Personal computing discussed

Moderators: Flying Fox, morphine

 
setaG_lliB
Gerbil First Class
Topic Author
Posts: 122
Joined: Wed Mar 03, 2010 6:02 pm

Let's party like it's 1999 and look at single-threaded x87 FPU performance

Wed Nov 01, 2017 4:03 pm

Over the course of the last several months I've been casually benchmarking my old and new computers with SuperPi. What makes SuperPi so great as a universal benchmark is that it is completely single-threaded, and doesn't make use of SIMD. It just tests the raw, x87 (FPU) performance of a CPU. Kind of like software PhysX. :lol: The latest whizbang Core i7 will not have any advantage over a 486DX in this benchmark, other than having a faster FPU.

I ran most of the processors at 2.4GHz (sometimes through overclocking or underclocking), to get a true sense of how each microarchitecture's FPU performed clock-for-clock. On the K6-2 and faster machines, I ran SuperPi 1.9 under Windows XP SP3 or Win7 SP1. The 486 and Pentium ran SuperPi 1.1 under Windows 95. Every machine ran the 1M test.

Am486 DX2, 80MHz - 32MB 70ns FPM DRAM - 40MHz FSB - 8KB L1 + 128KB external L2: 1h 51m 32s
Pentium, 100MHz - 64MB 60ns EDO DRAM - 66MHz FSB - 16K L1: 30m 16.710s
AMD K6-2, 500MHz: - 256MB CL2-2-2-6 PC100 SDRAM - 100MHz FSB - 64KB L1 + 1MB external L2: 8m 34.526s

Pentium III (Tualatin) at 1575MHz - 2GB CL2-2-2-5 300MHz single channel DDR - 150MHz FSB - 512KB L2: 1m 26.311s
Pentium 4 (Northwood) at 2400MHz - 2GB CL2-2-2-5 333MHz single channel DDR - 400MHz FSB - 512KB L2: 1m 14.547s
Athlon XP (Barton) at 2400MHz - 2GB CL3-3-3-8 436MHz single channel DDR - 436MHz FSB - 512KB L2: 54.482s
Atom X7 Z8700 (Cherry Trail) at 2400MHz - 4GB CL9-9-9-30 1600MHz dual channel DDR3 - 2MB L2: 37.125s
Athlon 64 (San Diego) at 2400MHz - 2GB CL2-2-2-6 400MHz dual channel DDR - 1MB L2: 33.814s
Pentium M (Dothan) at 2400MHz - 2GB CL3-3-3-8 533MHz dual channel DDR2 - 533MHz FSB - 2MB L2: 32.843s
Phenom II (Thuban) at 2400MHz - 16GB CL9-9-9-24 1600MHz dual channel DDR3 - 512KB L2+6MB L3: 24.722s
Core 2 Duo (Wolfdale-6M) at 2400MHz - 8GB CL6-6-6-18 1066MHz dual channel DDR3 - 1066MHz FSB - 6MB L2: 18.814s
Core i5 (Haswell-6M) at 2400MHz - 8GB CL11-11-11-31 1600MHz single channel DDR3 - 256K L2+6MB L3: 14.506s

Core i7 (Ivy Bridge-E) at 4600MHz - 32GB CL10-11-11-30 2400MHz quad channel DDR3 - 256K L2+12MB L3: 7.781s

Some interesting takeaways:
-Look at how much faster that P100 is than the 486-80!
-Without SSE kicking its sorry butt up the hill, Pentium 4 is quite a poor performer. I didn't include it here, but the Celeron (Northwood-128) version of this chip is slower than the PIII.
-I always thought that x87 FP was largely unchanged going from K7 (Athlon XP) to K8 (A64), but that doesn't appear to be the case. K8 is far, far ahead.
-The difference between Pentium M and Core 2 is startling, with Core 2 being something like 75% faster at equal clocks. However, Core 2 appears to be the last big leap, as there is relatively little difference between Wolfdale and Haswell. This isn't too surprising, as x87 FP was most likely an afterthrought by the time Nehalem launched.
Last edited by setaG_lliB on Fri Nov 03, 2017 3:20 am, edited 3 times in total.
 
