on 14nm for a few years and 22nm before that. Besides, Intel's process is more complex/advanced from what I read. 7nm FinFET from GLOFO etc is like 10nm on Intel 🙂 I highly doubt Ry/zen "2" will be 7nm, the 3rd might (2019ish). 2nd will probably be 14nm+/++ aka refined (but this is not always a bad thing.. result is lower power / higher clocks)

14nm finfet was more like Intel's higher node as for gate sizes, however, that is not the case with 7nm which is based on IBM process. It should be actually close to what Intel should use but much earlier. Don't expect 14nm zen 2 if it's being developed for 7nm, it would take a lot of time and costs to redesign it as this in not just die shrink.

This 7mm process is built for high frequencies too. Glofo buying IBM's division was one of the best decision they have ever made. Speaking of they do plan to have multiple tape outs on 7nm in 2018, not sure if Zen is one of those but would be lovely to have a 7nm Ryzen by 2019. Those griping about HBM2, just wait for Navi to come out and all answers as to why they are building on it now will be answered. Building using HBM2 today isn't really needed but when you start gluing more than one GPU together via Infinity fabric that bandwidth will be much needed.

interesting read, boss. i had missed this somehow. good post

* It is theorized that the manufacturing process of Ryzen hasn't been perfected, which is supposed to explain the underwhelming overclockability.

you can see where the problem is right here - its due to the immature process. power spikes suddenly outside of the normal quadratic scaling beyond 1.3v-ish. the stability issues beyond ~4ghz are caused by the increased power draw...the dies thermal resistance limit hits a brick wall. new steppings might smooth it out, or we may have to wait until zen2 for the wattage/amperage to calm the hell down with voltage bumps. either way its fixable. just depends on if they want to spend the time & effort to fix it, or double down on zen2 instead. http://i.imgur.com/QFsSBqj.png

" Building using HBM2 today isn't really needed but when you start gluing more than one GPU together via Infinity fabric that bandwidth will be much needed." Zactly. I think ~thats the reason amd went out on a bit of a limb w/ early hbm adoption - compact for use very adjacent to mcmS suits fabric also.