Question CPU Cores?

[Phaetos]

Considering I’ve been out of the nitty gritty tech side for a long time and my other thread on upgrading an FX-6300, I have a question.
My FX6300 is advertised as a 6-core processor. Windows reports it as 3 physical and 6 logical cores. So it actually only has 3 REAL cores, but divides itself out as 6 in software?

Newer processors that are 6/8/12(etc) cores, are they actually 6/8/12 physical actual cores on the die, or will I find an 8 core that reports as4 physical and 8 logical?

And if that is the case, how do they get away with advertising the higher core counts, but only deliver half of what is actually there?

 

[Yotsugi]

So it actually only has 3 REAL cores, but divides itself out as 6 in software?

No, these are 6 real cores, just narrow ones and pairs share the FPUs.

Newer processors that are 6/8/12(etc) cores, are they actually 6/8/12 physical actual cores on the die

They're %whatever% amount of big, fat, juicy x86 cores.

 

[Markfw]

No, these are 6 real cores, just narrow ones and pairs share the FPUs.

They're %whatever% amount of big, fat, juicy x86 cores.

Example. 3600 Ryzen for $200 i s6 real cores, 12 real threads. 3700x is 8 real cores, 16 real threads. 3900x is 12 real cores, 24 real threads.

Quad core are a thing of the past.

 

[Phaetos]

So my assumption was correct, those old processors were in essence false advertising. Were they ever called out on this BS?

 

[Markfw]

So my assumption was correct, those old processors were in essence false advertising. Were they ever called out on this BS?

Whick ones ? example. A quad core hyperthreading in 8 threads. Am AMD system that said 8 core might have been.

 

[VirtualLarry]

Were they ever called out on this BS?

There were class-action lawsuits. And I think, a settlement. That in the end didn't pan out for anyone.

Think of it this way. AMD's FX CPUs, were kind of like, duplex-townhouses. Only four buildings, but both halves had their own front entrance, but they shared a boiler in the basement. As opposed to "separate cores", which are like individual houses.

 

[Maxima1]

So my assumption was correct, those old processors were in essence false advertising. Were they ever called out on this BS?

Why didn't you research before buying? Why would you stake your purchase into just the trivial info of how many cores there are, especially when it's being sold cheaper than the competition? Two different cores could vary drastically in capability obviously, which is what occured here.

 

[scannall]

So my assumption was correct, those old processors were in essence false advertising. Were they ever called out on this BS?

It was 6 cores. Just narrow ones, as Yotsugi said. A few interesting ideas in the construction cores, but really bad implementation. They didn't go wide enough to keep it fed, amoungst other problems. It was an OK CPU, not great. But OK. Held fairly strong against Nehalem, behind a bit but once Sandy Bridge hit it was over. At that point AMD had to decide between actually fixing the Construction cores, or starting from scratch. Starting from scratch was likely a lot more painful, but in the long run a better way to go.

The next several years should be entertaining if nothing else. Intel is now getting to the place of choosing to start from scratch, or keep refining the Core architecture. Personally, I thing Core is pretty much done for. Too much work to do on security, before even thinking of increasing performance. They have the talent to do a from scratch architecture, but I'm not sure they have the corporate structure to pull it off.

 

[Thunder 57]

Why didn't you research before buying? Why would you stake your purchase into just the trivial info of how many cores there are, especially when it's being sold cheaper than the competition? Two different cores could vary drastically in capability obviously, which is what occured here.

I don't get how caveat emptor doesn't seem to be a thing anymore.

Anandtech discussed the whole core thing and the litigation here. There's also a helpful image there. There are six integer cores, but some resources are shared. It didn't pan out well.

 

[moinmoin]

So my assumption was correct, those old processors were in essence false advertising. Were they ever called out on this BS?

We recently (as in less than two months ago) had a thread about a lawsuit and its settlement regarding this very topic: https://forums.anandtech.com/thread...ht-core-bulldozer-advertising-fiasco.2569383/

 

[VirtualLarry]

Intel to Discontinue Nearly All Desktop Kaby Lake CPUs

www.anandtech.com

Then there's this. I was just looking at ebay at maybe getting a i5-6400 or i5-6500, for a B150 K4/Hyper (BCLK OC) board, they're around $100-120, but if Intel discontinues them, who knows what they will shoot up to.

