9 Years of AMD CPUs: From AMD FX to Ryzen 5000 Series, Tested

DZillaXx

Posts: 413   +530
M1 vs Ryzen 5800U is not valid comparison since M1 is not CPU. In other words, it has no support for external memory.
The M1 is a CPU

The memory sits on the same package... So it is external, not that it defines what a CPU is.

By definition a CPU is a CPU by having a integer unit and a floating point unit. Which is M1 has....

The FX Chips are considered quad core even though they have 8 integer units is because they only have 4 floating point units. Each module of two integer units share a single floating point unit.
 

HardReset

Posts: 1,310   +983
The M1 is a CPU

The memory sits on the same package... So it is external, not that it defines what a CPU is.

By definition a CPU is a CPU by having a integer unit and a floating point unit. Which is M1 has....
It's about definition. While AMD CPU's with integrated graphics can be considered CPU's, AMD defines them as APU's. To make clear that they have something you would not expect CPU to have.

Not every ASIC's are considered CPU even if they have integer unit/ALU. Floating point units are missing from many CPU's.

Point is that talking M1 as CPU is pointless since it has no support for externally added memory that makes it very different vs Ryzen CPU's, right, CPU's. Comparing CPU vs CPU when there is at least major difference makes comparison pretty invalid.
The FX Chips are considered quad core even though they have 8 integer units is because they only have 4 floating point units. Each module of two integer units share a single floating point unit.
This is very easy to probe with mathematics.

If FX-8350 is quad core, it has 4 cores. Disable 4 cores from FX-8350, it has zero cores then.

4-4=0

However it still has 4 cores.

0=4

Does not compute. I cannot see why there anyone could consider FX-8350 as quad core since even Windows 7 sees it as 8 core...
 
FX series is NOT SMT. Two cores are combined into a single module, which has some shared resources, but still separate processing units.

The FX-8350 has eight cores. Period. Just they come in pairs of cores.
 

Sausagemeat

Posts: 1,036   +863
FX series is NOT SMT. Two cores are combined into a single module, which has some shared resources, but still separate processing units.

The FX-8350 has eight cores. Period. Just they come in pairs of cores.
FX 8350 definitely had 8 cores on the die. However in most applications only half of them could be used.

It’s like selling a car with 8 wheels But half of them are on the top. Sure there are probably some tunnels you can drive down that use them all. But for most uses it’s stuck with just 4 wheels.

The lawsuit was about AMDs marketing. AMD definitely did mislead people into thinking they had a full 8 core when realistically it was a part time 8 core.

4 core SMT would have given most applications more cores to work with than the 8350. The judge was right to fine AMD.
 

HardReset

Posts: 1,310   +983
FX 8350 definitely had 8 cores on the die. However in most applications only half of them could be used.
Wtf is that supposed to mean? In most applications you can use only single core, no matter how many you have got.

FX-8350 utilized all 8 cores as long as software supports at least 8 cores.
It’s like selling a car with 8 wheels But half of them are on the top. Sure there are probably some tunnels you can drive down that use them all. But for most uses it’s stuck with just 4 wheels.

The lawsuit was about AMDs marketing. AMD definitely did mislead people into thinking they had a full 8 core when realistically it was a part time 8 core.
It was all time 8 core. If you even bother to read lawsuit, it was about someone believing 8 cores would be twice as fast than 4 cores. That never happens anyway.
4 core SMT would have given most applications more cores to work with than the 8350. The judge was right to fine AMD.
Once again, you have absolutely no idea about this.

And once again, judge didn't decide anything.
 

maxxcool7421

Posts: 64   +89
"Some might still argue that the FX-8350 is an 8 core CPU, but it’s not, and it’s certainly not according to the California Consumers Legal Remedies Act."

I don't care what some court says about tech, especially in some absurd civil case in a country where you can legally sue ANYBODY for ANYTHING. The CPU core was defined as an integer computing core since the original Intel 4004. According to the "California Consumers Legal Remedies Act", Intel, AMD, Motorola, Texas Instruments (Cyrix), SPARC, IBM and VIA could ALL be sued for selling CPUs that didn't have FPUs attached. The FPU isn't even x86 architecture, it's x87 architecture and used to be (pre-486DX) sold separately as a "math co-processor" for floating-point operations.

Steve, I'm surprised at you that you'd let some court in another country make you ignore all the knowledge that you have about CPUs. Since when are lawsuits in the USA determined by who is right as opposed to "who just doesn't want to deal with it anymore"?

