Basically AMD must have been lying their rear ends off as far as throughput (40% my rear). Each Zen core is indeed pushing between 2.5 and 2.7 K10 cores worth on blender... and one Piledriver core at same clock is only worth around .85 of a K10 core (at least on Cinebench).
You are way overestimating what 8 cores and 16 threads would do for the Core 2 chip and what hyperthreading (SMT) would do for it. 8 cores and 16 threads is not 4 times faster than 4 cores.A clean run at 150 samples on my C2Q @3.52GHz resulted in 2:45.75. A theoretical 16-thread Core 2 chip at the same clocks would then be 41.44 seconds. At 3.4GHz would then be 42.9 seconds. My CPU under full load uses 88W according to HWmonitor - so a 16 core version would be 352W. Zen with 16 threads is 36s, all in 95W (although momentary platform power during the test was ... 160W? My PC under CPU stress (OCCT) hovers around 190W. That adds up to roughly 450W platform power for my theoretical 16-thread Core 2, which still underperforms Zen by 20%. And half of those threads are virtual, of course.
No matter the talk of stagnation in the CPU space, I'd say this is pretty good progress in 8 years.
A theoretical performance increase of just 20% doesn't sound too great. Then again, my chip is clocked way faster than any that existed back then. Also, 4 cores were as high as you got back then. So what we're seeing is a 20% increase in raw IPC including simultaneous threads, while consuming ~1/4 the power, while packing 2x the cores and 4x the threads into the chip. Not to mention new instruction sets and various capabilities that have been added.
I'll be very happy to upgrade to Zen when the time comes.
edit: I ran the test on my A8-7600-based HTPC as well, which ended up at 4:36.15 for 150 samples. The chip is 3.1-3.7GHz (and power limited), but clocks fluctuated between 3.3 and 3.6 during the test. I'd guesstimate an average of ~3.45 or so. That's a very, very solid IPC increase on the part of Zen.
Yeah, but he was comparing it to an 8c/16t CPU's score.bystander36, he was calculating for 16C/16T C2Q chip, not for 8C/16T.
In gaming you get up to 100% boost if there are enough threads.Edit: Maybe someone could help him here with real world benefits of SMT on Intel platforms. In gaming, HT has at most 10-20% improvements. I think some synthetics may show as much as 30% performance gains from HT, but I have not been following this in quite a while.
Of course that game requires 4 cores to meet the minimums. The i3 mimics it enough. That said, as shown in this thread, 26% is all HT gives a boost to performance.In gaming you get up to 100% boost if there are enough threads.
In game benchmarks the boost is not that big because it's only rendering the graphics and rendering only benefits a little bit by HTT.
Also most tests of this sort are being done on i7 CPUs that are already fast enough even without HTT so the difference is very small.
Yes, as that is what Zen is. Core 2 never had any form of SMT, and as such making that up would be utterly useless. My point was to show how even when compared to an extrapolated theoretical best-of-he-best chip from way back when there have been huge gains both in IPC and power. Of course a 16c16t Zen chip would grow this lead even more - but there is no such chip. And SMT is one of the innovations that has given us these huge efficiency gains. That Zen can, even with half the "real" cores of my theoretical Core2Sedecim, outperform it by 20% while consuming 1/4 of the power, that's what makes this impressive in my eyes. Of course it would utterly crush a theoretical Core2Octo with 25% or so added for SMT. But beating a "true" 16-core is far more impressive IMHO.Yeah, but he was comparing it to an 8c/16t CPU's score.
I find it extremely unfair to compare an even theoretical 16 core chip to an 8 core chip with SMT. That just doesn't compute at all as a reasonable comparison, if you are going to say that the 8 core chip is unimpressive for not being a ton faster.Yes, as that is what Zen is. Core 2 never had any form of SMT, and as such making that up would be utterly useless. My point was to show how even when compared to an extrapolated theoretical best-of-he-best chip from way back when there have been huge gains both in IPC and power. Of course a 16c16t Zen chip would grow this lead even more - but there is no such chip. And SMT is one of the innovations that has given us these huge efficiency gains. That Zen can, even with half the "real" cores of my theoretical Core2Sedecim, outperform it by 20% while consuming 1/4 of the power, that's what makes this impressive in my eyes. Of course it would utterly crush a theoretical Core2Octo with 25% or so added for SMT. But beating a "true" 16-core is far more impressive IMHO.
I find it extremely unfair to compare an even theoretical 16 core chip to an 8 core chip with SMT. That just doesn't compute at all as a reasonable comparison, if you are going to say that the 8 core chip is unimpressive for not being a ton faster.
Sorry, when did I say that? I'm quite sure what I said was the exact opposite, as in that I'm very [bleep]ing impressed by Zen. So please, READ before you answer. Jeez.I find it extremely unfair to compare an even theoretical 16 core chip to an 8 core chip with SMT. That just doesn't compute at all as a reasonable comparison, if you are going to say that the 8 core chip is unimpressive for not being a ton faster.
Which was a rhetorical point to underscore that this is 8 cores (not 16!) with SMT (which, as everyone, their dog and their grandmothers have been pointing out adds ~25% perf) beating 16 "real" (theoretical) cores by 20%, all the while consuming 1/4 of the power. How is that not impressive? I think that's pretty damn impressive. That means, in essence, that an 8c8t Zen chip would tie 16 Core 2 cores. 16. That's twice as many! At speeds they never came close to in production. How is that not impressive? I'm sorry if my apparently excessively complex way of writing confuses you. Or, well, I'm really not.A theoretical performance increase of just 20% doesn't sound too great.
Now can you please stop your straw man argumentation please?I'll be very happy to upgrade to Zen when the time comes.
At the same clock? You said your i5 runs at 4,3GHz. The Ryzen demo is at 3,4 (6900k is 3,2-3,7 IIRC). That's quite a difference.Encode time: 1 minute 40 seconds.
I am just about x2.2 times slower than Ryzen at the same clock and about ~2x slower than 6900K at the same clock.
i7-5820k encode 00:01:20Since now we have a file to DL and instructions how to run Handbrake test, how about we post our results from Handbrake as well?
We can make a new topic if need be.
Anyway, my Handbrake results on 4690K i5 @ 4.3Ghz, DDR3 @ 2400Mhz 16GB, win10 pro x64 and following instructions summed up here by Dresdenboy and Car Merc:
https://twitter.com/Cat_Merc/status/810515806316560384
Encode time: 1 minute 40 seconds.
I am just about x2.2 times slower than Ryzen at the same clock and about ~2x slower than 6900K at the same clock.
I have already calculated the clock difference in there . So all things accounted for, my i5 (Haswell 4C/4T) is about ~2.2x slower than Ryzen 8C/16T if both were running at the same clock.At the same clock? You said your i5 runs at 4,3GHz. The Ryzen demo is at 3,4 (6900k is 3,2-3,7 IIRC). That's quite a difference.
Didn't Lisa Su state that they used the "Ryzen intro video" ("the video you just saw") as the test media in those Handbrake tests?
Ah. That wasn't clear from the way you wrote itI have already calculated the clock difference in there . So all things accounted for, my i5 (Haswell 4C/4T) is about ~2.2x slower than Ryzen 8C/16T if both were running at the same clock.
Ah. That wasn't clear from the way you wrote it
Yep, there's nothing there indicating that you've done calculations to equalize things. Those statements can just as easily mean that you've simply misread the specAre you sure?
"I am just about x2.2 times slower than Ryzen at the same clock and about ~2x slower than 6900K at the same clock."
I bolded out the important parts in my original post.