News AMD K6 and FX-8350 Re-Visited Against Modern CPUs: Ryzen up to 910X Faster

[Admin]

A pair of tech reviewers recently re-visited some of AMD's older processors, including the AMD FX-8350 and AMD K6 from 10 and 26 years ago, to see how they perform compared to modern CPUs and games.

AMD K6 and FX-8350 Re-Visited Against Modern CPUs: Ryzen up to 910X Faster : Read more

 

[bdcrlsn]

IMO, the AMD Bulldozer/Piledriver processors were good ideas on paper. They could hold their own in multi-threaded programs, but these processors were not designed for single-threaded ones and the era these processors were in was decidedly lacking multi-thread optimized applications.

 

[Metal Messiah.]

Ah, the good old FX bulldozer debacle I still remember. I have owned an FX 6350 CPU before.

AMD created confusion as they advertised FX-8350 as having 8 cores, and the claim that a shared FPU unit within a 'dual core' module does not constitute an actual core of performance similar to a separate core/FPU unit.

AMD based the core count on the number of integer cores, and pitched its Bulldozer processors as the first 8-core desktop chips. This was a bad move actually, but I liked the bulldozer's architecture/design at that time. The Legacy still lives on !

You know OLD is GOLD after all.

Actually, technically it is impossible for an eight-core Bulldozer-powered processor to truly execute eight instructions simultaneously – it cannot run eight complex math calculations at any one moment due to the shared FPU design.

On the other hand, Intel cores, for what it's worth, had their own separate floating-point math units.

 

[King_V]

Ah, this reminds me . . I've still got Old Ironsides sitting in a room now, with its FX-6300. Not as much grunt as the 8350, but still. Just sitting. That said, I fired it up a short while back, to install the fanless Sapphire HD 6670 card, which stays nice and cool due to the fan in the side of the case blowing right on it.

And, I really, REALLY need to fire up my old PC with the AMD K6-2+ 400Mhz CPU. It ran the last time I booted it, but, uh, that was a while ago. 192MB of RAM, if I recall correctly. Odd sort of configuration, but I remember that it was all SDRAM sticks with the same CL or same overall timings.

Quite a long while ago. I wonder if it'll still work...

 

[SunMaster]

How the 8350 performs probably illustrates how ridiculously overpowered todays CPUs are for gaming, At least the current generation of games.

 

[jeremyj_83]

What is impressive against the K6-500 is that the 7950X's clock speed in ST applications is only 11.5x higher. That means another 35x of performance gains are all from uArch developments over 26 years.

 

[jeremyj_83]

How the 8350 performs probably illustrates how ridiculously overpowered todays CPUs are for gaming, At least the current generation of games.

Or it shows that games today are better able to take advantage of the FX's strange uArch design.

 

[gfg]

Error in the publication: it was not a first generation AMD "K6" as indicated here, it was a "K6-2 500".
233 Mhz vs 500 Mhz makes the 46 times make more sense. Since it marks a IPC of 400% and not 200% if it were a first generation K6.
In any case, 4 times more IPC seems little to me, it is obviously not optimized software.
Ryzen 7000 vs the Athlon FX 8350 there is already an increase of ~2.6 times, against a K6 it should be around x10.

 

[hotaru.hino]

How the 8350 performs probably illustrates how ridiculously overpowered todays CPUs are for gaming, At least the current generation of games.

A lot of it I would argue has to do with how games over the past decade finally started spawning more threads to do work. They may not do a whole lot of work on their own (there's still usually 2-3 threads that are really active), but it's certainly better than having that all done on a single thread.

What is impressive against the K6-500 is that the 7950X's clock speed in ST applications is only 11.5x higher. That means another 35x of performance gains are all from uArch developments over 26 years.

I'm more curious about what sort of instructions were actually ran on the LFK benchmark to characterize where might these improvements come from. For instance, if they ran a version of it that took advantage of SIMD extensions, then that shows the strength of the uArch, but it's not necessarily an apples to apples comparison since the K-6 wouldn't even have SSE.

 

[SunMaster]

A lot of it I would argue has to do with how games over the past decade finally started spawning more threads to do work. They may not do a whole lot of work on their own (there's still usually 2-3 threads that are really active), but it's certainly better than having that all done on a single thread.

