Intel & AMD Processor Hierarchy

Page 3 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.

Awev

Reputable
Jun 4, 2020
89
19
4,535
The king is dead (Intel), long live the king (AMD). A new era has started - the post-Zen 3 era :p Can't wait to see what AM5, and PCIe gen 5 and 6 does for Zen 4 & 5 CPUs. Intel, please try to get back in the (1080p and 1440p) game.
 

Kamen Rider Blade

Distinguished
Dec 2, 2013
1,419
944
20,060

Awev

Reputable
Jun 4, 2020
89
19
4,535
@Kamen Rider Blade Just a few thoughts on the matter. First, the new Zen 3 chips are pulling 20 to 60 watts less than the 10900K (105 for the 5800X, 5900X, and 5950X, 65 watts for the 5600X, and 125 watts for the 10900K), granted that is at rest, yet they max out around 142 watts due to the limitations of the AM4 socket, and hence, motherboards. Watts is not a true indication of the heat it will generate, as other things come into play, such as die size, and in the case of the Zen family, how far apart the different chiplets are spaced, yet as a general rule of thumb higher watts means higher heat.

Would a Threadripper 3960 be a good choice for you? The Threadripper series is designed for the High End DeskTop (HEDT) market. Or how about their Epyc server chips - designed for what you are talking about - 24 hours a day x 7 days a week x 365 days a year, air cooled, mounted in a rack. In the phone industry they used to speak about the five nines (99999) of uptime, meaning that their equipment was so reliable that it would only be down for less than 6 minutes a year (60 minutes * 24 hours per day * 365 days per year = 525600 minutes per year, and then 99.999% of the time the equipment would be running, so it would be down for only 5 minutes and 16 seconds (I rounded up) or less). I have not seen the results of the uptime for an Epyc server chip, yet at 280 watts, and being a server that is air cooled, the cooler and fans have to be doing well for the five 9s of uptime.

Companies used to perform 72 hour burn-in tests, and if anything was going to fail it would be then (or five years down the road after the warranty ran out). So many people want to be able to order a custom PC, have it built and shipped the next day, that you don't hear about the burn-in tests in advertising like you used to.

A little research and you can find a good air cooler and thermal paste. Just make sure that the case you place the chip in has good ventilation, that can affect airflow more than the fan on the cooling fins of the cooler.
 

Kamen Rider Blade

Distinguished
Dec 2, 2013
1,419
944
20,060
Would a Threadripper 3960 be a good choice for you? The Threadripper series is designed for the High End DeskTop (HEDT) market. Or how about their Epyc server chips - designed for what you are talking about - 24 hours a day x 7 days a week x 365 days a year, air cooled, mounted in a rack. In the phone industry they used to speak about the five nines (99999) of uptime, meaning that their equipment was so reliable that it would only be down for less than 6 minutes a year (60 minutes * 24 hours per day * 365 days per year = 525600 minutes per year, and then 99.999% of the time the equipment would be running, so it would be down for only 5 minutes and 16 seconds (I rounded up) or less). I have not seen the results of the uptime for an Epyc server chip, yet at 280 watts, and being a server that is air cooled, the cooler and fans have to be doing well for the five 9s of uptime.

Companies used to perform 72 hour burn-in tests, and if anything was going to fail it would be then (or five years down the road after the warranty ran out). So many people want to be able to order a custom PC, have it built and shipped the next day, that you don't hear about the burn-in tests in advertising like you used to.

A little research and you can find a good air cooler and thermal paste. Just make sure that the case you place the chip in has good ventilation, that can affect airflow more than the fan on the cooling fins of the cooler.
Nah, I don't think the ThreadRipper line fits what I want, I'm perfectly happy with the normal mainstream line. I've run mainstream AMD CPU's in 24/7/365 configurations for years on Air in a Full Tower for beyond a Decade before the CPU started having issues. That's a testament to the build quality of AMD's main stream CPU that they can be worked that hard before dying out.

I only buy Noctua for HeatPipe and Fans.

As far as Thermal Paste, I prefer Artic Cooling MX-4 for it's 8 year TIM longevity.
 

Awev

Reputable
Jun 4, 2020
89
19
4,535
I see the 5600X has made it into the top five position in most listings. If I am counting correctly, Post Zen 3 has AMD with the first four of the top five, and seven of the top 10 places, in most of the Post Zen 3 (PZ3?) listings. Nicely done AMD.
 

dbob3

Reputable
Mar 9, 2020
5
1
4,515
This metric largely depends upon a mixture of instruction per cycle (IPC) throughput (the number of operations the chip can execute in one clock cycle) and frequency, which is the speed at which the transistors switch between on and off states

Just nit-picking. The frequency of the chip is the frequency at which the results are saved in their flops/latches. The transistors are actually orders of magnitude faster than the clock frequency, but between each 'save point' (latch/flip-flop) the transistors are doing maths and such. So in one clock cycle (which is what the frequency is controlling), the data may need to progress through 10-20 transistors (just making up a number, I don't work on the backend, just frontend). If there is too much logic (too many levels of transistors) between the latches/flip-flops, the data can't progress fast enough before the time the clock toggles. In that case, the data doesn't make it in time and the flop/latch gets the wrong data. So you get errors. You were trying to clock the processor too fast for that particular data path. So engineers will find out which data paths are the worst (slowest) and find ways to make them faster. Maybe they can use larger transistors that use more energy, but propagate faster. Or maybe they add a latch/flip-flop in between, and instead of being able to calculate a floating-point operation in one clock cycle, it takes two.
The Pentium 4, in order to get super fast clock speeds, ended up splitting part of their logic up into a lot of stages (20 for one of their pipelines). They ended up cancelling the Pentium 4 successor because doing this was causing way too much heat for the performance they got.
https://en.wikipedia.org/wiki/Instruction_pipelining#Number_of_steps

Anyway, long story short, the frequency is not the speed at which the transistors switch between on and off, it's how often the flip-flops/latches are saving new values (hmm, yeah, doesn't sound very cool the way I said it).

