AMD ZEN Engineering Sample AOS - Further Performance Analysis

exactly!
i mean i have HT enabled because any OS tasks, archiving, etc. profit from it obviously.
and yes, speaking of non-gaming stuff like encoding, video cutting etc...
this is where considerable gains are being brought in by HT.

but some people in here painted a picture as if HT has importance in games.
i just say: not yet.

 

i mean this in the politest way possible...but he said that because you throw all your thoughts together in a jumbled wall of text. nobody says anything to me about my lack of capitalization or missing punctuation simply because i first think about exactly wat i want to say - & then, more importantly, structure my post appropriately in order to convey that point.

i find the results of this testing quite intriguing & like everyone else have questions about the benches methodology & control variables (like threads utilized, hyperthreading, memory bandwidth throughput, lack of chipset/OS optimizations, etc etc etc), but it does us exactly no good to bicker about it all. the bench was rough around the edges & it approximates an interesting picture: if the leaked info is accurate, the ES is performing quite well compared to, well, everything. period.

 

You (and they) are COMPLETELY wrong about HT in gaming. The very tangible effect it has is that it reduces latencies and, in consequence, stutter. These will never show in the tests you link, because none of these guys analyzed frame latency data, nor they use any kind of more advanced statistics on the data they gather. You're gonna feel it while you play alright.
Only Overclock.net has done some tests about it, and the 99th-percentile numbers in Crysis 3 and the Witcher 3 show that the latencies were up to 50% lower on the i7. GTA V didn't seem to have any statistically significant difference between them, but that's an engine written for the PS3 and the Xbox 360 with some upgrades for the PC. Everything was running much better on the i7 CPUs, when all else was equal:



That's the reason that no one should ever get i3's over FX-8xxx CPUs, but no one listens anyways.

SMT is actually bad for most scientific workloads due to resource contentions that lead to crazy latencies. Rendering videos, or non-realtime applications can benefit from it, as code that works on unpredictable workloads (like a game). For the non-realtime apps it can make sure that resources are 100% used, and for games it functions a bit like out of order execution on "parallel" instead of "ahead". It's actually really well suited for games, since draw calls are inherently parallel operations. This is from your own link:

As shown above it's really the latencies that make a big big difference with HT on, and with more modern APIs/Engines there are also very tangible differences on submitting draw calls to the GPUs, often reaching almost EDIT: +60%(not 80) of performance in comparison to a similarly clocked i5.

 

Prminister, where did you get that 80% more drawcalls? In that chart there is 24,1% benefit on 4C/8T compared to 4C/4T. Or did you mean some another example?

But anyway we have sidetracked a bit from topic. My original point when I started hyperthreading tests with aots was to show that is really scales with threads. Therefore it makes no sense that Zen is only "40% faster with same clock than bulldozer is, when it's actually 8C/16T chip compared to 4C/8T chip. Even if hyperthreading wouldn't work properly on Zen, those real cores should already give it a bigger edge, and that fact that it's so far from 5960x running at same speed on same core config tells that there is deffinately something "wrong". Let it be bios, windows cpu scheduler or memory configuration that can't feed Zen.

 

80% more draw calls does not mean 80% more performance.

 

Erm? 80% more drawcalls does not mean 80% more drawcalls? That last chart I was referrering was drawcalls.

 

My bad I misunderstood what you were referncing.

 

I corrected it. It's +60% with latency mainly, not FPS in most cases. There seems to be something wrong indeed there. We'll see soon enough I guess. There can be a million things.

PS: Or there's nothing wrong and if we extrapolate it on a 3.2GHz FX, it is indeed 40-60% faster, and this is final IPC, with final performance depending on final clockspeed. I don't buy that yet though.