Found the source I guess for the mame numbers, they noticed the oddity about results matching with different versions as well(poster probably botched a cut n' paste.) http://www.mameworld.info/ubbthreads/showflat … =&Number=343433

Wonder what their reasoning is for using 0.65

I downloaded 0.65 to bench but max cycles didn't work, would set to 2500 with "cycles max" and alt-f12 had no effect. I wanted to compare with my 3570k at 4ghz, it does from 159 to 162.5 with 0.74. edit: Ah, 0.65 uses "cycles auto." I surmise they're using 0.74 and just using an old graphic because with 0.65 I only get 87.5 fps and it doesn't square that a 3570k@4ghz would perform 32% worse than a 3770k@3.5ghz and I pretty much match their results using 0.74.

I'm really curious what the 5775c would do at 4 ghz and max oc now, the ultimate chip for emulators?

The MAME results... Yeah, that's taz-nz, one of MAMEWorld's favorite benchtesters. Funny how that blog picked up his results and reposted them...

They used much outdated 0.65. The results may differ from the current 0.74. Wich Quake's version is unknown and how they've measured fps is unknown too (according to screenshot it could be demo1).

For demo1 with 4770K 4 GHz we had 174 fps, what to interpolate to 3.9 GHz (as DOSBox will work in turbo) and with the console at bottom (viewsize 100) would be 174 * (39/40) * 1.133 = 192 fps. While they got only 139 fps, - 28% lower. Probably because of different DOSBox and Quake's versions.

Compared to 4770K (3.9 GHz), that 5775C (3.7 GHz) is 11% higher, 6700K (4.2 GHz) is 17% higher. Hence in 0.74 we may expect: 188 fps and 198 fps for them.

So, the touted "inverse hyperthreading" gimmick doesn't lead to 4x performance on single-threaded apps likes dosbox yet 🙁 that's a shame.

Interlace wrote:

So, the touted "inverse hyperthreading" gimmick doesn't lead to 4x performance on single-threaded apps likes dosbox yet 🙁 that's a shame.

The shame is CPUs after 8 years are only +60% faster in a single core process. In multicore, as we have same 4 cores on top CPUs, I doubt the situation is much better.

Well, you could also translate that to today's cpu's being 60% more efficient per cycle/instruction since clock speeds haven't been gaining too much ground.. my q6600 ran on 3,2ghz in 2007.. still a very small increase for such a long time. I was hoping inverse HT would change that in a bigger way, like DX12 has a huge boost thanks to multithreading, but in reverse 😵

Interlace wrote:

you could also translate that to today's cpu's being 60% more efficient per cycle/instruction

Ok. 60% after 8 years. Compare with P4 -> Core2 140% efficiency jump made in 6 years. We see shame anyway.

If you'll look at Trail Bay Z3736G - 4 cores, 2.16 GHz, GPU, low TDP - it costs $20 only! It has artificial limitations like some options not included or blocked, microcode and MB BIOS with settings wich reduce its speed without good reasons, the frequency wich I'm sure can be ~1.5 times more with close selfcosts for the producer. Now look at 6700K sold for $340 - it's very close from producing technology side, but costs 17 times more! Besides strange technology stagnation the producer may to get astronomic % of profit, wich may be masked or lied in official reports (with help of state bureaucrats this may be done easily, while strategic and big producers have links with them). Such situation lasts from middle of 2000s.

my q6600 ran on 3,2ghz in 2007

in 2007 there were already 4 core CPUs wich overclocked on typical ~30% to get ~4 GHz. after 8 years we have same 4 cores on same 4 GHz wich work only 60% faster, but have top prices >$300. near on a shelf are same CPUs underclocked and with cut some options with their real price ~$20

like DX12 has a huge boost thanks to multithreading

When DX12 games will massively appear a possible huge boost may be explained by that most of these games were made for consoles on 2010 year level of technology. While I'm not sure in "huge boost", even if it will be possible technically. Like recently we din't see "huge boost" in games made for ancient Xbox360 wich were ported to modern computers a couple years ago.

No, seriously.. up to now PC's with more than a single core have been severely bottlenecked for GPU rendering. Everything before DX12 could only allocate a single thread to handle draw calls. They should've implemented that way back when the Pentium D came out. Check out "a boy and his kite" tech demo. They needed a Titan X back in March to run that on DX11 at 30fps, don't be surprised if it turns out a gtx970 with enough memory would have been able to do that all along.