Redocbew
Gold subscriber
Gerbil Jedi
Posts: 1665
Joined: Sat Mar 15, 2014 11:44 am

Re: Let's party like it's 1999 and look at single-threaded x87 FPU performance

Wed Nov 01, 2017 4:10 pm

I remember reading that about the K8. Despite the block diagram being pretty much the same AMD said there had been a lot of work done on the FPU, but didn't really go into detail about it. I guess they weren't kidding.

I also remember x87 assembly nearly stealing my will to live when I had to learn it in college. That was around the time when I decided that maybe low level languages weren't going to be my thing. :P
Do not meddle in the affairs of archers, for they are subtle and you won't hear them coming.
 
chuckula
Gold subscriber
Gerbil Jedi
Posts: 1862
Joined: Wed Jan 23, 2008 9:18 pm
Location: Probably where I don't belong.

Re: Let's party like it's 1999 and look at single-threaded x87 FPU performance

Wed Nov 01, 2017 4:19 pm

I'm surprised that there was as big a difference at there was between Core 2 and Haswell when clockspeeds were held constant. The x87 FPU is pretty ugly internally and there aren't really many if any architectural tricks left that will improve its performance while also being considered canonically correct implementations that actually jump through all the flaming hoops required to be considered backwards-compatible with all the earlier x87 implementations.
4770K @ 4.7 GHz; 32GB DDR3-2133; GTX-1080 sold and back to hipster IGP!; 512GB 840 Pro (2x); Fractal Define XL-R2; NZXT Kraken-X60
--Many thanks to the TR Forum for advice in getting it built.
 
DrDominodog51
Gerbil Elite
Posts: 696
Joined: Sat Jan 17, 2015 12:23 pm
Location: Silicon Valley

Re: Let's party like it's 1999 and look at single-threaded x87 FPU performance

Wed Nov 01, 2017 4:38 pm

What memory speed and timings were used for each CPU?

IME the 32 million size can be extremely sensitive to memory frequency and latency. I'm not sure how much the 1 million size changes this.
"Look real close cause strobe lights lie" -Anonymous
 
Duct Tape Dude
Gerbil Elite
Posts: 699
Joined: Thu May 02, 2013 12:37 pm

Re: Let's party like it's 1999 and look at single-threaded x87 FPU performance

Wed Nov 01, 2017 4:41 pm

I can't believe Core 2 is a decade old already. Gone are the days of huge performance increases every year, and yet I still find a way (or maybe an excuse?) to buy a new system periodically.
 
just brew it!
Gold subscriber
Administrator
Posts: 51530
Joined: Tue Aug 20, 2002 10:51 pm
Location: Somewhere, having a beer

Re: Let's party like it's 1999 and look at single-threaded x87 FPU performance

Wed Nov 01, 2017 5:00 pm

Redocbew wrote:
I also remember x87 assembly nearly stealing my will to live when I had to learn it in college. That was around the time when I decided that maybe low level languages weren't going to be my thing. :P

X87 FPU is really a bizarre architecture. It isn't easy for compilers to generate optimal code for, or for the CPU to optimize internally either, and is completely unsuitable for vectorizing repetitive calculations. That's why it's been deprecated in favor of SSE and AVX.
Nostalgia isn't what it used to be.
 
Pancake
Gerbil First Class
Posts: 156
Joined: Mon Sep 19, 2011 2:04 am

Re: Let's party like it's 1999 and look at single-threaded x87 FPU performance

Wed Nov 01, 2017 8:35 pm

just brew it! wrote:
Redocbew wrote:
I also remember x87 assembly nearly stealing my will to live when I had to learn it in college. That was around the time when I decided that maybe low level languages weren't going to be my thing. :P

X87 FPU is really a bizarre architecture. It isn't easy for compilers to generate optimal code for, or for the CPU to optimize internally either, and is completely unsuitable for vectorizing repetitive calculations. That's why it's been deprecated in favor of SSE and AVX.