No court in some US state is going to define what a CPU core is for me when I've known damn well what a CPU core is for more than 20 years before that frivolous lawsuit case was brought forward. I wouldn't be surprised if it turned out that Intel had something to do with it.

Both iterations of Bulldozer and piledriver could not execute simultaneous streams on the shared pipeline resources .. if you cannot run 8 threads on 8 cores , but only run 4 threads on 8 cores and ''occasionally' run extra cycles of shared data... that is a quad core cpu.
 

Avro Arrow

Posts: 1,851   +2,205
TechSpot Elite
That is because Bulldozer was supposed to reach at least 5 GHz clock speeds with decent cooling. Due to some design flaws (L2 cache limiting clock speed) that didn't happen.
I agree which is why I said it wasn't good for mobile. Decent cooling in a craptop is hard to find and who wants the heat output of a 5GHz CPU on their lap? The battery life would probably also be reduced to 30 minutes unless they used a battery that weighed as much as the craptop itself. :laughing:
 

Avro Arrow

Posts: 1,851   +2,205
TechSpot Elite
"Some might still argue that the FX-8350 is an 8 core CPU, but it’s not"

Pentium 4 has zero cores.
Actually, going by the argument that was used against AMD, all Pentiums would have the number of "CPU cores" advertised because all CPUs starting with the 80486DX had the math-coprocessor on the same die as the CPU. The math co-processor would become known as the FPU.
 

HardReset

Posts: 1,310   +983
Both iterations of Bulldozer and piledriver could not execute simultaneous streams on the shared pipeline resources .. if you cannot run 8 threads on 8 cores , but only run 4 threads on 8 cores and ''occasionally' run extra cycles of shared data... that is a quad core cpu.
Your logic is flawed since shared resources on front and are not execution units. And execution units are doubled and can run two threads simultaneously and independently.

Also like I explained before, you can disable 4 cores from 8 and still you have 4 cores. Not possible if you don't have 8 cores before disabling.
 

Targonis

Posts: 14   +14
AMD came from the literal bottom, scrapping everything they had in development followed by 5 years off, so of course their improvements look better compared to a company that struggled with 10nm they were responsible for designing and manufacturing, but DIDN'T quit. Yet AMD gets all the praise?

LMAO.
Intel came from the literal top, and did everything wrong to lose its advantage. Intel couldn't come up with a new design for 14nm until Rocket Lake(which didn't even perform better) for example.

AMD is still much smaller than Intel, doesn't have as many resources or employees, yet has been pushing very good performance improvements year after year. So yea, AMD is getting praise for not just going from near bankruptcy to profitability, but CPUs that are faster than anything Intel has. All this talk about Alder Lake sounds a lot like Rocket Lake, which wasn't faster than previous Intel chips outside of benchmarks.
 

Avro Arrow

Posts: 1,851   +2,205
TechSpot Elite
Both iterations of Bulldozer and piledriver could not execute simultaneous streams on the shared pipeline resources .. if you cannot run 8 threads on 8 cores , but only run 4 threads on 8 cores and ''occasionally' run extra cycles of shared data... that is a quad core cpu.
Except that's not what the issue was (and you're wrong regardless). They were whining about the fact that two integer cores (the x86 CPU cores) were sharing a single floating point core and so they tried to argue that it wasn't a "true" 8-core CPU despite the fact that Windows still recognised it as such. It was a case of computer illiterates who thought that they were geniuses and so they screwed up royally but tried to blame AMD for it.

I HAVE an FX-8350 and used it for five years. I know that it's an 8-core CPU because it functioned EXACTLY as an octocore CPU should. Do you or did you actually OWN one of these things or are you just assuming that you know these CPUs better than people who actually own them?

Calling an FX-8350 a quad-core CPU is nothing that an intelligent person well-versed in tech does. It's what someone says when they have no clue but WANT it to be the case.
 

DZillaXx

Posts: 413   +530
Your logic is flawed since shared resources on front and are not execution units. And execution units are doubled and can run two threads simultaneously and independently.

Also like I explained before, you can disable 4 cores from 8 and still you have 4 cores. Not possible if you don't have 8 cores before disabling.
While I do agree that what we normally refer to "cores" are just the integer core of the Chip, there is a lot more to a "core" than just the integer unit.

