Re:AMD passes Intel in market cap
by NotZachari on Tuesday August 02, 2022 @08:50PM
(#32326329)
It's not that I think that shorter pipelines are a bad thing when implemented properly, but historically, AMD used that as a way to match or beat Intel on raw benchmark numbers. A shorter pipeline and pipeline flushing when ops hang or whatever means you get more operations per cycle on paper. If there's an issue, they'd just say "fuck it" and flush the pipeline and run it back, and because it's shorter, the pipeline is full and progressing fairly quick. That leads to higher benchmark scores for inferior hardware or stats that the average casual gamer or enthusiast sees. That leads to sales.
As I said, Intel's not perfect and I don't hate AMD. I want the best possible products. I just feel like market share isn't a great indicator of quality here. Threadripper and top tier AMD products are on the same level as Intel's products, but I personally tend to trust Intel's performance at the price point far more in real world situations. It was the worst kept secret in tech for years that AMD was gaming raw numbers at lower price points to sell units.
Gonna be real. This is all just my take on it, and I could be wrong or biased. I know they're not the same as they were 10 years ago and produce legitimate contenders at the top level. I just don't trust their low to mid-tier products.
Pentium 4 and Bulldozer both had (very) long pipelines and look how they turned out. Long pipelines certainly doesn't fundamentally equal better real-world performance, short pipelines != better benchmarks. It depends on the combination of architectural decisions considered together, right?
You're going to have to expand on this concept a bit to convince people that these design decisions are a deliberate attempt to cheat the benchmarks, throwing out lingo doesn't quite suffice.