Dat 80-bit register file though... nom nom nom

The 8087 did become the universal IEEE-754 standard for floating point format used in just about everything today. As an Intel-hater I was never happy with that.
 
just brew it!
Gold subscriber
Administrator
Posts: 51530
Joined: Tue Aug 20, 2002 10:51 pm
Location: Somewhere, having a beer

Re: Let's party like it's 1999 and look at single-threaded x87 FPU performance

Wed Nov 01, 2017 8:47 pm

Pancake wrote:
just brew it! wrote:
Redocbew wrote:
I also remember x87 assembly nearly stealing my will to live when I had to learn it in college. That was around the time when I decided that maybe low level languages weren't going to be my thing. :P

X87 FPU is really a bizarre architecture. It isn't easy for compilers to generate optimal code for, or for the CPU to optimize internally either, and is completely unsuitable for vectorizing repetitive calculations. That's why it's been deprecated in favor of SSE and AVX.

Dat 80-bit register file though... nom nom nom

The 8087 did become the universal IEEE-754 standard for floating point format used in just about everything today. As an Intel-hater I was never happy with that.

At least the IEEE-754 data format is reasonable (unlike the architecture of the x87 FPU itself). So the way I look at it is, they got it half right, and we've kept the good half.
Nostalgia isn't what it used to be.
 
meerkt
Gerbil Elite
Posts: 969
Joined: Sun Aug 25, 2013 2:55 am

Re: Let's party like it's 1999 and look at single-threaded x87 FPU performance

Wed Nov 01, 2017 10:15 pm

setaG_lliB wrote:
Look at how much faster that P100 is than the 486-80!
It's also AMD vs Intel.
 
Mr Bill
Gold subscriber
Gerbil Jedi
Posts: 1557
Joined: Mon Jan 21, 2002 7:00 pm
Location: Colorado Western Slope
Contact:

Re: Let's party like it's 1999 and look at single-threaded x87 FPU performance

Thu Nov 02, 2017 12:41 am

I know you didn't forget but there was the 8MHz 8087 coprocessor that went with the original 8086 systems circa 1983. I still have that system. Have no idea if it would boot.
X6 1100T BE | Gigabyte GA-990FXA-UD3 AM3+ | XFX HD 7870 | 16 GB DDR3 | Samsung 830/850 Pro SSD's | Logitech cherry MX-brown G710+ | Logitech G303 Daedalus Apex mouse | SeaSonic SS-660XP 80+ Pt | BenQ 24' 1900x1200 IPS | APC Back-UPS NS-1350 | WinXP64 Pro
 
SkyWarrior
Gerbil
Posts: 48
Joined: Wed Jun 21, 2006 10:27 pm
Location: Turkey

Re: Let's party like it's 1999 and look at single-threaded x87 FPU performance

Thu Nov 02, 2017 3:42 am

Super-PI likes low latency high amount high bandwidth L2 cache than small L2 with high Latency large L3 or small bandwidth L2 with large L3.

I wonder how that works under Skylake-X vs Skylake or Broadwell-EP.

I remember I was getting the same score with a FX57 clocked at 3.0ghz with 2 super-pi's running at the same time against Pentium 4 LGA 775 2M cache 3.0 ghz HT enabled running at 3.0GHZ with 2 super-pi's running at the same time.

