Sujet : Re: Historical evolution of CPU perf
De : cr88192 (at) *nospam* gmail.com (BGB)
Groupes : comp.archDate : 09. Oct 2024, 19:23:22
Autres entêtes
Organisation : A noiseless patient Spider
Message-ID : <ve6hmr$2nbq0$1@dont-email.me>
References : 1
User-Agent : Mozilla Thunderbird
On 10/9/2024 11:33 AM, Stefan Monnier wrote:
I'm looking for a chart illustrating the evolution of CPU performance
(e.g. single-threaded or maybe performance per watt) over the years,
covering something like 1990-2020.
Any good candidates?
Yeah, I would also like something like this, or maybe some way to "sensibly" compare the relative performance of modern stuff with vintage stuff.
Like, for example, I can't make sense of whether the performance of my current project is similar to similarly-clocked vintage hardware, or potentially significantly faster.
Based simply on Dhrystone score, it would likely be placed in a similar area to a 90s era PowerPC in terms of perf/MHz.
But, if I add an early 2000s laptop as a reference point, stuff gets weird. In various benchmarks, the difference in performance is significantly smaller than the relative difference in clock-speed.
Though, the laptop is also break-even with a RasPi2 in terms of general perf (in theory, the laptop should be faster). Seems like the laptop suffers a relative deficit in terms of memory bandwidth (*).
But, can note that Dhrystone doesn't really measure memory bandwidth...
*: The 100MHz DDR1 RAM in the laptop gets roughly 7x the memory bandwidth of a 16-bit DDR2 chip being run at 50MHz. Sort of makes sense if one assumes 4x the width and 2x the clock-speed.
I am not sure if just the laptop, or if RAM access in general was proportionally slower in the 90s. Or, if it is just a case that late 90s / early 2000s, CPUs had gotten faster much faster than RAM had gotten faster, so there was a performance lag here.
I suspect it may be the latter, if one linearly extrapolated backwards, it would mean 486 PCs running at ~ 10-16 MB/sec for RAM bandwidth, which in my own testing seems insufficient to run Doom at acceptable speeds (actual 486's having no issues running Doom).
...
Stefan