News AMD vs Intel 2020: Who Makes the Best CPUs?

[spongiemaster]

1. How many of those can be overclocked by the user, both in absolute numbers and as a percentage?

Theoretically? 100%, you can change the multiplier on their CPU's
In the real world? Close to 0%, because you'll typically end up with worse performance which defeats the purpose of OC'ing

5000 series aren't available, and remember that 3000 series at launch couldn't even hit their advertised boost rates. AMD clocks their CPU's to their absolute limits leaving no headroom for any useful overclocking. As M42 mentioned, if you can't get the all core overclock at least to the maximum single core boost rate, there's no real reason to OC your CPU as you're often going to end up with worse performance than leaving it stock and letting AMD's Precision Boost control the clocks. That's why you will find multiple major review sites recommending to not OC AMD CPU's.

 

[AndrewJacksonZA]

It appears that there is a communication issue here. I'm going to try something different, let's break down Paul's conclusion on the overclocking section:

Winner: Intel. When it comes to AMD vs Intel CPU overclocking, Team Blue has far more headroom and much higher attainable frequencies. Just be prepared to pay for the privilege. AMD's approach is friendlier to entry-level users, rewarding them with hassle-free overclocking based on their system's capabilities, but you don't gain as much performance. Also, you need to be aware that Intel's overclocking prowess can be a bit of a hollow win - At stock settings, AMD's Ryzen 5000 processors are often faster than Intel's chips, even after aggressive overclocking.

Team Blue has far more headroom and much higher attainable frequencies.

I agree.

------------------------------------------------------------------------------------------

Just be prepared to pay for the privilege.

I have no idea what the CPU sales counts are. The fact that Intel chooses to lock some CPUs is probably a marketing decision, not a technical decision.

Whether it's a technical reason or a marketing reason, this is still a negative. A big one. This is what I am trying to communicate. I am not saying "How many CPUs have Intel sold?" I am saying "How many CPU models have Intel made available for sale, and out of those, let's say 20 models, how many can be overclocked by the user?" Maybe three out of the 20, and they're the costliest three models - and only on the costliest of the motherboard series.

AMD's approach is friendlier to entry-level users, rewarding them with hassle-free overclocking based on their system's capabilities,

With AMD, that option is available on all of their CPUs and all of their motherboards except for the base series.

This is the main point that I am trying to make. If one company locks down all but their highest end and costliest models and need the highest end and costliest motherboards and the other company makes it available on all their CPUs and the vast majority of their motherboards, for whatever reason, then the company that allows it should be awarded the winner. This also goes in the spirit of an enthusiast from the beginning, trying to extract more performance from lesser products.
------------------------------------------------------------------------------------------

but you don't gain as much performance.

I agree.

------------------------------------------------------------------------------------------

Also, you need to be aware that Intel's overclocking prowess can be a bit of a hollow win - At stock settings, AMD's Ryzen 5000 processors are often faster than Intel's chips, even after aggressive overclocking.

I agree. Whether this point should be included in the decision about overclocking is up to Paul.

------------------------------------------------------------------------------------------
------------------------------------------------------------------------------------------

Now as to why Intel is leaving so much performance on the table, like others have also asked, that another story.

 

[M42]

Now as to why Intel is leaving so much performance on the table, like others have also asked, that another story.

Agreed! I think it's because Intel didn't need to do anything until AMD became a threat. I think that may have woken up Intel.

Possible evidence of this is on Tom's website now:

https://www.tomshardware.com/news/intel-core-i7-11700K-geekbench-5-leak

Competition between AMD and Intel is good for us consumers!

 

[cryoburner]

I think you totally missed my point! How can clock rates not matter when the topic is overclocking prowess? The fact is AMD CPUs seem to be inferior overclockers compared to Intel's.

As for performance per clock cycle, if one believes the benchmarking rumors, it looks like intel is working on improving that now for the 11000-series CPUs. For the sake of competition, let's hope they do!

