Anyone remembered FX-4200?

[Vya Domus]

Look at that photo

There is no way for that to be correct, you can't have a module without it's L2 cache, look at diagrams of the uarch.

 

[Sabotaged_Enigma]

Okay, I've gone collecting some stats.

Here's what I've got:
According to Zambezi block diagram and those stats, relations are: 1 core / 1 thread / 16 KB L1 data cache, 1 module / 64 KB L1 instruction cache / 2 MB L2 cache.

So I guess that fella set it to 2M/4C rather than 4M/4C, and our guess on FX-4200 core config still stands. If FX-4200 is really this way, it not only has 8 MB of L2 cache but also 320 KB of L1 cache, which is truly interesting!

It could probably be explained if that user just misunderstood what he was doing and went the disable 2M route instead of disabling CMT. Hence 2x2MB.

Yeah I think so

I'm intrigued...
I have a friend w/ AM3+ kit, are these available on fleabay, affordably?
edit: nvm, 'bout ~$25-30+.

idk man isn't that too expensive for an outdated Bulldozer chip?

I have one, it's indeed 4M/4C. 8MB L3

MachineLearning`s CPU Frequency score: 5218.33 MHz with a FX-4200

The FX-4200 @ 5218.3MHzscores getScoreFormatted in the CPU Frequency benchmark. MachineLearningranks #null worldwide and #5 in the hardware class. Find out more at HWBOT.

hwbot.org

Thank you!

The only weird one is splave's run...2C/2T but only 1x2MB L2.

2C in 1 module does have 1 x 2 MB I think it's correct... according to my table

yup i had ole bulldozer! wiith an hd 4870 sapphire vapor x

nice bro, Vapor X was ace back then

 

[Zyll Goliat]

Yes L3 cache is all the same 8 MB, but I don't know what's going on with the L2.

Here's what I've found on the web
This guy did side-by-side comparison of FX-4100 and FX-4200 at the same frequency. And it showed FX-4200 has double the performance of FX-4100.
View attachment 324998View attachment 324993
View attachment 324997View attachment 324994

One snapshot on CPU-world website, indicating 4 x 2 MB L2 cache instead of 2x2.
View attachment 324999

I wonder what's going on here.

Yeah that's strange you are correct then it's most likely that FX 4200 have 4x2 MB L2 Cache.....

 

[Sabotaged_Enigma]

I'm still open to any more evidences and ideas.

 

[ShrimpBrime]

Cpuworld link in the original post was on track.

https://www.cpu-world.com/CPUs/Bulldozer/AMD-FX-Series%20FX-4200%20(FD4200FRW4KGU).html#google_vignette

And @Bones may have contributed with this one.

https://www.cpu-world.com/cgi-bin/CPUID.pl?CPUID=36474

 

[Bones]

Cpuworld link in the original post was on track.

https://www.cpu-world.com/CPUs/Bulldozer/AMD-FX-Series%20FX-4200%20(FD4200FRW4KGU).html#google_vignette

And @Bones may have contributed with this one.

https://www.cpu-world.com/cgi-bin/CPUID.pl?CPUID=36474

Yep - That's the one I contributed.
Until I did that there was no reference for the chip so I did it.

To clarify, it's a Zambezi chip that has the full useable L3 cache an 8 series chip would have in use, it's only one core per module (4 total) and that's what makes it different than the others.
It still suffers from some deficiencies that a Zambezi has such as media (Videos) sometimes not working correctly (Porky Pig effect/stuttering) but Vishera fixed that at least when it came out.

 

[LabRat 891]

I verified that I was indeed freebie'd a 990FX board (that at least POSTs). Now, I'm trying to track down one of these 4200s for under $20.
As Bones went over, (IMO) this is quite the 'curio' for AMD FX's era. (Kinda like Ryzen 3300x)

 

[ShrimpBrime]

I verified that I was indeed freebie'd a 990FX board (that at least POSTs). Now, I'm trying to track down one of these 4200s for under $20.
As Bones went over, (IMO) this is quite the 'curio' for AMD FX's era. (Kinda like Ryzen 3300x)

Almost as interesting as Athlon 5000 that unlocked to a Phenom-FX??