G4560, I hardly knew ye! Well, we shared some good times together, in my Deskmini PCs. Sad to see you go. Such value for a CPU, was virtually unheard of in Intel-land. (i3 performance, for half price? Yes, please!)

Now i3 CPUs, are quad-core, and the i3-9100F, even has Turbo now! Another great bargain CPU, if you don't mind being limited to 4 cores. (You can get Ryzen R5 2600 6C/12T that will OC to 4.0Ghz in most cases, for $120 these days. A much better deal.)

 

[ondma]

It was 6 cores. Just narrow ones, as Yotsugi said. A few interesting ideas in the construction cores, but really bad implementation. They didn't go wide enough to keep it fed, amoungst other problems. It was an OK CPU, not great. But OK. Held fairly strong against Nehalem, behind a bit but once Sandy Bridge hit it was over. At that point AMD had to decide between actually fixing the Construction cores, or starting from scratch. Starting from scratch was likely a lot more painful, but in the long run a better way to go.

The next several years should be entertaining if nothing else. Intel is now getting to the place of choosing to start from scratch, or keep refining the Core architecture. Personally, I thing Core is pretty much done for. Too much work to do on security, before even thinking of increasing performance. They have the talent to do a from scratch architecture, but I'm not sure they have the corporate structure to pull it off.

Actually Sandy Bridge was released before the construction cores. Phenom was fairly competitive, but Bulldozer was a failure from the beginning.

 

[Phaetos]

Why didn't you research before buying? Why would you stake your purchase into just the trivial info of how many cores there are, especially when it's being sold cheaper than the competition? Two different cores could vary drastically in capability obviously, which is what occured here.

Well, it was built several years ago as a VM server and based on info and fundingat the time, it was fine for its purpose and worked well. I never really looked at how it was reporting.

 

[Phaetos]

We recently (as in less than two months ago) had a thread about a lawsuit and its settlement regarding this very topic: https://forums.anandtech.com/thread...ht-core-bulldozer-advertising-fiasco.2569383/

Well, I have been out of the deep technical details for a few years. Read up enough to get by and sorta keep up.

 

[Thunder 57]

Well, it was built several years ago as a VM server and based on info and fundingat the time, it was fine for its purpose and worked well. I never really looked at how it was reporting.

That was one of the few scenarios the FX series was good at. There were some other use cases that it was also good at, especially for the price. The problem was that for things people do most often, it really wasn't all that great.

 

[Phaetos]

That was one of the few scenarios the FX series was good at. There were some other use cases that it was also good at, especially for the price. The problem was that for things people do most often, it really wasn't all that great.

It’s definitely a beast compared to my other pc which used an AMD CPU/GPU combo. Even when I got a 750Ti in it, it was still a horrible POS.

 

[scannall]

Actually Sandy Bridge was released before the construction cores. Phenom was fairly competitive, but Bulldozer was a failure from the beginning.

There was more potential in those cores than was ever used. This was back in the Hector Ruinz day. He was busy building golden parachutes for himself, and actual performance didn't matter. I have no concrete info or anything, just an old nerds speculation. But to actually feed the thing would have used a lot more die area. And that's cost. Cuts into the comfy cushions you want, ya know.

 

[NostaSeronx]

My FX6300 is advertised as a 6-core processor. Windows reports it as 3 physical and 6 logical cores. So it actually only has 3 REAL cores, but divides itself out as 6 in software?

It is physically a six-core processor. The issue is that AMD/Intel have been calling "processors" as "cores" since K8/Core.



^-- this details everything, especially the text.
"The FPU interfaces to the AMD-K7 core, which sends it instructions, load data, and guides the retirement of instructions."
"Complex x86 FPU instructions (e.g. transcendentals) are received from the core directly as a series of microcoded ops which are easily mapped into FPU execution ops."
"On completion of an operation, the result is written to the destination register, and completion status is sent to the core, which enables the core to retire the op."

The Core has always been where the retire queue is at. Family 15h calls it "the instruction scheduler and retirement control" and there is two per module.

^-- there are two of these per-module. The LSUs are replicated per core, also solving the load data. Retire quere + Load-store = dispatch of fpu instructions, load data, and retirement for both core contexts in the FPU. Bulldozer is contextually the AMD-K10 processor module(all things), and each core is the AMD-K10 processor core(Retire, Execution, Data). With the FPU and Front being components of the module, not the core. What people call K10 is not actually K10, but K9 hence the reference to dogs; Greyhound(Agena), Greyhound+(Deneb, Thuban), Deerhound(Barcelona), Husky(Llano), etc.