Pentium 4 sucked at single but when you have 2 of them running at the same time even FX57 was suffering.
Rare is common where I work...
 
whm1974
Gerbil Elder
Posts: 5761
Joined: Fri Dec 05, 2014 5:29 am

Re: Let's party like it's 1999 and look at single-threaded x87 FPU performance

Thu Nov 02, 2017 5:52 am

just brew it! wrote:
Redocbew wrote:
I also remember x87 assembly nearly stealing my will to live when I had to learn it in college. That was around the time when I decided that maybe low level languages weren't going to be my thing. :P

X87 FPU is really a bizarre architecture. It isn't easy for compilers to generate optimal code for, or for the CPU to optimize internally either, and is completely unsuitable for vectorizing repetitive calculations. That's why it's been deprecated in favor of SSE and AVX.

Isn't x86/x87 one of the worst ISAs to write assembly(or any language) code for? Every time I look up the history of the IBM PC there is always more then few people who keep saying that IBM should have chosen the Motorola 68000 over the Intel 8088. Of course then again even if they did, IBM probably would have crippled the PC with the 68008 instead. :roll:
 
just brew it!
Gold subscriber
Administrator
Posts: 51530
Joined: Tue Aug 20, 2002 10:51 pm
Location: Somewhere, having a beer

Re: Let's party like it's 1999 and look at single-threaded x87 FPU performance

Thu Nov 02, 2017 5:56 am

meerkt wrote:
setaG_lliB wrote:
Look at how much faster that P100 is than the 486-80!

It's also AMD vs Intel.

Irrelevant. AMD's 486 chips had identical IPC to Intel's. They had a license to use Intel's microcode (dating back to when they were a "second source" manufacturer for Intel's designs) in those days; their 386 and 486 chips were reverse-engineered clones of Intel's designs instead of an independently developed architecture.

In fact, AMD's clones had more clock speed headroom. Imagine that! :wink:
Nostalgia isn't what it used to be.
 
the
Gerbil Elite
Posts: 924
Joined: Tue Jun 29, 2010 2:26 am

Re: Let's party like it's 1999 and look at single-threaded x87 FPU performance

Thu Nov 02, 2017 7:40 am

There is a good chance the performance jump from Pentium M Dothan to Core 2 Duo Wolfdale is the increase in L2 cache. With 6 MB on-die, that Wolfdale chip has enough to keep everything in cache.

While per core cache sizes haven't increased that much since Wolfdale when looking at the entire cache hierarchy (L1 + L2 +L3), since then Intel has worked in decreasing the cache's latency and/or increasing its bandwidth. Even without touching x87, these changes still provide a small performance boost all around.

It would be interesting to see how SkyLake-EP handles SuperPi in this test. The cache topology has again changed and I don't think the per core amount has enough to hold everything. It might actually end up worse than Haswell-EP/Broadwell-EP.
Dual Opteron 6376, 96 GB DDR3, Asus KGPE-D16, GTX 970
Mac Pro Dual Xeon E5645, 48 GB DDR3, GTX 770
Core i7 3930K@4.2 Ghz, 32 GB DDR3, GA-X79-UP5-Wifi
Core i7 2600K@4.4 Ghz, 16 GB DDR3, GTX 970, GA-X68XP-UD4
 
bthylafh
Maximum Gerbil
Posts: 4084
Joined: Mon Dec 29, 2003 11:55 pm
Location: Southwest Missouri, USA

Re: Let's party like it's 1999 and look at single-threaded x87 FPU performance

Thu Nov 02, 2017 8:18 am

meerkt wrote:
setaG_lliB wrote:
Look at how much faster that P100 is than the 486-80!
It's also AMD vs Intel.


Ignoring the infamous bug, the Pentium's FPU was well known for being much faster than the 486's.

/old
Hakkaa päälle!
i7-8700K|Asus Z-370 Pro|32GB DDR4|Asus Radeon RX-580|Samsung 960 EVO 1TB|1988 Model M||Logitech MX 518 & F310|Samsung C24FG70|Dell 2209WA|ATH-M50x
 
whm1974
Gerbil Elder
Posts: 5761
Joined: Fri Dec 05, 2014 5:29 am

Re: Let's party like it's 1999 and look at single-threaded x87 FPU performance

Thu Nov 02, 2017 9:15 am

bthylafh wrote:
meerkt wrote:
setaG_lliB wrote:
Look at how much faster that P100 is than the 486-80!
It's also AMD vs Intel.