My point is that overclocking CPUs in general makes less sense these days. As AMD has shown, chips can be binned to get the most out of the silicon without the user having to waste time going through a potentially involved process just to achieve something similar. There's not as much point in overclocking their current processors, because they are already optimized to perform near their maximum potential out of the box. That method also ensures optimal efficiency, so the CPUs don't need to be running cores at higher clocks than is necessary when not needed, and can boost to higher clocks for lightly-threaded tasks than would be possible across all cores. That should be seen as a positive thing compared to shipping processors in a less-optimized state and leaving it up to the end-user to get the most out of them, often at the expense of efficiency.

And again, the actual numbers for clock rates don't matter much. Who cares if one CPU architecture can hit higher clocks if it performs less computation per clock? AMD had a Piledriver processor that boosted to 5GHz with a 4.7GHz base clock over 7 years ago. Much of the FX lineup in general could be considered "good overclockers" in terms of performance left on the table, but that wasn't exactly a great thing, since the performance-per-clock was significantly lower than the competition, and those higher clocks required a lot more power and heat output to achieve. As the article puts it, that makes for "a bit of a hollow win".

Don't get me wrong, I think Intel's 10-series processors are fine enough, especially since the pricing and availability of AMD's competing processors currently leaves a lot to be desired, at least for the time being. Making SMT available across the lineup certainly helped, and makes this generation a notably better value than their 9th-gen counterparts were. But the relatively high power draw and heat output is a bit of a turn-off, and it's hard to consider a little bit of additional overclocking headroom on a few processors to be much of a positive, especially since they charge extra for CPUs that don't artificially lock users out from overclocking.

Even Intel's current unlocked processors are not great overclockers, since they too have been pushing boost clocks closer to the limits of the chips. In general, the performance gains from overclocking these chips are not going to be noticeable. It's not like some processors of the past, where overclocking could get you 50% higher clocks. With recent unlocked i9s and i7s, we're talking about maybe 5-10% higher clocks over stock for heavily multithreaded workloads, and almost no gains for lightly-threaded ones.

 

[M42]

My point is that overclocking CPUs in general makes less sense these days. As AMD has shown, chips can be binned to get the most out of the silicon without the user having to waste time going through a potentially involved process just to achieve something similar. There's not as much point in overclocking their current processors, because they are already optimized to perform near their maximum potential out of the box. That method also ensures optimal efficiency, so the CPUs don't need to be running cores at higher clocks than is necessary when not needed, and can boost to higher clocks for lightly-threaded tasks than would be possible across all cores. That should be seen as a positive thing compared to shipping processors in a less-optimized state and leaving it up to the end-user to get the most out of them, often at the expense of efficiency.

And again, the actual numbers for clock rates don't matter much. Who cares if one CPU architecture can hit higher clocks if it performs less computation per clock? AMD had a Piledriver processor that boosted to 5GHz with a 4.7GHz base clock over 7 years ago. Much of the FX lineup in general could be considered "good overclockers" in terms of performance left on the table, but that wasn't exactly a great thing, since the performance-per-clock was significantly lower than the competition, and those higher clocks required a lot more power and heat output to achieve. As the article puts it, that makes for "a bit of a hollow win".

Don't get me wrong, I think Intel's 10-series processors are fine enough, especially since the pricing and availability of AMD's competing processors currently leaves a lot to be desired, at least for the time being. Making SMT available across the lineup certainly helped, and makes this generation a notably better value than their 9th-gen counterparts were. But the relatively high power draw and heat output is a bit of a turn-off, and it's hard to consider a little bit of additional overclocking headroom on a few processors to be much of a positive, especially since they charge extra for CPUs that don't artificially lock users out from overclocking.

Even Intel's current unlocked processors are not great overclockers, since they too have been pushing boost clocks closer to the limits of the chips. In general, the performance gains from overclocking these chips are not going to be noticeable. It's not like some processors of the past, where overclocking could get you 50% higher clocks. With recent unlocked i9s and i7s, we're talking about maybe 5-10% higher clocks over stock for heavily multithreaded workloads, and almost no gains for lightly-threaded ones.

I agree with much of what you have written above. However, whether it makes sense or not, many people overclock their CPUs. Motherboard makers have made it very easy to do so. I have overclocked almost every system I have built in the last 7 years, and the increased performance is measurable (I usually use Passmark). Many web sites, including Tom's, will include overclocked variants of CPUs in game and system benchmark charts. Intel still has the lead over AMD in terms of overclocking headroom.