I loved the article, btw. Makes you wonder where Intel's been putting its money. The MBAs over there have been doing MBA-type work and not keeping the experts. I remember 6 years ago they were handing out severance packages to anyone who would leave (they wanted to downsize by 5-10%). Imagine you have a team with experts, noobs, and people in-between, all living in the bay area where there are lots of other places to work. And then you told everyone that they could take a big money package and leave. How much more likely is an expert (confident in their abilities) to take the $$$$ (the severance package is also larger if they are more experienced) and leave, versus someone newer and less secure about their abilities to get a new job? At the same time period, they were hiring lots of new recent college graduates (cheaper pay when they're right out of school). You're now doing the same amount of work with roughly the same number of people, but paying them a lot less. Profit! ...and you've lost a lot of experts. Around that time they laid off a lot of their fab R&D team who were working on the next processes. More ways to save money! ...and now they are still struggling at the next process nodes, strange...
 
Last edited:
  • Like
Reactions: Makste

kato128

Distinguished
Feb 23, 2009
160
5
18,685
So no 3700x in the post zen3? Yet all the way back to a 1600x is in there. Love how review sites seem to think no one bought them.
 
Dec 24, 2020
1
1
15
Description says "All of today's consumer, desktop CPUs compared " but it skips a lot of processors even current gen desktop stuff, like the i5 10400. Makes it hard to compare mid range parts.
 
  • Like
Reactions: Soaptrail

clutchc

Titan
Ambassador
Is anyone ever going to fix the "Intel Core i7-7700 " typo? It still appears twice; once in its correct location on the Intel side and once in the wrong location way down on the AMD side of the Legacy Desktop Processor Hierarchy chart.
 

dayzero

Prominent
Jan 3, 2021
3
0
510
What does "1440p Gaming Score" actually mean? Is it a relative FPS score? How were the figures in this column produced?
 

UWguy

Commendable
Jan 13, 2021
68
54
1,610
What you also need is a column with the processors ranked taking into account the average overclockablity. Now you are talking and this should change the rankings significantly. How many people are really going to game using out of the box specs?
 

King_V

Illustrious
Ambassador
What you also need is a column with the processors ranked taking into account the average overclockablity. Now you are talking and this should change the rankings significantly. How many people are really going to game using out of the box specs?
How many people actually overclock?

Overclockability is unpredictable - it's luck of the draw. It has no business in the charts.
 

Awev

Reputable
Jun 4, 2020
89
19
4,535
What you also need is a column with the processors ranked taking into account the average overclockablity. Now you are talking and this should change the rankings significantly. How many people are really going to game using out of the box specs?
Good idea, this way Until can be at the bottom again, as all AMD CPUs are unlocked, while only a few of the Until CPUs let you overclock them. That is what you mean, right?

And lets ignore that even Tom's Hardware suggests that instead of overclocking you step up to what you actually need - unless you are at the very top then consider overclocking, for example step up from an iCore 5 to an iCore 7, or go from i9-10850? to i910900k. I have quoted and linked to Tom's Hardware comments and video on this subject in prior posts.
 
If you have a local Microcenter, then the 10850k is by far the best overall value CPU at $349.
Except they also have the 8-core, 16-thread 10700K for $280. Paying 25% more for 25% more cores might not be bad if you have some application-specific use for them, but when you consider that today's games and the vast majority of applications don't even fully utilize an 8-core, 16-thread processor, and currently tend to perform pretty much the same on a 6-core, 12-thread one (like the 10600K they have for just $190), that could still easily be an extra $70-$160 spent on additional cores that might not have much of an effect on performance for years to come.

And the usual caveats for Microcenter apply, namely that they only have 25 locations in the US and most people live hours away from one, so such a recommendation would not be applicable to many.

And of course, if the upcoming Rocket Lake processors manage to perform notably better on a per-core basis, one might be better served by trading those extra core for additional per-core performance.
 

JRHERITA

Distinguished
Dec 31, 2007
9
2
18,515
How is the 10700K @ 5.1 GHz so much faster (18%?) than the 9900KS? They're the same chip, and the only difference is 5.1 GHz all core vs. 5.0 GHz all core?
 
Mar 4, 2021
2
0
10
intel prices are low nowadays compared to ryzen. like 10850k is almost $150 cheaper than 5600x. I'm still using xeon 1245vs (i7-3770 class). been checking out CPU at online shopping sites for weeks, but my current cpu still always feels fast. makes me wonder, if there's anything I can spend on my current desktop without having to "upgrade" ? maybe an NVME drive ? but my motherboard doesn't support nvme boot