Ignoring the infamous bug, the Pentium's FPU was well known for being much faster than the 486's.

/old

Didn't Intel fixed that bug shortly after everyone made a big stink about it?
 
srg86
Gerbil Team Leader
Posts: 245
Joined: Tue Apr 25, 2006 7:57 am
Location: Madison, WI

Re: Let's party like it's 1999 and look at single-threaded x87 FPU performance

Thu Nov 02, 2017 9:15 am

Great benchmarking!!

Redocbew wrote:
I also remember x87 assembly nearly stealing my will to live when I had to learn it in college. That was around the time when I decided that maybe low level languages weren't going to be my thing. :P


Don't worry, I love low level programming, but I've never liked x87. BTW There was one optimization that CPUs to to try and lessen the performance pain. The FCH instruction to swap the current top of stack value with another can be done kinda in parallel, to make the instruction be "free" as in takes no clock cycles.

whm1974 wrote:
Isn't x86/x87 one of the worst ISAs to write assembly(or any language) code for? Every time I look up the history of the IBM PC there is always more then few people who keep saying that IBM should have chosen the Motorola 68000 over the Intel 8088. Of course then again even if they did, IBM probably would have crippled the PC with the 68008 instead.


I think IBM had an existing contract with Intel or something.

Not to defend x86 too much, but what they are referring to was Real Mode Segmentation, where each segment could overlap and the same linear address could be addressed using multiple sets of segment:offset pairs. It means that you can't just make a pointer and write to it, you have to go through extra hoops to set it as a segment and offset. Also the instruction set of the 8088-80286 isn't very orthogonal. Protected Mode x86 programming from the 80386 is much, much nicer imho (and probably no real worse than 68k, though I've not done any 68k assembly). The main weakness is the lack of registers, corrected by AMD64.

SkyWarrior wrote:
Pentium 4 sucked at single but when you have 2 of them running at the same time even FX57 was suffering.


There's no-other way of putting it really, Netburst just sucked.

bthylafh wrote:
Ignoring the infamous bug, the Pentium's FPU was well known for being much faster than the 486's.


I'd be curious to see what his benchmark shows on a Cyrix 6x86

/Keen student of computer history.
Intel Core i7 4790K, Z97, 16GB RAM, 128GB m4 SSD, 480GB M500 SSD, 500GB WD Vel, Intel HD4600, Corsair HX650, Fedora x64.
Thinkpad T460p, Intel Core i5 6440HQ, 8GB RAM, 512GB SSD, Intel HD 530 IGP, Fedora x64, Win 10 x64.
 
Vhalidictes
Silver subscriber
Gerbil Jedi
Posts: 1776
Joined: Fri Jan 07, 2005 2:32 pm
Location: Paragon City, RI

Re: Let's party like it's 1999 and look at single-threaded x87 FPU performance

Thu Nov 02, 2017 1:00 pm

I owned a Cyrix 6x86 PR-200 (166MHz overclocked to 175MHz IIRC, talk about running hot...), and it was a beast for Integer operations but essentially a 486 DX-100 for floating point.

Acceptable for the time but not a contender against the Pentium.

On that note I miss high-IPC designs and CISC.
 
just brew it!
Gold subscriber
Administrator
Posts: 51530
Joined: Tue Aug 20, 2002 10:51 pm
Location: Somewhere, having a beer

Re: Let's party like it's 1999 and look at single-threaded x87 FPU performance

Thu Nov 02, 2017 1:07 pm

Vhalidictes wrote:
I owned a Cyrix 6x86 PR-200 (166MHz overclocked to 175MHz IIRC, talk about running hot...), and it was a beast for Integer operations but essentially a 486 DX-100 for floating point.