The reason why the Bios for a FX chip shows 8 core is simply because of it having 8 integer units.

But the issue with the FX chip is that two integer chips per module shares resources. In a Module the two Integer units share the fetch and instruction decoder. This is something that each single "Core" would normally have. You wouldn't have two "Cores" sharing the front end.

This shared front end is really a bottleneck if not able to feed both integer units 100%, which clearly did not happen. In fact while AMD made big gains with Excavator, we never saw a high end product using those cores.

On top of the shared front end they share a single FPU. Normally in a "Core" there is a single FPU and a single integer unit. AMD's Module had two integer units and a single FPU. In later revisions on the Microarch, the FPU was made bigger and had the ability to act as two independent FPU's or a single large unit.

The last big issue was the shared L2 Cache and limited amount of L1 Cache. Unlike most "Cores" where the single integer units has it's own dedicated L2 Cache. In AMD's Module two integer units share a single pool of L2 cache. The amount of L2 cache was pretty small per module considering the two integer units.

So while the FX Chip does have 8 integer units, it only has 4 modules. And the modules are configured more like traditional cores we come to think of. A Core being a single integer unit and a floating point unit with a dedicated front end and backend. A integer unit by itself is not consider a core.

The bottlenecks put in place in the FX Arch kept the chips from performing like a real 8 core chip. These are Quad Module chips.

Integer units performance most of the general purpose computing, and these are the core Window's looks at for core count. This is where most of the processing is done. The only issues is being able to feed data into both integer units without slowdown, the module idea could have worked if AMD kept pushing it. But it really is a picky design and is a better idea on paper than it is in real life.

We never got a high end Excavator chip, so we have no idea if AMD was able to help the balance issue.
 

hahahanoobs

Posts: 3,835   +1,908
Intel came from the literal top, and did everything wrong to lose its advantage. Intel couldn't come up with a new design for 14nm until Rocket Lake(which didn't even perform better) for example.

AMD is still much smaller than Intel, doesn't have as many resources or employees, yet has been pushing very good performance improvements year after year. So yea, AMD is getting praise for not just going from near bankruptcy to profitability, but CPUs that are faster than anything Intel has. All this talk about Alder Lake sounds a lot like Rocket Lake, which wasn't faster than previous Intel chips outside of benchmarks.
Fabless AMD had 5 years off. Intel was doing everything and they hit a snag. AMD's snags lead to acquisition rumours. Remember those.

I can't brag about going 30-5 against players with their controllers disconnecting all the time in a MP game. Not the best example, but you get the idea.

I'll be impressed when both are firing on all cylinders and AMD CPUs can remain on top. Wake me when that happens. Everything you said I've heard before. Been hearing it since Conroe. When AMD is consistent in a fair fight I'll be impressed. I said when Zen came out and better capatilize, and then we saw Intel basically hold AMD off with one node versus the three AMD used.

Alder lake, Meteor lake and Zen 4 are the chips to watch now.
 
Last edited:

HardReset

Posts: 1,310   +983
While I do agree that what we normally refer to "cores" are just the integer core of the Chip, there is a lot more to a "core" than just the integer unit.

The reason why the Bios for a FX chip shows 8 core is simply because of it having 8 integer units.

But the issue with the FX chip is that two integer chips per module shares resources. In a Module the two Integer units share the fetch and instruction decoder. This is something that each single "Core" would normally have. You wouldn't have two "Cores" sharing the front end.

This shared front end is really a bottleneck if not able to feed both integer units 100%, which clearly did not happen. In fact while AMD made big gains with Excavator, we never saw a high end product using those cores.
While I partially agree, shared resources cannot determine amount of cores. Since virtually every CPU that has more than single core have shared resources, at least some way.

Easiest example is memory controller that is almost every time shared with some or all cores. Another example is L3 or even L2 cache. Intel's Core 2 shares L2 cache with two cores, most modern CPU's and AMD's Phenom/Athlon64 etc do not. Core 2 L2 cache is very vulnerable to cache pollution and even leaving that out of equation, Core 2 L2 cache performs much worse with two cores than with one core. Therefore we can easily argue Core 2 CPUs are not dual core because they share L2 cache. Normally core has own L2 cache, this is very similar example of Bulldozer's shared front end.

So I don't agree. CPU's always share something that almost every time mean dual core performance is not single core performance *2.