 

[LabRat 891]

Almost as interesting as Athlon 5000 that unlocked to a Phenom-FX??

View attachment 325421

Deneb core, PhII uArch.
I have and have-had fun with those; never owned anything Bulldozer/Piledriver (Socketed, on desktop)
But, that's still awesome! (I DD'd an X3 720BE quad-unlocked, prior to my 3770k.)

IIRC, AMD had radically different naming planned-out for both end-run Phenom II and FX. I think there were plans for Phenom-branded Bulldozer chips, too.

 

[ShrimpBrime]

Deneb core, PhII uArch.
I have and have-had fun with those; never owned anything Bulldozer/Piledriver (Socketed, on desktop)
But, that's still awesome! (I DD'd an X3 720BE quad-unlocked, prior to my 3770k.)

IIRC, AMD had radically different naming planned-out for both end-run Phenom II and FX. I think there were plans for Phenom-branded Bulldozer chips, too.

Let it be known that Phenom was originally going to have the FX naming, but they pushed it back to be bulldozered.

If you didn't LLC unlock it, that would be an Athlon 5000+ with a particular partnumber I can't remember at this time.

 

[LabRat 891]

Let it be known that Phenom was originally going to have the FX naming, but they pushed it back to be bulldozered.

If you didn't LLC unlock it, that would be an Athlon 5000+ with a particular partnumber I can't remember at this time.

https://www.cpu-world.com/CPUs/K10/AMD-Athlon%20X2%205000+%20-%20AD5000ODJ22GI%20(AD5000ODGIBOX).html

I may have to keep an eye out for one, in the future.

 

[tabascosauz]

2C in 1 module does have 1 x 2 MB I think it's correct... according to my table

1 x 2MB doesn't seem right.........if FX4200 has 4 modules, with 2MB per module but only 1 core per module, how can you get the cores from 2 modules but the L2 of 1? Unless somehow you halve each module's L2 cluster, but that's not how the FX4200 is by default

Maybe just old CPU-Z derping around

 

[ShrimpBrime]

1 x 2MB doesn't seem right.........if FX4200 has 4 modules, with 2MB per module but only 1 core per module, how can you get the cores from 2 modules but the L2 of 1? Unless somehow you halve each module's L2 cluster, but that's not how the FX4200 is by default

Maybe just old CPU-Z derping around

8 cores is quad modules, with each core getting its own L1 and L2 cache and doesn't need to share.

Each module is an FPU with 2 integer cores.

Because each core doesn't have its own FPU, they couldn't call it an 8 core chip. Even though each integer core doesn't even need an FPU to operate.

Each FX processor I've delidded, are all made from 8 core dies (quad module). All of em. Quad and hexa cores are just chopped up 8 cores. I've delidded maybe a dozen of them.

 

[tabascosauz]

8 cores is quad modules, with each core getting its own L1 and L2 cache and doesn't need to share.

Each module is an FPU with 2 integer cores.

Because each core doesn't have its own FPU, they couldn't call it an 8 core chip. Even though each integer core doesn't even need an FPU to operate.

Each FX processor I've delidded, are all made from 8 core dies (quad module). All of em. Quad and hexa cores are just chopped up 8 cores. I've delidded maybe a dozen of them.

Yes, I'm aware of how Zambezi is laid out and that it's all the same die. I was referring to this HWBot result here:

Splave`s SuperPi - 32M score: 15 min 16 sec 953 ms with a FX-4200

The FX-4200 @ 5002.5MHzscores getScoreFormatted in the SuperPi - 32M benchmark. Splaveranks #null worldwide and #2 in the hardware class. Find out more at HWBOT.

hwbot.org

As there is functionally no CMT sharing shenanigans in FX-4200, each core/module gets its own 2MB L2......but the screenshot shows 2MB for 2 cores/modules. Whereas the full chip reports as 4x2MB and one would expect two cores/two modules to be 2x2MB.