Agreed. And eventually I think the incredible focus on single thread performance will if not "go away" so at least be far less meaningful. I'm not talking about tomorrow

 

[gfg]

What is impressive against the K6-500 is that the 7950X's clock speed in ST applications is only 11.5x higher. That means another 35x of performance gains are all from uArch developments over 26 years.

11.5x ???
35x ???

7950X / K6-2 500 = (1294 pt / 5700mhz) / (28 pt / 500 mhz)
7950X / K6-2 500 = (227 pt x Ghz ) / (56 pt x Ghz)
7950X / K6-2 500 = 4,05x IPC or +305% IPC

this benchmark does not do justice to the instruction set on a 64-bit Ryzen 7000

 

[drtweak]

still have my 8320. Runs as my work bench PC now after upgrading 2 years ago. Only reason why I upgrade is 3 of the VRMs blew up on me one day. Motherboard still works, CPU Still works. I no longer have an overclock on it and turned off turbo but it still runs perfectly fine. I did notice that once i took off the 4.4GHz over clock, that is had for almost 7 years, there was a big performance difference to the stock 3.5.

 

[Deleted member 2838871]

Interesting.

My first and till just recently only AMD processor was the Athlon XP 1800+ (1.53ghz) in 2001... paired with a GeForce 3 Ti 200 and Windows XP.

Then Intel for 20 years before my recent 7950x3D build paired with a RTX 4090 and Windows 11 Pro.

1000x the upgrade?

 

[zipspyder]

Looking at his video quickly it turns out that he over clocked the crap out of the 8350 and used those benchmark numbers. Like the 8350 all cores at 4.80Ghz @ 1.51v and 2000MHz memory. Yeah that is not going to be stable or sustainable long term. Nor is it indicative of a normal default AMD system with even moderate OC of the time. Interesting video just a little misleading.

 

[SonoraTechnical]

I remember getting a hold of an AMD-K6-III+ .... Ran it at 6 * 103 for 618Mhz on the then AMD MVP3+ chipset boards... That chipset allowed upto 112Mhz... but 618Mhz was faster than 560Mhz. I recall either a half multiplier wasn't supported for 5.5*112 = 616... or it just wasn't stable... But 618Mhz was fine and rock solid.

I got a hold of some K6-II+ chips and put them in really cheap Toshiba laptops where I could run them at 5*95 for 570Mhz, but that was the quickets I could do.. Turned those useless K6-400 laptops into something useful at that speed with the extra L2.
Anyway, I had allot of fun for a few years with those K6, K6-II+, K6-III, and K6-III+ cpus on the mVP3+ mATX Super7 boards back in the day...

Would be curious to see how that chip would do in this comparison with the built in 256KB of L2. (The K6-2+ had 128Kb, The K6 had no onchip L2 cache).

 

[SunMaster]

11.5x ???
35x ???

7950X / K6-2 500 = (1294 pt / 5700mhz) / (28 pt / 500 mhz)
7950X / K6-2 500 = (227 pt x Ghz ) / (56 pt x Ghz)
7950X / K6-2 500 = 4,05x IPC or +305% IPC

this benchmark does not do justice to the instruction set on a 64-bit Ryzen 7000

I think your calculation is correct. It has a factor of 4,05 IPC, so a net gain of 305%,

 

[hotaru.hino]

I think your calculation is correct. It has a factor of 4,05 IPC, so a net gain of 305%,

I think they're pointing out the quoted post had some really odd math that actually doesn't make sense.

With regard to the LFK benchmark, the Ryzen got an ST score of 1294, while the K-6 go 28. This is about 46x better performance. They then took the clock speed (5700Mhz vs 500MHz) and got ~11 out of that. Then figured they could subtract that from 46.

 

[evdjj3j]

"The K6 was originally launched in April of 1997 as the successor to the K-III"

K5 not K-III.

 

[SunMaster]

I think they're pointing out the quoted post had some really odd math that actually doesn't make sense.

With regard to the LFK benchmark, the Ryzen got an ST score of 1294, while the K-6 go 28. This is about 46x better performance. They then took the clock speed (5700Mhz vs 500MHz) and got ~11 out of that. Then figured they could subtract that from 46.

