soccerballtux
Lifer
- Dec 30, 2004
- 12,553
- 2
- 76
that was without HT6.8x scaling, with HT, for 8 cores on a synthetic benchmark is pretty poor.
that was without HT6.8x scaling, with HT, for 8 cores on a synthetic benchmark is pretty poor.
*Sigh* Again, look at that last bench I ran and compare it to my original bench at same clocks with HT on in post #50. HT is disabled, 8 cores, 8 threads. The scaling with HT is what is poor. 6.8X scaling with 8 threads (no HT), it only increase to 7.9X with 16 (HT on).6.8x scaling, with HT, for 8 cores on a synthetic benchmark is pretty poor.
We'll see. Went downhill from there
Set multiplier for 48x upped vcore 0.020v's booted windows did some quick testing which seemed fine....Hmm.
Went for the illusive 50x upping vcore 0.040v's more which probably was too little....Boot loop from hell is what I got! Bsod and reboot followed by the loop. Clearing cmos was a no go! Pulling battery and clearing cmos was a no go! Pulled 1 stick of ram and she booted with bios warning screen. Set to 42x profile booted and shutdown. Put stick back in boot loops. Swapped to other slots boot loops. Go figure!
Yesterday I had a first look at the assembler code found in the hotspots (64b version). The benchmark creates hundreds of threads for the MT test (although only a small number overlapping at any single point in time). The calculations use scalar single precision SSE ops to do something. I'll write down the formulas. Maybe it's some fractal calculation. But there are also things like (integer) a=(b*c)/b.Yeah, I'm guessing it's just really light on FPU use allowing the FX to really act as 8 pure cores. Eight real Intel cores just demolishes the 4c/8t Intel's.
*Sigh* Again, look at that last bench I ran and compare it to my original bench at same clocks with HT on in post #50. HT is disabled, 8 cores, 8 threads. The scaling with HT is what is poor. 6.8X scaling with 8 threads (no HT), it only increase to 7.9X with 16 (HT on).
*Sigh* Again, look at that last bench I ran and compare it to my original bench at same clocks with HT on in post #50. HT is disabled, 8 cores, 8 threads. The scaling with HT is what is poor. 6.8X scaling with 8 threads (no HT), it only increase to 7.9X with 16 (HT on).
the difference surprised me, they are usually a lot closer on benchmarks
now I know this benchmark is influenced by memory, but the Athlon with single channel has lower latency and copy/read not to far from the pentium with dual channel, so I guess there is something else that makes the k8 so slow...
Yeah, why I don't push this old thing too far these days.
I pretty much know what it can do more or less all ready, when ya start having to pop the battery etc just to go farther and learn a new rig setup it can be a real PITA.
It also makes heavy use of int and fp division. CPUs with HW int divider have another advantage.now I know this benchmark is influenced by memory, but the Athlon with single channel has lower latency and copy/read not to far from the pentium with dual channel, so I guess there is something else that makes the k8 so slow...
the difference surprised me, they are usually a lot closer on benchmarks
now I know this benchmark is influenced by memory, but the Athlon with single channel has lower latency and copy/read not to far from the pentium with dual channel, so I guess there is something else that makes the k8 so slow...
Welcome to this forum! Is "qookap" related with the ever smiling animal Quoka?I won't agree those point about K8 but still waiting more bench.
I m finish my work about processor microstructures research yesterday. might be share it later..
My research and benchmark tell me there is no relationship between CPU-Z and DRAM,
it's a success pure CPU performance benchmark tool just like sisoft sandra(before fully commercialize, the best fault as benchmark tool like passmark lol)
I m really thankful for bench data. hope more AMD side bench to help me review the CTP and Flops/Cycle. AMD's CPU never have a clear data like Intel...
now I know the A10-7850K just like Core 2 Quad performace, it's basic 8 single precition floating point opreations per second, 4 DP Flops/sec. Theoretically their performance should very close, streamroller(APU) had 8 DP in 1 module, each module including 2 cores.
------
ps. this is my first post at here and forgive my poor english...
As often as the integer code uses idiv, there surely is some fighting for related ressources (HW divider or ex units needed for the ucode) on a hyperthreaded core.Logical cores are not scaling well, physical cores scale like mad, see post #123 above.
I think the disadvantage for the athlon is single channel memory and lower frequencies.
I still have my opteron 170 being used as an HTPC. That has DDR1 500mhz models in it if I can run the bench on that we will see if it closes the gap.
FX8150 @ 4.6GHz
win 10 64bit