 

[ShrimpBrime]

Yes, I'm aware of how Zambezi is laid out and that it's all the same die. I was referring to this HWBot result here:

Splave`s SuperPi - 32M score: 15 min 16 sec 953 ms with a FX-4200

The FX-4200 @ 5002.5MHzscores getScoreFormatted in the SuperPi - 32M benchmark. Splaveranks #null worldwide and #2 in the hardware class. Find out more at HWBOT.

hwbot.org

As there is functionally no CMT sharing shenanigans in FX-4200, each core/module gets its own 2MB L2......but the screenshot shows 2MB for 2 cores/modules. Whereas the full chip reports as 4x2MB and one would expect two cores/two modules to be 2x2MB.

Probably, and just a wild guess here, CPU-Z doesn't report correctly because it may have never been added to the roster.

 

[LabRat 891]

Yes, I'm aware of how Zambezi is laid out and that it's all the same die. I was referring to this HWBot result here:

Splave`s SuperPi - 32M score: 15 min 16 sec 953 ms with a FX-4200

The FX-4200 @ 5002.5MHzscores getScoreFormatted in the SuperPi - 32M benchmark. Splaveranks #null worldwide and #2 in the hardware class. Find out more at HWBOT.

hwbot.org

As there is functionally no CMT sharing shenanigans in FX-4200, each core/module gets its own 2MB L2......but the screenshot shows 2MB for 2 cores/modules. Whereas the full chip reports as 4x2MB and one would expect two cores/two modules to be 2x2MB.

That particular HWbot run appears to be a core-reduced OC-run.
edit 2: it also appears to be a "single module, dual-core". IIRC, there was some configurability on the core-layout via firmware.
Agreed that it's 'odd'; I wouldn't have expected a 4xmodule 4xcore 1:1 module:core 'stock' CPU to 'unlock+lockdown' to a 1x module 2x core CPU...

• Model: AMD FX-4200 'Zambezi'
• Cooling: Single Stage Phase Change
• Cores: 5,002.49 MHz (+51.59%)
• 

edit:
This User's PCMark: Vantage run, is on air, and CPU-z is reporting 'correctly', w/ an older ver.:

zx896.ru`s PCMark Vantage score: 15654 marks with a FX-4200

The FX-4200 @ 4855.7MHzscores getScoreFormatted in the PCMark Vantage benchmark. zx896.ruranks #454 worldwide and #1 in the hardware class. Find out more at HWBOT.

hwbot.org

 

[Sabotaged_Enigma]

1 x 2MB doesn't seem right.........if FX4200 has 4 modules, with 2MB per module but only 1 core per module, how can you get the cores from 2 modules but the L2 of 1? Unless somehow you halve each module's L2 cluster, but that's not how the FX4200 is by default
Maybe just old CPU-Z derping around

Yes this is what I meant. I thought that was an error.

 

[LabRat 891]

Yes this is what I meant. I thought that was an error.

As far as I can tell, it's not an error.
It's a supported and utilized feature for xOC runs.



I wouldn't have expected a 4xmodule 4xcore 1:1 module:core 'stock' CPU to 'unlock+lockdown' to a 1x module 2x core CPU...

So, there may still be a slight error w/ the older CPU-z ver,
or that's just how his Sabertooth configured it, when he disabled 2 cores.

 

[Sabotaged_Enigma]

As far as I can tell, it's not an error.
It's a supported and utilized feature for xOC runs.
View attachment 325456

I knew it was only running two cores probably. The problem we've been confused about is that if FX-4200 has only two cores running, it should be two cores in two separate modules which should be 2 x 2 MB of L2 cache. FX-4200 has four cores in four modules from the start, and I don't think it's possible to unlock the disabled second int unit in any module by a single mainboard setting to make it 1M/2C/2T. It's different from ACC function unlocking disabled K10.5 cores, isn't it?