Not subtract.

The single core on the ryzen had a 46,2 times higher score than the K6-2 (1294 / 28)
Comparing the clock speed difference, the ryzen core was 11,4 times higher (5700 / 500)'

So dividing 46,2 on 11,4 you get around 4,05, which means 305% faster. Faster in terms of IPC.

 

[hotaru.hino]

Not subtract.

The single core on the ryzen had a 46,2 times higher score than the K6-2 (1294 / 28)
Comparing the clock speed difference, the ryzen core was 11,4 times higher (5700 / 500)'

So dividing 46,2 on 11,4 you get around 4,05, which means 305% faster. Faster in terms of IPC.

But that doesn't get you 35x, which would be 3500%.

In any case, they're just pointing out the weird math that again, doesn't make sense.

 

[SunMaster]

But that doesn't get you 35x, which would be 3500%.

In any case, they're just pointing out the weird math that again, doesn't make sense.

All I did was post I agreed with his way of calculating the IPC gain. I too think it is 305%, as per the scores and frequencies.

 

[Zerk2012]

Never bought a FX my 2500K clocked at 5.0 spanked it anyhow paired with SLI 6970's back then.

 

[LazyTirade]

LFK is a floating unit benchmark. The K6-2 has 1 FPU unit while Zen 4 has 4 (per core, 2 add and 2 mul/acc), so theoretically the single core improvement should be <= 4 * 5.85 Ghz / 500 MHz = 46.8x, which lines up pretty well with the results. It's a testament to how efficient a modern CPU frontend is, it's almost fully utilizing the 4 units, despite the fact that it's running single threaded code from the 1980's and 2 of the units are add-only.

On the other hand the K6-2 has 2 "extra" FPUs in the form of 3DNow units, and I suspect the compiler did not target 3DNow instructions. In theory, with the right optimizations, the K6-2 could achieve 50-75% of the IPC of a single Zen 4 core.

TLDR, there's been far less CPU architectural improvement than you would have guessed, if you take out the gains from scaling frequency and execution width, i.e. gains from node improvements.

 

[abufrejoval]

I don't know any more how many iterations of K6-II and K6-III+ I went through: I must have owned pretty much every variant at one point in time.

Most of it was the easy upgrades, all you had to do was to swap the CPU to get so much more power!

Has me wonder now, where the older CPUs went: I didn't have kids or in-laws at the time to pass them on to and I'm not into selling things that I no longer consider good enough...

The main machine was a luggable box with an AT-sized mainboard at the time, 192MB RAM I think it had, too, and an assortment of EISA GPUs, including a TI TMS 34020 TIGA card I had ported X11R4 to, if I remember correctly.

It also ran VMware on top of NT 3.51 or 4.0 and a Linux on that.

I didn't touch Intels for along time after a Pentium Overdrive (had started with an 80286), I think the Q6600 eventually won me over again, which competed against a Phenom II x6 next to it, probably the better machine to test gaming on today.

Both are still around in storage somewhere, most likely still operable, at least until Windows 10 is finally turned off.

I guess the main reason Windows 11 is killing all these older machines is that they are really good enough for near everything 1920x1080 as long as the GPU isn't a total drag. Some of my kids are still running Ivy Bridge i7 with GTX 980ti and not only are they not complaining, I regularly don't stand a chance against them or with them: they need to protect me like an elderly citizen, when we're out on the Killing Floor or similar.

Part of it is my 4k screen. Part of it is simply that I started gaming on an Apple ][... when it was new.

 

[SkyBill40]

My first full build was a K8 series Thunderbird. That thing was awesome for that time. Following that, I went to an Athlon 64 4000+ San Diego core which was excellent and blew Intel out of the water. Next came the FX8350 which plowed on like a champ (and still does/will if I ever build a HTPC setup) until I built my 5800X based rig.

For all the talk about how terrible the FX series was, I loved my 8350 and it aged supremely well. Yeah, it wasn't a true 8 core as AMD had advertised but despite that it more than held it's own against other processors. Seeing how it still pulls decent frames as referenced in the story is a testament to the longevity of that CPU.