BTW, a fast CPU alone does not guarantee an overall fast system. You have to balance CPU, memory, disk, and graphics. Here are Passmark's top fastest 20 desktop systems, and so far only a couple are AMD-powered:

https://www.pcbenchmarks.net/fastest-desktop.html

 

[CerianK]

BTW, a fast CPU alone does not guarantee an overall fast system. You have to balance CPU, memory, disk, and graphics. Here are Passmark's top fastest 20 desktop systems, and so far only a couple are AMD-powered:
https://www.pcbenchmarks.net/fastest-desktop.html

A few weeks ago, the top system was an AMD TR 3990X, but it looked to me like the 2D performance was being gamed, so Richard looked into it for me and decided to remove the benchmark submission.

It remains an open question whether others are gaming the 2D benchmark (since its weight can 'make' a top system):
https://forums.passmark.com/pc-hardware-and-benchmarks/49425-2d-graphics-mark-scaling

 

[M42]

A few weeks ago, the top system was an AMD TR 3990X, but it looked to me like the 2D performance was being gamed, so Richard looked into it for me and decided to remove the benchmark submission.

It remains an open question whether others are gaming the 2D benchmark (since its weight can 'make' a top system):
https://forums.passmark.com/pc-hardware-and-benchmarks/49425-2d-graphics-mark-scaling

I believe other systems have been removed for reasons related to "gaming" the results. If the Passmark developers believed any of the Top 20 systems were also gaming the 2D results, they should remove those entries as well. Unless they do, they must be considered valid. 2D performance should be an extremely important factor, as almost everything (except 3D applications) use 2D rendering.

 

[cryoburner]

I believe other systems have been removed for reasons related to "gaming" the results. If the Passmark developers believed any of the Top 20 systems were also gaming the 2D results, they should remove those entries as well. Unless they do, they must be considered valid. 2D performance should be an extremely important factor, as almost everything (except 3D applications) use 2D rendering.

The "combined" scores for benchmarks like Passmark (along with Userbenchmark and others) are kind of a joke anyway, since not everyone is using their system to run the same kinds of software, and the weighting between the various tests (and choice of precisely what to test) is ultimately completely arbitrary. So the results really don't mean much. They are there just to drive traffic to their site, and little more.

The current top results being for a system with an i9-9960X and 1080 Ti should make that clear. There are CPUs that perform notably faster at most light to moderately-threaded tasks that would provide superior performance for anyone not primarily using their system for certain heavily-multithreaded workloads. And if one is mostly using a system for those kinds of multithreaded tasks, then other processors can be significantly faster. And there is certainly much faster graphics hardware available at this point. It's kind of silly to consider such a system to be the "fastest" in their list, but it might very well manage to achieve the fastest average of results at the specific workloads they arbitrarily decided to test for.

And that's not even so much a "Top 20 list", as many of the same systems each make an appearance multiple times. That same 9960X system appears 6 times out of those 20 results, submitted on various occasions over the course of a little over half a year. Another single 10900K system takes 4 positions. Others take multiple positions as well. These are people who are repeatedly running the test to get higher scores to outdo one another at an arbitrary benchmark, not so much an accurate representation of what the fastest systems are. It's worth pointing out that all 5 of the Ryzen systems currently in that list appear to be unique though, so actually, the list is almost even split between Intel and AMD when we ignore the duplicates and compare the 10 or 11 unique systems that remain.

 

[M42]

The "combined" scores for benchmarks like Passmark (along with Userbenchmark and others) are kind of a joke anyway, since not everyone is using their system to run the same kinds of software, and the weighting between the various tests (and choice of precisely what to test) is ultimately completely arbitrary. So the results really don't mean much. They are there just to drive traffic to their site, and little more.

The current top results being for a system with an i9-9960X and 1080 Ti should make that clear. There are CPUs that perform notably faster at most light to moderately-threaded tasks that would provide superior performance for anyone not primarily using their system for certain heavily-multithreaded workloads. And if one is mostly using a system for those kinds of multithreaded tasks, then other processors can be significantly faster. And there is certainly much faster graphics hardware available at this point. It's kind of silly to consider such a system to be the "fastest" in their list, but it might very well manage to achieve the fastest average of results at the specific workloads they arbitrarily decided to test for.