 

[MachineLearning]

Yes, I'm aware of how Zambezi is laid out and that it's all the same die. I was referring to this HWBot result here:

Splave`s SuperPi - 32M score: 15 min 16 sec 953 ms with a FX-4200

The FX-4200 @ 5002.5MHzscores getScoreFormatted in the SuperPi - 32M benchmark. Splaveranks #null worldwide and #2 in the hardware class. Find out more at HWBOT.

hwbot.org

As there is functionally no CMT sharing shenanigans in FX-4200, each core/module gets its own 2MB L2......but the screenshot shows 2MB for 2 cores/modules. Whereas the full chip reports as 4x2MB and one would expect two cores/two modules to be 2x2MB.

I would assume it's a bug with that CPU-Z version. Probably has been fixed since 1.76

 

[LabRat 891]

I knew it was only running two cores probably. The problem we've been confused about is that if FX-4200 has only two cores running, it should be two cores in two separate modules which should be 2 x 2 MB of L2 cache. FX-4200 has four cores in four modules from the start, and I don't think it's possible to unlock the disabled second int unit in any module by a single mainboard setting to make it 1M/2C/2T. It's different from ACC function unlocking disabled K10.5 cores, isn't it?

Correct. I didn't mean to take a step backwards, from the what already had been figured out. my bad.
that, I don't know.

I would assume it's a bug with that CPU-Z version. Probably has been fixed since 1.76

That would be my (first) assumption, also. (I was just a 'lil slow on the uptake )
But I suppose it's not impossible the Sabretooth is responsible.

 

[fusseli]

fx4200 ought to half of the turd that was fx8350, which was the last AMD cpu I owned. It was slow. Embarrassingly slow even compared to my friend's older at the time e8500 core2duo that still beat the ipc years later. Years later AMD paid me $26 in a class action lawsuit over the cpu architecture being not a true 8 core / 8 thread architecture no regertz going intel after that bs. Oh yeah and they made the fx9390 or whatever that was just a binned fx8350 and not even a new architecture or series. Jesus.

 

[ShrimpBrime]

fx4200 ought to half of the turd that was fx8350, which was the last AMD cpu I owned. It was slow. Embarrassingly slow even compared to my friend's older at the time e8500 core2duo that still beat the ipc years later. Years later AMD paid me $26 in a class action lawsuit over the cpu architecture being not a true 8 core / 8 thread architecture no regertz going intel after that bs. Oh yeah and they made the fx9390 or whatever that was just a binned fx8350 and not even a new architecture or series. Jesus.

Only slow on a general concensus.
They where much faster unzipping

 

[LabRat 891]

Only slow on a general concensus.
They where much faster unzipping

Also, since Intel hopped-aboard the many-core train, and Msft re-did the scheduler...
they've gotten faster, in the same uses, on newer Windows. It's, kinda crazy.

They're no competitor to even a Ryzen 1st gen (IMHO), but they're interesting...

Years later AMD paid me $26 in a class action lawsuit over the cpu architecture being not a true 8 core / 8 thread architecture

Also interesting, because of stuff like this.



Had the FX-4200 been better-marketed (and was a cheap AF platform), I'd have considered it over the 3570k+Z77 I started-out with (that era).
which, it wouldn't have been
since, it shows the 2 "core" module design/strategy was not ready for prime time.

 

[tabascosauz]

Had the FX-4200 been better-marketed (and was a cheap AF platform), I'd have considered it over the 3570k+Z77 I started-out with (that era).
which, it wouldn't have been
since, it shows the 2 "core" module design/strategy was not ready for prime time.

Their hands were probably tied, though, since making FX-4200 essentially would have admitted that Bulldozer was a failure. It's not CMT at all anymore. There were people who did the whole make-your-own-4M/4T thing with FX-8150 but pretty much all decided it wasn't actually worth doing even for gaming, once Vishera and and scheduler fixes came out.

Also, having zero viable higher end mini-ITX options for years...........those were some dark times.