CPU question

[mCode]

Hi, 

 

Why isn't a higher base / boost clock necessarily making it a better CPU? I noticed people recommending AMD over Intel despite its lower boost clocks.

 

Please educate me. A trustworthy source is fine too. Everything is gibberish at this moment for me, so I hope to learn. 

[minibois]

It's not possible to compare two different brands of CPU's or even two different generations of CPU's together just by their clockspeed, because there is something else that influences the 'actual speed' of a CPU: the IPC, which stands for "instructions per clock".

 

If GHz is the (billion) amounts of ticks per second, IPC is the amount of things it can do per tick.

IPC isn't a definite certain number, but rather something that can be deduced through some advanced testing and comparing. This IPC will differ between programs though, as CPU's have certain instruction sets which a program can or cannot make use of.

 

For example, let's compare the 5600X (3.7/4.6GHz base/boost clock respectively and about 4.3-4.5GHz all-core boost (as the 4.6GHz is only a single/dual core boost)) and the 10600K (4.1GHz/4.8GHz base/boost and 4.5GHz all-core).

Those are all-together very similar GHz numbers, with the same amount of cores, but if you look in certain reviews, you'll see the 5600X pull ahead of the 10600K: https://www.guru3d.com/articles_pages/amd_ryzen_5_5600x_review,8.html (this page even features the IPC, for this specific benchmark).

Same goes for Cinebench R20, the newer Cinebench benchmark: https://www.guru3d.com/articles_pages/amd_ryzen_5_5600x_review,9.html

 

But there will also be scenarios where the two CPU's are equal: https://www.guru3d.com/articles_pages/amd_ryzen_5_5600x_review,24.html

 

So in conclusion, don't just look at the base/boost/all-core speeds, but rather what these CPU's mean in the context of what you use the PC for.

IPC (the amount of things it can do per Hz) influences the 'actual speed'.

[tikker]

Depends on the workload. Parallellisable workloads can benefit more from having more cores available than from a single speedy core. Multiple fast cores will of course be even better than multiple slow cores.

 

The actual "speed" is a combination between clock speed and the instructions per clock (IPC) it executes. This is why you can't compare a 5 GHz Intel CPU to a 5 GHz AMD CPU. For for illustration a 5 GHz clockspeed CPU with an IPC of one can process as many instructions per second as a 2.5 GHz clockspeed  CPU with an IPC of 2 as the latter executes twice more per cycle. If everything else is the same, they will then theoretically perform the same in this simple example.

 

AMD offers more cores for the same or less money than Intel and their speeds aren't massively apart anymore either, so they are an attractive option in multiple ways. I believe Intel often still has the edge in single-core performance though, but we're not talking factors anymore. NVM @miniboisexamples above show AMD is pulling ahead as well

[Chris Pratt]

1 hour ago, tikker said:

AMD offers more cores for the same or less money than Intel and their speeds aren't massively apart anymore either, so they are an attractive option in multiple ways. I believe Intel often still has the edge in single-core performance though, but we're not talking factors anymore. NVM @miniboisexamples above show AMD is pulling ahead as well

Yes. 10th Intel had higher single core perf than Zen 2, but Zen 3 had a massive gen on gen increase of 20%. It is now the single core perf leader, though Intel made up ground with 11th gen.

[Chris Pratt]

In addition to IPC, prefetch is extremely important to CPU performance. The CPU will attempt to fetch instructions preemptively from RAM and store them in the much faster cache in order to process those instructions more quickly. Like any form of cache, though, there's cache hits and misses. The more hits you get from cache, the faster instructions are processed, and hence the better performance is. AMD has been adding relatively massive amounts of cache to it's CCDs for a while now, and it's also working on technologies like 3D V-cache that will potentially double the amount of CPU cache. The more cache, the more chance for hits vs misses. Intel, by comparison, has relatively paltry amounts of cache, which is a big contributing factor to AMD pulling ahead.

[tikker]

38 minutes ago, Chris Pratt said:

Yes. 10th Intel had higher single core perf than Zen 2, but Zen 3 had a massive gen on gen increase of 20%. It is now the single core perf leader, though Intel made up ground with 11th gen.

Ah I see. I haven't kept up with Intel's 11th gen or Zen 3 very much.