Originally Posted by Kuivamaa
You are doing quite the cherry picking here.
They are cherry picked lightly threaded comparisons. That was the point I originally made and the point that those benchmarks prove. If you can provide benchmarks of the exact same tests that show a different result due to "conditions" being different, be my guest. The only way anyone can be mislead by the "cherry picked" benchmarks here, is not to consider the context that they are defending. The context here, is obviously in desperate need of a reminder.
I made a broad, general point about IPC. That is absolutely accurate in compute bound comparisons that both chips can compete in. Now lets see how many excuses we can come up with to take light off of the point I made in order to bury it under the rug. We wouldn't want anyone to stumble along and find this thread, and learn anything about the differences between AMD and Intel IPC, rather, we need to muddle this up as much as possible so that nobody finds out.
Unless I am mistaken this is based upon rankings provided by users. Unsurprisingly it fails to mention turbo frequencies (doesn't even mention any frequency on many FX chips) and/or overclocking which might explain why results are often all over the place (eg. i7-4960HQ@2.6 appears to have higher ST performance than another haswell cpu, i5-4670k clocked at 3.4Ghz which is laughable.). There's a reason passmark is generally ignored.
A minor presentation problem is being taken literally and without any reasonable insight by you. The chip speed
is listed based on the advertised "base" chip speed as part of the chip's "name" in the list in many cases. Performance results do include turbo frequencies, which is evident by the i7-4960HQ you pointed out.
The i7-4960HQ, scores just shy of the i7-4770, because it has a turbo frequency of 3.8ghz, just shy of the 3.9ghz turbo found in the few chips that out-perform it. The SINGLE SAMPLE size of the HQ chip actually falls inline with what one would expect. Impressive how accurately even just a single sample for a chip winds up in the charts where it belongs.
I do not believe overclocked results are effecting the average passmark scores significantly because there are no "leaps" in results where "K" edition chips exist vs locked chips. Furthermore, the passmark results mimic the ratios of IPC I have seen everywhere else in benchmarks around the web. The only reason to suspect foul play here is if you are TRYing to find it where it doesn't exist in order to sweep a well known trade-off of AMD chips under the rug.
Cinebench 11.5 is notoriously compiled with ICC which actively checks what kind of cpu it is ran on-If it identifies it is on AMD it will run the slowest code possible on purpose.Nothing to do with specific workload,it flatout ignores instructions present on non-intel processors effectively handicapping them. It is like having two runners, one running 100m and the other 110m hurdles, checking their times and declaring a winner.
I see AMD enthusiasts look to this benchmark all the time to test chip performance. I've even seen some OC results bring some richland chips up to within spitting distance of i7's. I have never heard that it favors intel heavily until now. In the multi-threaded benches, the modern quad-module FX chips perform about on par with a sandy bridge i7, which makes perfect sense considering the cache latencies, raw clocks, etc.
Purely your assumptions here.
First of all, single thread=/= dual thread.For example it was frequent in the early era of BD architecture to have applications that would load a single module ,triggering shared resource penalty instead of spreading workload around a bit. It still occurs today.
If AMD had marketed the "modules" as "cores" from the get go and treated them that way all along, (a "tandem threaded" module, like hyper-threading), then this penalty caused by improper scheduling would have been solved before it were ever a problem. Any applications still taking a hit because of this scheduling problem have AMD to blame for it by my figuring. If this problem is still going on after 2 years now, then you can't play this "card" anymore. I think this particular bathroom pass was used up for this period, sorry.
Other than that you include Skyrim, (which is dual threaded)
It's up against dual module
chips so so the comparison is still valid. You should be applauding this particular comparison because it's one of those that clocks in showing less IPC advantage to intel than many others. Not sure why you would want to argue with this specific metric as it is one of AMD's best showings of the bunch. I feel you're trying to cherry pick nitpicky problems
to try to discredit a very obvious trend in the benchmarks I am pointing out. YES my benches are cherry picked to include lightly threaded CPU bound scenarios. Any benchmarks which are more parallel, or bottleneck somewhere else, need not apply as those positions do not define single threaded IPC performance.
and ofc notoriously used obsolete x87 which AMD seems to have abandoned (there is a whole story about The Stilt and his research on the subject, you can google for it).