Your FX-6300 is physically three processors, each processor in this case contains two processor cores. The central processor being a FX-6300 is thus a six-core processor.

Newer processors that are 6/8/12(etc) cores, are they actually 6/8/12 physical actual cores on the die, or will I find an 8 core that reports as4 physical and 8 logical?

There have been no new CMT DT-CPU processors since Vishera/Centurion in x86. All central processors thus should be 1(processor):1(core)(8 cores equal 8 processors). There should be no cases of 1(processor):>2(cores)(8 cores equals 4/2/1 processor(s)).

 

[crashtech]

When maintaining a benchmark thread as I did for several years, it became evident that it was actually to AMD's advantage to consider FX CPUs as having half the cores as claimed in marketing materials. In reality, the FX-6300 was a 3C/6T CPU which did quite well against its contemporaries if considered as such. If it was accidentally considered an actual 6C CPU, it was destroyed by the likes of the the 3930K et al, which were actual 6C/12T CPUs.

That said, AMDs current implementation of SMT in the form of Zen 2 needs no marketing magic, it is the equal or better of Intel (clock for clock), so picking a CPU today is much less fraught with marketing hype than in the past.

 

[NostaSeronx]

When maintaining a benchmark thread as I did for several years, it became evident that it was actually to AMD's advantage to consider FX CPUs as having half the cores as claimed in marketing materials. In reality, the FX-6300 was a 3C/6T CPU which did quite well against its contemporaries if considered as such.

However, in actual reality, the FX-6300 is a 6C/6T CPU. Benchmarks don't determine core counts. Actual cores do.

 

[crashtech]

However, in actual reality, the FX-6300 is a 6C/6T CPU. Benchmarks don't determine core counts. Actual cores do.

Benchmarks are where the rubber meets the road. FX got killed in the category AMD (and, seemingly you) purported it should run in. It's simply a matter of historical record, cut and long dried. Not worth fighting again, especially now that promises have been kept with current iterations.

 

[NostaSeronx]

Benchmarks are where the rubber meets the road.

However, they do not determine quantity of cores.

For example, in your argument a SMT8 core that has each thread with the performance of the previous single-threaded core, would be an eight core. Which isn't what is actually going on, physically there aren't eight cores, it is just one core instead. In the other direction, a SMT2 core has 1/3rd the performance per thread than the previous single-threaded core, it would be a zero core(1/6 rounded down).

 

[soresu]

However, in actual reality, the FX-6300 is a 6C/6T CPU. Benchmarks don't determine core counts. Actual cores do.

Nooooo, just no.

It was 3 separate 'modules', that each had 2 narrow (2 wide?) integer cores and a shared 2 part FPU. Putting an exact thread count on it on further confuses the issue because INT and FP were inequal resources in the module.

Consequently a combination of narrow integer execution power and FPU/FMAC resource contention made for a practical implementation far inferior to whatever high efficiency nirvana they had promised during the days of K10.

I had both a Deneb (K10.5) quad core and a Piledriver 8320, so I'm certainly not confused on that score.

Having said that, it would seem that Windows did not adequately schedule for the Bulldozer uArch for some time, given that Windows 10 provided improvements in this area supposedly, I never found out myself as I put off installing W10 until I bought my Ryzen 1700 in 2017.

 

[NostaSeronx]

Nooooo, just no.

FX-6300 is a six-core.

It was 3 separate 'modules', that each had 2 narrow (2 wide?) integer cores and a shared 2 part FPU. Putting an exact thread count on it on further confuses the issue because INT and FP were inequal resources in the module.

It is three active replicated module microarchitectures, which contain two core within. The FPU/FE is part of the module of itself, not part of any cores within Bulldozer. A single of implementation of Bulldozer has exactly two cores which of course, exactly processes two physical threads.

 

[soresu]

FX-6300 is a six-core.It is three active replicated module microarchitectures, which contain two core within. The FPU/FE is part of the module of itself, not part of any cores within Bulldozer. A single of implementation of Bulldozer has exactly two cores which of course, exactly processes two physical threads.

I mentioned that the FlexFPU was a 2 part FPU/FMAC shared between/across the module.

My mention of the confusion part comes from what happens if one core has all the FMAC tied up, it means that no matter how meagre the FP workload it cannot be done until the FMAC is free again to split or be used as a whole for a new workload.