And that's not even so much a "Top 20 list", as many of the same systems each make an appearance multiple times. That same 9960X system appears 6 times out of those 20 results, submitted on various occasions over the course of a little over half a year. Another single 10900K system takes 4 positions. Others take multiple positions as well. These are people who are repeatedly running the test to get higher scores to outdo one another at an arbitrary benchmark, not so much an accurate representation of what the fastest systems are. It's worth pointing out that all 5 of the Ryzen systems currently in that list appear to be unique though, so actually, the list is almost even split between Intel and AMD when we ignore the duplicates and compare the 10 or 11 unique systems that remain.

As I said before, a fast system requires high performance in all areas. One thing you might not be aware of is 2D performance does not necessarily increase between series of graphics cards. It can be more about the CPU's single-core speed and the GPU's clock-rate. Thus, a highly overclocked 1080Ti may be able to easily outperform a regular 3090 in 2D. And again, 2D performance is extremely important, as that determines how fast most Windows GUI applications update, and even command-line text output performance in a command prompt window.

Since there are very many fewer 3090's available than 1080Ti's, the chance that someone is going to get a highly overclockable 3090 sample is not very good.

 

[M42]

Also, regarding duplicates, maybe some are. But I have on occasion selected exactly the same parts as people on the list. And because of that, I have built several systems that have been in the top 20, the highest being #4 for about one month.

 

[Conahl]

a i9-9960X and 1080 Ti are NOT really high performance any more. as cryoburner said " The "combined" scores for benchmarks like Passmark (along with Userbenchmark and others) are kind of a joke anyway, since not everyone is using their system to run the same kinds of software, and the weighting between the various tests (and choice of precisely what to test) is ultimately completely arbitrary. So the results really don't mean much. They are there just to drive traffic to their site, and little more. "

sorry to say, but if you look at reviews, 98% of them dont even test 2D any more, as chances are, any card released in the last few years, have fast enough 2D speeds, and is not needed to test any more. even my old GTX 560s are fast enough for general usage.

 

[M42]

a i9-9960X and 1080 Ti are NOT really high performance any more. as cryoburner said " The "combined" scores for benchmarks like Passmark (along with Userbenchmark and others) are kind of a joke anyway, since not everyone is using their system to run the same kinds of software, and the weighting between the various tests (and choice of precisely what to test) is ultimately completely arbitrary. So the results really don't mean much. They are there just to drive traffic to their site, and little more. "

sorry to say, but if you look at reviews, 98% of them dont even test 2D any more, as chances are, any card released in the last few years, have fast enough 2D speeds, and is not needed to test any more. even my old GTX 0s are fast enough for general usage.

That's your opinion. The reason why 2D is not often tested is that most people care about GAME performance, not Windows 2D GUI performance. I have built quite a number of systems, as well as purchased desktops and laptops for my software business. In my experience, the test results pretty much reflect how performant a system feels.

BTW, it seems like you feel the Passmark test is a joke because AMD CPUs are not showing very well. Maybe, you really think that CPU speed alone makes a system fast? Maybe it does for some things, but not everything. Most applications are not 3D., thus 2D performance is important.

 

[M42]

a i9-9960X and 1080 Ti are NOT really high performance any more. as cryoburner said " The "combined" scores for benchmarks like Passmark (along with Userbenchmark and others) are kind of a joke anyway, since not everyone is using their system to run the same kinds of software, and the weighting between the various tests (and choice of precisely what to test) is ultimately completely arbitrary. So the results really don't mean much. They are there just to drive traffic to their site, and little more. "

sorry to say, but if you look at reviews, 98% of them dont even test 2D any more, as chances are, any card released in the last few years, have fast enough 2D speeds, and is not needed to test any more. even my old GTX 560s are fast enough for general usage.

Sorry, one more post about this.