Acceptable for the time but not a contender against the Pentium.

On that note I miss high-IPC designs and CISC.

Not sure why you'd miss CISC. It only matters if you do a lot of assembly language coding, and almost nobody does that any more. Even if you're an assembly programmer, the current x86-64 ISA is still outwardly CISC, even if it gets translated into an internal RISC instruction stream for execution.
Nostalgia isn't what it used to be.
 
mikewinddale
Gerbil
Posts: 52
Joined: Sat Jan 07, 2017 2:22 am

Re: Let's party like it's 1999 and look at single-threaded x87 FPU performance

Thu Nov 02, 2017 4:07 pm

Intel Core i7-7500U @ 3.48 GHz: 11.000 seconds

I also have an old laptop, an AMD A10-5750M, but it has Linux Mint 18.1, and I can't get the SuperPi Linux version to run. I've tried both the executable and the command-line, and nothing happens. Just nothing. (I also tried SuperPi with Wine. No success.) If anyone can offer any suggestions, then please. But in the meantime, I only have results for the Intel, where I have Windows installed.

On Linux, I was able to get System Stability Tester (SST) to run, which claims to be a SuperPi clone. However, using SST, I got **very*** different results, to wit:

SuperPi, 1M, Intel: 11.000 seconds
SST, 1M, Intel: 2.146 seconds
SST, 1M, AMD: 4.488 seconds

So within SST, comparing Intel to AMD, the Intel is about twice as fast, which is consistent with a variety of other benchmarks I've run (e.g. Cinebench, Blenchmark, x264, x265). But between SST and SuperPi, we can see that the results are not at all comparable.

Since SST is twice as fast on Intel as AMD, we can extrapolate that if SuperPi is 11 seconds on Intel, it would be 22 seconds on AMD. But that is an extrapolation. I'd love to get SuperPi to run on the AMD-Linux.
Last edited by mikewinddale on Thu Nov 02, 2017 4:09 pm, edited 1 time in total.
 
Vhalidictes
Silver subscriber
Gerbil Jedi
Posts: 1776
Joined: Fri Jan 07, 2005 2:32 pm
Location: Paragon City, RI

Re: Let's party like it's 1999 and look at single-threaded x87 FPU performance

Thu Nov 02, 2017 4:08 pm

just brew it! wrote:
Not sure why you'd miss CISC. It only matters if you do a lot of assembly language coding, and almost nobody does that any more. Even if you're an assembly programmer, the current x86-64 ISA is still outwardly CISC, even if it gets translated into an internal RISC instruction stream for execution.


You're right - micro-ops are a symptom, not a root cause. The disease here is "speed demon" CPU design, most recently the Bulldozer and Northwood before it. Side note, there's no way that the internal translation is cost-free.
 
just brew it!
Gold subscriber
Administrator
Posts: 51530
Joined: Tue Aug 20, 2002 10:51 pm
Location: Somewhere, having a beer

Re: Let's party like it's 1999 and look at single-threaded x87 FPU performance

Thu Nov 02, 2017 5:13 pm

Vhalidictes wrote:
just brew it! wrote:
Not sure why you'd miss CISC. It only matters if you do a lot of assembly language coding, and almost nobody does that any more. Even if you're an assembly programmer, the current x86-64 ISA is still outwardly CISC, even if it gets translated into an internal RISC instruction stream for execution.

You're right - micro-ops are a symptom, not a root cause. The disease here is "speed demon" CPU design, most recently the Bulldozer and Northwood before it. Side note, there's no way that the internal translation is cost-free.

It has a hidden advantage though: The arcane variable-length x86 instruction format allows code to be packed very efficiently into memory, reducing pressure on the internal caches and DRAM interface. Since no modern CPUs execute x86 instructions directly, it has essentially become a specification for object code compression. The instruction stream gets "compressed" by the compiler, and "decompressed" into multiple micro-ops when it is decoded in the CPU.

