I understand you totally, however I'm using the same settings as everyone else and unless you can suggest a way to prove your conspiracy theory then I would simply say it is but a synthetic benchmark that is flawed. Below are my full system specs - OS - Windows 7 Ultimate x64 Intel i5 6600k day to day at a sedate 4.8GHz @ 1.325V AsRock Z170M OCF Motherboard Crucial Ballistix Elite 16Gb kit (2x8Gb) running at 12 13 13 28 1T @ 2666MHz Zotac 1070 Amp Samsung NVMe 960 M.2 Evo SSD (You may wish to change your tin foil hat about now also)
I think this is getting a bit out of hand when you start talking about conspiracy theories and tinfoil hats - I'm just evaluating the results that are posted in the thread, and your results deviate significantly from every other submission. It's obvious that something's up with your install/configuration of Blender.
In a word, whatever. Unless you know of the PC software equivalent of a cavity search or drugs test you might have me perform...?? You have to consider variables such as OS, overclocks, latency, bandwidth etc etc etc in all of this also... In the interim it is what it is.
I think Paul's score shows that perhaps the Blender test isn't an entirely reliable metric - I've spent the last couple of hours running various configurations. Each time I make a change, I run the benchmark three times: Bumping the clock speed from 4.6 to 4.7GHz with the RAM at 2400MHz gains me another second ~ 46.6 secs. I'm running in a SFF case so I'm not willing to push the CPU further without putting it in my spare case with an AIO. I then ran the benchmark at that speed with the ram at 2666MHz with both tight and loose timings , and the gains were negligible - I did see 46.4 secs on one run but the average was still 46.5-46.6 secs. Running the same tests with HT disabled adds approx 20 secs to the time. If such a discrepancy exists between 6600K and 6700K on Z170 boards, it casts doubt on the performance claims about Ryzen too.
OK a little more info. Apparently this is what is believed to be a 1700x running the Integer test in CPUMark. So I decided to run this on my Xeon (8c 16t 3.4). I got this. You can get it here. https://www.passmark.com/products/pt.htm And only have to run the one test. Takes a couple of seconds.
Yep, just goes to show that a single benchmark (and unknown hardware configuration) doesn't make for a reliable indication of Ryzen's capabilities. Let's wait and see how Ryzen fares in replicable, directly comparable real-world testing.
It's not an unknown configuration dude. The file provided by AMD is exactly as they ran it. There may be other variables coming into play here, like cache for example. Don't you overclock it on the new processors? well that could be why they fare so well compared to higher C/T CPUs. Any way, we now have some more info and another bench to try. Let's see if we get the same sort of hierarchy using this test.
i5 6600K at default speed ( topped at 3.89 Ghz turbo during test) + 16 GB RAM at 2400 Mhz (2x8). Ran it 5 times and the best result was 52:80.
Pretty much what I get with my 5820K at stock (51.18), exact same RAM speed too. Very strange results from these shifty 6600K users.
I don't know dude. I do know that Skylake was up to 15% faster than DC i some things. I just put it down to DDR4 vs the DDR3 but now I am not so sure. As I say, doesn't the cache clock a lot higher on Sky/Kaby? CPU mark seems to make sense. The 6800k stock scored ever so slightly less than my 8c 16t Ivy at the same clocks. Which would make sense, given I have 10 core cache on an 8 core CPU a couple of gens behind. Oh I see. Ah well, we've got another one to try now
Problem solved - the ridiculously fast 6600K renders are not comparable because they use the most recent version of Blender which was released 8th Feb (ver 2.78b). This version appears to be around 40% faster than the previous one, LOL. It would probably have been a good starting point to stipulate we all use the same version of the software. Oh, hindsight how I love thee. Here's my score (HT off) with the new version of Blender: So can people please use version 2.78a if they want their results in the table? Or is there any point any more?
In light of the above information, I think the comparisons are OK for this sort of thing - a lot like we did with Geekbench all dem years ago.
Interestingly enough, just a tad slower than six physical cores without HT. Going by what other people have been saying, background programs also have quite an impact on results as well. I used to run Geekbench in diagnostic mode... I wonder if Blender would benefit from the same.