Here's an old Tom's Hardware article that looks into 2D performance of graphics cards at the time. The actual graphics cards mentioned are probably not relevant to most people now, but the content might be enlightening to some that think that the expensive 3D graphics cards are also always the best 2D performers:

https://www.tomshardware.com/reviews/2d-windows-gdi,2539.html

Here's a newer, but still old, article comparing 2D performance of graphics cards on LINUX. Note that not always does the most powerful graphics card win every benchmark:
https://www.phoronix.com/scan.php?page=article&item=amd_nvidia_2dhigh&num=1

My point is that you cannot judge 2D performance by the price nor the 3D performance of a graphics card.

 

[Conahl]

The reason why 2D is not often tested is that most people care about GAME performance, not Windows 2D GUI performance

nope, its because for most of the people that use a comp, the 2d performance that we have now, is fast enough. so there is no need to test it. those reviews are 6 and 10 years old now, i dont recall anyone testing 2d performance in the last 2-3 years, so that should also tell you, there is no need to test it any more.

and that is also your opinion, and i quote you M42 : " Maybe, you really think that CPU speed alone makes a system fast? " but considering you were pushing intel and how well those chips overclock, vs AMD., its seems like you were the one that seems to think cpu speed is what makes a PC fast. seems to me like you are for all intents and purposes, saying that if something doesnt overclock, to a certain point, it performs like crap. most of those posts you were making, made it sound like you either believe intel's marketing crap, or know for a fact, in order for intel to have any performance, they have to and need to overclock their cpus, or you work for intel's marketing department.

contrary to what you believe, and what your own opinions are, cryoburners post, pretty much refutes what you say, benchmarks like those, are pretty pointless, and meaningless. BTW it seems like you believe passmark and the like are not a joke, and need to be taken seriously, cause its one of the few tests, where intel is a head of amd, maybe thats why you are pushing these types of benchmarks, as intel has pretty much lost in all metrics, where, only 3 years ago, they were king.

 

[M42]

nope, its because for most of the people that use a comp, the 2d performance that we have now, is fast enough. so there is no need to test it. those reviews are 6 and 10 years old now, i dont recall anyone testing 2d performance in the last 2-3 years, so that should also tell you, there is no need to test it any more.

Yea, but "fast enough" is not what the "top 20" is about.

If you don't care about benchmarks, then don't do them, but I think they can be useful minimally to confirm your system is running at the performance level of computers with similar components.

and that is also your opinion, and i quote you M42 : " Maybe, you really think that CPU speed alone makes a system fast? " but considering you were pushing intel and how well those chips overclock, vs AMD., its seems like you were the one that seems to think cpu speed is what makes a PC fast. seems to me like you are for all intents and purposes, saying that if something doesnt overclock, to a certain point, it performs like crap. most of those posts you were making, made it sound like you either believe intel's marketing crap, or know for a fact, in order for intel to have any performance, they have to and need to overclock their cpus, or you work for intel's marketing department.

Oh, I would have loved to build a Zen 3 AMD system. I was online early enough the first day they went on sale to try to get a 5900x or 5950x, but all the bots scooped them up before any of the retail sites would respond. When availability and prices are back to retail I definitely plan on building a few Zen 3 systems.

That being said, I can't help but hope that the Intel 11000-series beats out the AMD CPUs so that all those AMD scalpers lose out on making money on future sales. I also hope that the retailers will work out a way to prevent the new Intel CPUs from being sold out by bot purchases. Either way, I will likely build a few AMD and Intel systems this year to replace some of the older systems we have.

contrary to what you believe, and what your own opinions are, cryoburners post, pretty much refutes what you say, benchmarks like those, are pretty pointless, and meaningless. BTW it seems like you believe passmark and the like are not a joke, and need to be taken seriously, cause its one of the few tests, where intel is a head of amd, maybe thats why you are pushing these types of benchmarks, as intel has pretty much lost in all metrics, where, only 3 years ago, they were king.

Whatever... I'm just pointing out that CPU performance has to be balanced with other components. This is a universal rule, not something specific to Intel or AMD.

And, BTW, for the type of software development work we do, 2D graphics performance is way more important than 3D performance. I am not opposed to AMD at all. I am just not willing to pay over retail so scalpers can make a profit.