I believe this more efficient use of DRAM space and bandwidth was one of the factors which helped x86 achieve its dominant market position.
Nostalgia isn't what it used to be.
 
setaG_lliB
Gerbil First Class
Topic Author
Posts: 122
Joined: Wed Mar 03, 2010 6:02 pm

Re: Let's party like it's 1999 and look at single-threaded x87 FPU performance

Thu Nov 02, 2017 8:39 pm

DrDominodog51 wrote:
What memory speed and timings were used for each CPU?

IME the 32 million size can be extremely sensitive to memory frequency and latency. I'm not sure how much the 1 million size changes this.

I've added all of the memory and cache information for each CPU. :)

SkyWarrior wrote:
Pentium 4 sucked at single but when you have 2 of them running at the same time even FX57 was suffering.

The P4 I used doesn't have HT, so when I try to run 2 instances of SuperPi at the same time, well...it takes twice as long. :P

the wrote:
There is a good chance the performance jump from Pentium M Dothan to Core 2 Duo Wolfdale is the increase in L2 cache. With 6 MB on-die, that Wolfdale chip has enough to keep everything in cache.

While per core cache sizes haven't increased that much since Wolfdale when looking at the entire cache hierarchy (L1 + L2 +L3), since then Intel has worked in decreasing the cache's latency and/or increasing its bandwidth. Even without touching x87, these changes still provide a small performance boost all around.

It would be interesting to see how SkyLake-EP handles SuperPi in this test. The cache topology has again changed and I don't think the per core amount has enough to hold everything. It might actually end up worse than Haswell-EP/Broadwell-EP.

Didn't even think of that! Wolfdale's huge cache may very well be what's giving it such a huge advantage. I'm also quite interested to see how Bulldozer and Ryzen based CPUs handle SuperPi. The fastest AMD CPU I have is the Phenom II X6, and underclocked to 2.4GHz it doesn't do half bad. I wonder if Bulldozer would end up being slower.
 
chuckula
Gold subscriber
Gerbil Jedi
Posts: 1862
Joined: Wed Jan 23, 2008 9:18 pm
Location: Probably where I don't belong.

Re: Let's party like it's 1999 and look at single-threaded x87 FPU performance

Thu Nov 02, 2017 9:04 pm

mikewinddale wrote:
Intel Core i7-7500U @ 3.48 GHz: 11.000 seconds

I also have an old laptop, an AMD A10-5750M, but it has Linux Mint 18.1, and I can't get the SuperPi Linux version to run. I've tried both the executable and the command-line, and nothing happens. Just nothing. (I also tried SuperPi with Wine. No success.) If anyone can offer any suggestions, then please. But in the meantime, I only have results for the Intel, where I have Windows installed.

On Linux, I was able to get System Stability Tester (SST) to run, which claims to be a SuperPi clone. However, using SST, I got **very*** different results, to wit:

SuperPi, 1M, Intel: 11.000 seconds
SST, 1M, Intel: 2.146 seconds
SST, 1M, AMD: 4.488 seconds



If I had to guess I'd say that SST is "accidentally" using newer instructions than the classic x87 FPU and is therefore not a direct comparison for the purposes of this discussion. Of course, that means it's also probably a more useful real-world tool than classic SuperPi too.
4770K @ 4.7 GHz; 32GB DDR3-2133; GTX-1080 sold and back to hipster IGP!; 512GB 840 Pro (2x); Fractal Define XL-R2; NZXT Kraken-X60
--Many thanks to the TR Forum for advice in getting it built.
 
