Personally I still rate past experience as a good indicator. We all know what happened last time so lets look at the sort of stuff written at about the same time about Bulldozer:
http://www.portvapes.co.uk/?id=Latest-exam-1Z0-876-Dumps&exid=threads/rumour-bulldozer-50-faster-than-core-i7-and-phenom-ii.2134911/
Amusingly we have a number of the same posters saying exactly the same sorts of things with the same fervour.
Anyway then it was 50% faster at this point pre-release, which ties in nicely with the latest "AMD is saying above 40% IPC." this time around.
It would be great if we could trust what AMD are saying, but lets be honest they have earned a reputation for talking out of their bottoms, so I'd reign in my expectations a bit and wait for the independent reviews before getting too carried away.
Actually if you read that thread you'll notice that a lot of the discussion revolved over multi threaded benchmarks and people comparing CMT vs SMT. IPC was not discussed much because AMD hadn't said anything about IPC pre-bulldozer. People assumed 10-20% IPC gains. Yes they were wrong, but it was hard to guess Bulldozer was going to have worse IPC than Phenom II. Personally I was shocked about that on the launch day. But JF posts on HardOCP had made me suspicious even before the launch due to something he said but I can't remember what exactly now, I just remember being concerned about single threaded performance. AMD was really tight lipped about IPC.
Where Zen launch is different from Bulldozer launch:
- No one really expects Zen to beat Intel outright in single thread performance, but most everyone is optimistic it will come close, and perhaps even overtake Intel in certain disciplines. Personally I have maintained cautious optimism about it, but Haswell level ST perf. would be satisfactory to me.
- We know more about Zen today than we did about Bulldozer before launch. We know base clocks, we've seen power consumption, we've heard the HotChips architecture overview..
- We aren't going by marketing slides. AMD hasn't showed us a lot but they did show us 2-3 mainstream open source CPU intensive applications, and I am pretty sure everyone here thinks the results look promising.
- We are also seeing comparisons between apples and apples. The thing with Bulldozer this whole gimmick that you got "honest to goodness" real cores for free. You get 8 cores for the price of four. This is not the case with Ryzen. They are pitting 8 AMD cores vs 8 Intel cores. Apples to Apples, nothing is getting lost in translation.
There are still unknowns, like what will the dreaded single threaded perf. look like? If we assume AMD SMT scaling is similar or worse than Intel's it could be pretty darn good. If SMT somehow shares more resources than on the Intel side it could be worse. Or if branch predictor is bad. By how much we don't know, but as I said it looks promising.
Problem with Bulldozer was the following:
- Single thread perf. straight up sucked, couldn't even match Thuban, which was a regression, first time ever for AMD. (actually not different from Pentium III vs Pentium 4).
- The chips was a monster. 1.2B transistors (at first we thought it was 2B, but 1.2B was still more than double of that of Sandy Bridge 504M). Inefficient, expensive for not much benefit.
- It had poor perf/watt. Bulldozer basically had no redeeming quality. At least we know Ryzen can render or transcode stuff by being more power efficient.
So far we're pretty confident Ryzen isn't a Bulldozer. It could still completely suck at single thread performance, but I think they are being genuine about the 40% or greater IPC uplift (Lisa Su seems alright), and we know clocks aren't horrible either (3.4Ghz base is pretty solid for an 8 core part, while maintaining the power consumption lead in those tests).
I give Ryzen 80/20 chance of success at this point.