 

[Conahl]

Yea, but "fast enough" is not what the "top 20" is about.
If you don't care about benchmarks, then don't do them, but I think they can be useful minimally to confirm your system is running at the performance level of computers with similar components.

and that top 20, is based on benchmarks that very few would even look at, as it and others like it, use questionable methods that also seem to lean heavily on certain tasks, that favor one side or the others strengths and that makes it a little invalid. IE, leaning on AVX 512 to get a performance advantage, that amd cant match, as it doesnt use that.

i do care about benchmarks, only when i am looking to buy new hardware, and passmark, userbench, and geek bench, are ones i practically ignore for the small paragraph i just typed above

That being said, I can't help but hope that the Intel 11000-series beats out the AMD CPUs so that all those AMD scalpers lose out on making money on future sales.

they dont have to beat the AMD cpus., the cpu's just need to be in stock, if they are, then no need to go to scalpers. im looking at a 5900x myself, but im in no rush to get it, and even if i was, i sure has he-double-hocky-sticks, wouldn't go to a scalper to get it.

Whatever... I'm just pointing out that CPU performance has to be balanced with other components. This is a universal rule, not something specific to Intel or AMD.

the funny part, you are the only one that is pushing this. i dont recall, or see, and one else doing this, so it seems to be 100% only your opinion.

 

[M42]

the funny part, you are the only one that is pushing this. i dont recall, or see, and one else doing this, so it seems to be 100% only your opinion.

It comes from many years of experience building and testing computers, and common sense. It also applies to many other things in life.

 

[dayzero]

Why do the numbers on the "Average FPS (Geomean), Entire Test Suite 1440p" chart in section "AMD vs Intel CPU Gaming Performance" not match the numbers in the article "CPU Benchmark and Hierarchy 2021: Intel and AMD Processors Ranked"?
For example, the chart in this article shows that Intel Core i9-10900K has the top performance of all, whereas the other article shows Ryzen 9 5900X having top performance.

 

[King_V]

Why do the numbers on the "Average FPS (Geomean), Entire Test Suite 1440p" chart in section "AMD vs Intel CPU Gaming Performance" not match the numbers in the article "CPU Benchmark and Hierarchy 2021: Intel and AMD Processors Ranked"?
For example, the chart in this article shows that Intel Core i9-10900K has the top performance of all, whereas the other article shows Ryzen 9 5900X having top performance.

Because they're using out of the box scores, and not overclocked scores. Rightfully so.

 

[timf79]

So why is that in Excel my i7-10610U is about 4 times faster than my Ryzen 9 5950x?

 

[King_V]

So why is that in Excel my i7-10610U is about 4 times faster than my Ryzen 9 5950x?

 

[spongiemaster]

So why is that in Excel my i7-10610U is about 4 times faster than my Ryzen 9 5950x?

Make sure your 5950x system is turned on before you start trying to use it.

 

[timf79]

Make sure your 5950x system is turned on before you start trying to use it.

Haha,

I have a theory that Excel only uses one core, but the time difference is staggering.
test was conducted as follows:

• Excel file with chained formulas
• VBA code to loop multiple times
• VBA code uses random number at beginning of Formula chain, measures time to completion
• time of cycles is averaged.

Run this multiple times (same file) and the i7 was consistently 4 times faster

 

[TerryLaze]

I have a theory that Excel only uses one core, but the time difference is staggering.

Run process explorer double click on the task that does the work go to the threads tab and have a look, it will show you how many threads the task runs and how much CPU power each one uses.
There are definitely some things in excel that do use multithreading but not everything can be multithreaded.
Chained formulas as in the second one needs the results of the first one would definitely not be able to use much, if any, multithreading.

 

[timf79]

Run process explorer double click on the task that does the work go to the threads tab and have a look, it will show you how many threads the task runs and how much CPU power each one uses.
There are definitely some things in excel that do use multithreading but not everything can be multithreaded.
Chained formulas as in the second one needs the results of the first one would definitely not be able to use much, if any, multithreading.

Interestingly enough it shows 244 threads an only one goes to ~3%
looking at the CPU graph and selecting to show "one graph per CPU", none of the 32 cores maxes out.
In excel it shows calculating using 32 cores...