If AMD has abandoned something and the result is a regression in performance in real world apps/games, that does not excuse them from being compared in their regressed state of performance. By the same standard you are asking me to accept here, if AMD were to abandon the production of CPUs, they would automatically be better performing because they exist in a purely hypothetical construct.
When this particular game title, is run with every video setting awkwardly skewed to something very unusual in order to produce hundreds of frames per second, the bottleneck is not actually landing on instruction level performance, rather, it is landing on IO traffic limitations somewhere. Understanding what you are looking at is half of the battle. Unfortunately, I have to call foul here as this benchmark you have chosen is more cherry picked to show a result that is not representative of real world comparisons than ANY of the benches I pointed out, because it is not measuring compute performance at all. The Pentium would get crushed here because it is turning at a slower speed, not because of any difference in IPC. If they threw an old Pentium 4 clocked to 4ghz in there it would perform pretty middle of the road in this benchmark, which would REVEAL the misleading nature of this sort of benchmark.
THIS IS NOT A CPU INSTRUCTION
BOUND BENCHMARK, IT IS HIGHLY MISLEADING, DO NOT BE FOOLED.
Starcraft II also famously runs better on intel ,after all as an e-sport it is actively promoted by intel.
By that standard then Supreme Commander II, which starts up to an AMD logo, should run better on AMD chips. Unfortunately, that standard doesn't hold up as Intel chips beat AMD chips in the AMD promoted SC2 by similar margins as they do in the Intel promoted SC2. Funny how that works out. Realistically, what is going on here is simple. These games are greatly effected by L3 performance. Intel's L3 cache latency is far lower, and the prediction algorithms are more refined. These game engines thrive on that L3 performance and suffer on chips without it or with poor latency.
Seems like this is inline with the point I was making. Your find here actually helps average things "down" to my original "4.8ghz" ballpark figure. Since most of the benches I wound up finding to prove the point actually pointed to a more extreme difference in IPC.
Looks like another example more or less backing up my position, with FX needing a lot more clock speed to match the Intels.
The reason I am so hostile towards pentium chips is that they are overrated to no end.Nothing to do with my supposed "love" for AMD (btw vishera brought 10% + generic ipc boost and it was often more than 20%+ in games). Tom's BF3 review 2 years ago ("cores don't matter in this game") lead many people to build machines around pentiums only to find out they would run the game like a dog and not just bf3, plenty of online titles.
Poorly executed benchmarks are common. Understanding how to "spot" when a benchmark has been botched comes with experience. The benchmark you are talking about here is a great example of how a benchmark can be easily misinterpreted. I find that more often than not, over at TomsHardware, they use pretty good test methodology but fall short of a complete follow-through to confirm that the method is good from all angles, and then draw all sorts of bad conclusions based on assumptions that could have been disproved had they ran a few more tests. More often than not, when I see their benches, I am able to see where the bottleneck in a set of results actually is, but then they go on to say it is something different.
I would point out, that this sort of "bad conclusions" problem is the same exact problem you had with the Torchlight II benchmark. You "wanted" to see a particular thing in that benchmark, so forgot to even consider the possibility that you weren't looking at a result of instruction based performance. Had the test been re-run with many more settings and with different GPUs, and with various bus speeds tinkered with, they would have eventually "found" what they actually were
bottlenecking on there. Compute performance and traffic-directing are different issues.
I have honestly looked at hundreds, if not thousands of benchmarks comparing CPU's over the last 3 years. The trend has been consistent since SB came along. The only place where AMD chips ever appear to match or beat Intel in lightly threaded workloads, are in cases where the bottleneck is somewhere else. The "cherry picked" comparisons I have made, are cherry picked to point out real world compute-bound comparisons. AMD defenders love to point to benchmarks that are bottle-necked on IO, event timing, disk speed, or GPU, to show that AMD is "just as fast" as intel. Hiding the CPU performance does not make it better.
They are limited in so many ways, instruction sets,[etc etc...]
completely out of the scope of issues related to the original point.... You're just trying to trash a chip in order to sweep the big picture under the rug here. In case you had forgotten, nowhere in this thread was I suggesting that anyone BUY or USE a Pentium for any specific tasks. I merely pointed out the IPC vs parallel compute advantages/disadvantages of intel vs AMD and made what I now realize was an enormous mistake of using the Pentium chip to make the point. Apparently, people get very very butt hurt when they find out that celerons and pentiums can do many things just as fast as FX chips.Edited by mdocod - 11/20/13 at 3:38am