Excavator have also shared front end (fetch). How does that affect on core count?
On top of the shared front end they share a single FPU. Normally in a "Core" there is a single FPU and a single integer unit. AMD's Module had two integer units and a single FPU. In later revisions on the Microarch, the FPU was made bigger and had the ability to act as two independent FPU's or a single large unit.
Like I stated above, there are many examples where "normally" something exists or does not exist. And then there are exceptions. If only "normal" cores count, we probably have only single core CPU's right now.
The last big issue was the shared L2 Cache and limited amount of L1 Cache. Unlike most "Cores" where the single integer units has it's own dedicated L2 Cache. In AMD's Module two integer units share a single pool of L2 cache. The amount of L2 cache was pretty small per module considering the two integer units.

So while the FX Chip does have 8 integer units, it only has 4 modules. And the modules are configured more like traditional cores we come to think of. A Core being a single integer unit and a floating point unit with a dedicated front end and backend. A integer unit by itself is not consider a core.
Bulldozer had quite big L2 cache, 2MB total or 1MB per integer unit is pretty big. That's double vs Zen2 or quadruple vs Sandy Bridge.

Like I said, you can have infinite definitions for "core". You may well define core must have straight access to system memory, that each core must have FPU, that each core must have own L2 cache, that each core must have...

Problem is that integer unit simply cannot work just itself, it needs much more. And Bulldozer integer units have everything it needs to work as a core.
The bottlenecks put in place in the FX Arch kept the chips from performing like a real 8 core chip. These are Quad Module chips.

Integer units performance most of the general purpose computing, and these are the core Window's looks at for core count. This is where most of the processing is done. The only issues is being able to feed data into both integer units without slowdown, the module idea could have worked if AMD kept pushing it. But it really is a picky design and is a better idea on paper than it is in real life.

We never got a high end Excavator chip, so we have no idea if AMD was able to help the balance issue.
Quad module but also octa core. No matter how badly it performs, you can:

- Disable 4 cores and still have 8 cores
- Execute 8 threads simultaneously

I consider those two things more than enough to prove FX-8350 is octa core. Because on SMT CPU you cannot do first one (unless counting logical cores as cores).

AMD was supposed to have Excavator successor where problems were solved but since most efforts were put on Zen, we never saw what FX series could be. Also Bulldozer was at least third major architecture after K8. Those two (or more) never saw light. That also explains why FX was probably rushed one.
 

DZillaXx

Posts: 413   +530
While I partially agree, shared resources cannot determine amount of cores. Since virtually every CPU that has more than single core have shared resources, at least some way.

Easiest example is memory controller that is almost every time shared with some or all cores. Another example is L3 or even L2 cache. Intel's Core 2 shares L2 cache with two cores, most modern CPU's and AMD's Phenom/Athlon64 etc do not. Core 2 L2 cache is very vulnerable to cache pollution and even leaving that out of equation, Core 2 L2 cache performs much worse with two cores than with one core. Therefore we can easily argue Core 2 CPUs are not dual core because they share L2 cache. Normally core has own L2 cache, this is very similar example of Bulldozer's shared front end.

So I don't agree. CPU's always share something that almost every time mean dual core performance is not single core performance *2.

Excavator have also shared front end (fetch). How does that affect on core count?

Like I stated above, there are many examples where "normally" something exists or does not exist. And then there are exceptions. If only "normal" cores count, we probably have only single core CPU's right now.

Bulldozer had quite big L2 cache, 2MB total or 1MB per integer unit is pretty big. That's double vs Zen2 or quadruple vs Sandy Bridge.

Like I said, you can have infinite definitions for "core". You may well define core must have straight access to system memory, that each core must have FPU, that each core must have own L2 cache, that each core must have...

Problem is that integer unit simply cannot work just itself, it needs much more. And Bulldozer integer units have everything it needs to work as a core.

Quad module but also octa core. No matter how badly it performs, you can:

- Disable 4 cores and still have 8 cores
- Execute 8 threads simultaneously

I consider those two things more than enough to prove FX-8350 is octa core. Because on SMT CPU you cannot do first one (unless counting logical cores as cores).

AMD was supposed to have Excavator successor where problems were solved but since most efforts were put on Zen, we never saw what FX series could be. Also Bulldozer was at least third major architecture after K8. Those two (or more) never saw light. That also explains why FX was probably rushed one.
Just because the FX Bios calls them cores, doesn't mean they are. You are doing nothing more than disabling 4 integer units.