MileageMayVary
Gerbil Team Leader
Posts: 281
Joined: Thu Dec 10, 2015 9:18 am
Location: Baltimore

Re: Let's party like it's 1999 and look at single-threaded x87 FPU performance

Fri Nov 03, 2017 9:37 pm

Ryzen 1600 @ 2.4GHz
Image
Image
Image

And at my usual 3.75GHz
Image
Main rig: Ryzen 1600, R9 290@1100MHz, 16GB@2933MHz, 1080-1440-1080 Ultrasharps.
 
setaG_lliB
Gerbil First Class
Topic Author
Posts: 122
Joined: Wed Mar 03, 2010 6:02 pm

Re: Let's party like it's 1999 and look at single-threaded x87 FPU performance

Sat Nov 04, 2017 10:48 am

Thanks for the Ryzen results!
 
setaG_lliB
Gerbil First Class
Topic Author
Posts: 122
Joined: Wed Mar 03, 2010 6:02 pm

Re: Let's party like it's 1999 and look at single-threaded x87 FPU performance

Sun Nov 05, 2017 10:39 am

I overclocked my Phenom II x6 1090T to just over 4GHz.
Phenom II (Thuban) at 4070MHz - 16GB CL9-9-9-24 1760MHz dual channel DDR3 - 512KB L2+6MB L3: 15.063s

Which is still slower than Haswell at 2.4GHz with a single channel of cheap (CL11 DDR3-1600) memory. :-?
It's nice to see that Ryzen is a good bit faster than Ph2 at SuperPi. Though I don't have one to confirm, I'm thoroughly convinced that a Bulldozer-derived chip would end up being well behind Ph2 at this particular benchmark.
 
thecoldanddarkone
Gold subscriber
Minister of Gerbil Affairs
Posts: 2401
Joined: Wed Mar 26, 2003 4:35 pm

Re: Let's party like it's 1999 and look at single-threaded x87 FPU performance

Sun Nov 05, 2017 12:49 pm

intel 7800x / currently in dual channel 2933@17,18,18,35 2t

@2400 = 15.125
@4600 = 7.906

*I changed this to 2400mhz for comparison sake.
Last edited by thecoldanddarkone on Sun Nov 05, 2017 5:15 pm, edited 1 time in total.
I7 4930k, 32 GB Ballistix DDRL3@2133 , 1.2 TB Intel 750 AIC, 500 GB mx200, Sapphire R9 Fury, asus x79 ws, HP ZR24w, edifier s730
HP Pro x2 612- i5-4302Y, 8 gigs of memory, 256 ssd
 
APWNH
Grand Gerbil Poohbah
Posts: 3189
Joined: Thu Oct 31, 2002 9:10 pm
Location: Boston MA
Contact:

Re: Let's party like it's 1999 and look at single-threaded x87 FPU performance

Sun Nov 05, 2017 2:57 pm

You mention P100 and I keep thinking of the Tesla. That thing can crunch some floating points...
i7-5820K@3.8, 64GB, X99 Sabertooth, 1080Ti, 1TB Samsung 960PRO
Xeon 12core ES@2.7, 64GB, X99-P SLI, 980Ti, Vega 56
MBP15 2017 touch bar (i7-7820HQ) Radeon 560, 16/512GB
MBP15 2015 (i7-4870HQ) R9 M370X, 16/512 (FS!)
MB12 2017 (i7-7Y75@1.4) 16/512
 
mikewinddale
Gerbil
Posts: 52
Joined: Sat Jan 07, 2017 2:22 am

Re: Let's party like it's 1999 and look at single-threaded x87 FPU performance

Sun Nov 05, 2017 3:55 pm

MileageMayVary wrote:
Ryzen 1600 @ 2.4GHz
...
And at my usual 3.75GHz


Thanks for the Ryzen results. But I'm a little bit confused. I thought the Ryzen 5 1600 had a base clock of 3.2 GHz and a turbo of 3.6 GHz. (See AMD's product description: https://www.amd.com/en/products/cpu/amd-ryzen-5-1600).

Can you please clarify? Is this an under-clock and an over-clock? Or did AMD change the specs at some point?

Who is online

Users browsing this forum: No registered users and 1 guest