AMD's FX Chips had their user cases, and for awhile were decent chips for the price. For gaming they were horrible from the start and even the old Sandy Bridge 2600k in modern games still packs a decent punch with a small OC and good RAM.

They really were AMD's own Pentium 4. A horrible choice that nearly bankrupted the company. If AMD wasn't so mismanaged at that point in time we would have just gotten Phenom III.

Sadly FX Chips at their best workloads were only little faster than Intel Quad cores. With twice the integer cores.... Luckily for us Zen is the best thing that came out of the entire situation, and if it wasn't for bulldozer AMD would probably be on Phenom VI by now and like intel stuck with a dated arch that needs some major changes. Especially in terms of cheap scaling.
 

HardReset

Posts: 1,310   +983
Just because the FX Bios calls them cores, doesn't mean they are. You are doing nothing more than disabling 4 integer units.
Units that can independently execute integer code. That's about best definition for core I can make.
AMD's FX Chips had their user cases, and for awhile were decent chips for the price. For gaming they were horrible from the start and even the old Sandy Bridge 2600k in modern games still packs a decent punch with a small OC and good RAM.

They really were AMD's own Pentium 4. A horrible choice that nearly bankrupted the company. If AMD wasn't so mismanaged at that point in time we would have just gotten Phenom III.

Sadly FX Chips at their best workloads were only little faster than Intel Quad cores. With twice the integer cores.... Luckily for us Zen is the best thing that came out of the entire situation, and if it wasn't for bulldozer AMD would probably be on Phenom VI by now and like intel stuck with a dated arch that needs some major changes. Especially in terms of cheap scaling.
AMD chose between Phenom III and Bulldozer. You wouldn't just throw away years of development. Intel did same with Pentium 4.

Probably yes. Failure of Bulldozer encouraged AMD to basically abandon FX-development, although some remaining projects were still completed (Steamroller and Excavator). However, remembering Bulldozer and Zen had same lead architect, many things could have happened...
 

BMAN61

Posts: 15   +7
Say what you will about the Phenom series; my brother's PC is equipped with my first PC build's CPU, a Phenom 965BE (Black Edition) and an ASUS 990FX motherboard. He doesn't game or do video editing or anything like that, between him and his wife they browse the net, and watch Netflix - and it's been working just fine since the day I built my first PC.

Phenom II gets all the attention, but IMO I think the original Phenom series was better tan the second.
 

HardReset

Posts: 1,310   +983
Say what you will about the Phenom series; my brother's PC is equipped with my first PC build's CPU, a Phenom 965BE (Black Edition) and an ASUS 990FX motherboard. He doesn't game or do video editing or anything like that, between him and his wife they browse the net, and watch Netflix - and it's been working just fine since the day I built my first PC.

Phenom II gets all the attention, but IMO I think the original Phenom series was better tan the second.
965BE is Phenom II series...
 

DZillaXx

Posts: 413   +530
Say what you will about the Phenom series; my brother's PC is equipped with my first PC build's CPU, a Phenom 965BE (Black Edition) and an ASUS 990FX motherboard. He doesn't game or do video editing or anything like that, between him and his wife they browse the net, and watch Netflix - and it's been working just fine since the day I built my first PC.

Phenom II gets all the attention, but IMO I think the original Phenom series was better tan the second.
^ Yes the 965BE is Phenom II

The Original Phenom was horrible. It launched with a bug, that delayed the rollout big time. They had fab issues that really limited clock speed. It really was a mess. Clock Speed alone was a big issue with Phenom "1" chips. Phenom II was not a major change up, and started off with low clock speeds as well. But Late in Life Phenom II chips with that last stepping revision were great overclockers and honestly held higher clocks easier than late in life C2Q's and first gen i7's. The higher end chips also had great base clocks.

The Phenom II X6 was by far one of the best chips of its era. The 1090T BE was amazing. @4ghz+ that is. Only thing it lacked with SMT.

Sadly Sandy Bridge came along and anything AMD was just pointless if you wanted your machine to have the best performance while gaming. The FX series didn't come close.
 

ZackL04

Posts: 744   +566
I enjoy these “classic” articles more than new tech articles lately.

How about a revisit of the 1st, 2nd, 4th, 7th, and 11th gen intel core products vs their amd counter parts too?

Fun stuff. What I miss most is the game benchmarks that had popular CPU’s on the last page of the article….