I am completely
CPU limited in several games (including Planetside 2) with both my 4.3GHz 3930k and 4.7GHz 2700k, even with a only a single 7950, as most games, even supposedly well threaded ones, tend to be limited by a single thread.
Obviously, my FX-8150 is even more dramatically affected by this, due to it's individually weak cores.
Originally Posted by Moustache
Most remotely modern Intel parts have more than one logical core.
Originally Posted by geoxile
True, but HT doesn't scale nearly as well as the cores in AMD's "modules". AMD's 8-core processors will probably see a bigger proportional performance increase than Intel's i7s.
Only if you have more demanding
threads than physical cores, which is probably a very unlikely scenario for many of these games.
Even going to four demanding threads is four times more than what many games have now; to expect them to be able to divvy things up further than this in the short term is far fetched.
Originally Posted by JassimH
Cinebench is flawed and biased due to specific instruction sets only intel can use (not all games/programs etc. will) and quite possibly due to other reasons as-well.
Therefore it is only useful for comparing within the same direct archetecture, like 2500k vs 2600k vs 3930k or 3770k vs 3750k etc. on intel and an 8 core 8350 vs an fx 6300.
Cinebench has little bearing on gaming performance because it's not a game nor is it based on a game engine.
However, Cinebench is a pretty accurate reflection of Cinema4D performance, specific optimizations and all. This is what it's designed for, and such comparisons are meaningful, even across brands and architectures.
Originally Posted by jspeedracer
The 7zip bench will let you run up to 8 threads on any quad core cpu and you always get a higher score with more threads even if the cpu doesnt have HT. I ran 4 threads on my i3 530 with HT on and off and got the same score. Certain things just benefit more when you increase the thread count, for example dvdflick will let you run 8 threads on any quad core. An i5 2500k and i7 2600k at the same clocks perform the same when encoding with 8 threads, to me HT is practically a software hack that automatically tells programs to double the threads nothing more. AMD actually has more cores so they will benefit more than Intel when games and programs become more parallel and multi-threaded, but depending on the task, once that happens HT will be almost useless.
You can run more threads than logical cores in 7-zip and many other programs, and often get a benefit, both with an without SMT. 24 threads offers optimal performance on my 3930k in large archival jobs.
However, as others have stated, many, probably most, loads are not anywhere near as parallelizable as 7-zip, or other archival programs.
Even the most well threaded games aren't going to scale with core count any where near as well as 7-zip.
Originally Posted by Slaughtahouse
Are games finally going to be 64 bit now?
I've been playing 64-bit Crysis for years, and currently play the x64 version of Warframe.
Originally Posted by sumitlian
If parallelization is so hard, why don't they start to utilize latest instruction sets like AVX and FMA at its max throughput capability.
Many of these instructions are highly situational.
You get no speed boost from AVX when adding two 16-bit numbers, for example, you just waste precious memory in your caches.
Originally Posted by lacrossewacker
Each of those cores is probably around 1/4th-1/2 of a i5-3570k core (AT STOCK)
Probably less. Jaguar cores probably has only a bit more than half the IPC of a Vishera core, and are clocked at half the speed of a 3570k.
An i3 has more total CPU performance than the parts in PS4 or XBox One.
Originally Posted by thegreatsquare
Well, as they say ...necessity is the mother of invention.
And a welcome change it is, because we all tend to have enough cores to take advantage of the aftermath, regardless of what those cores happen to be.
Originally Posted by Usario
How is 2 weak cores vs 2 strong cores any different than 4 weak cores vs 4 strong cores?
Originally Posted by maarten12100
What is needed is a thread layer that gives proc time to all threads while looking like a single one.
This sort of reverse HT just is not possible. It's oft been rumored, but there is simply no practical way for more than one CPU core to work on a single thread efficiently.
Too many coherency issues, no where near enough inter-core bandwidth, and way too much inter-core latency.
I'm not one to say "never" often, but if there is any place where it's appropriate, it's here.
Originally Posted by ghostrider85
The FX-8350 is faster than a 3770k, or even 4770k in several encryption/decryption tasks.
The differences is typically not huge, and overall the FX is still slower, but there are certainly cases where eight Vishera cores can best four Ivy or Haswell cores.
Originally Posted by MerkageTurk
Maybe in next gen games and AMD optimised.
Next gen games are Jaguar optimized.
Jaguar is not Vishera or Steamroller. Jaguar architecture is very different from AMD's desktop architectures.
Multithreaded optimizations will help everything with more threads, anything more specific than that will help the consoles and ultra low power PCs, not mainstream or enthusiast parts of either brand.
Originally Posted by Usario
Gentoo with aggressive optimizations will even allow the FX to rival 2011 CPUs in Blender.
Against a build just as deliberately optimized for the LGA-2011 part?
I find this hard to believe, especially since Blender is helped enormously by HT. Indeed, the Windows build of Blender could well be the poster-child for hyperthreading.
Originally Posted by Tippy
I believe FX8350 out-performs i7 3770K in Crysis 3. As far as I know that's the only game so far which extensively
uses multi-threading, from what I've seen in benchmarks.
It's pretty close, but the 3770k tends to pull ahead with current versions of the game.
Originally Posted by Tippy
I'm still confused how an FX8350 is able to keep up with a 3770K in a game... 3770K is a quad-core hyperthreaded chip (so 8 virtual cores). I guess 8 physical Piledriver cores > 8 virtual Ivy cores?
SMT doesn't give you more cores, it allows well threaded, but not particularly ILP efficient, programs to utilize resources within a single core that would otherwise go unused.
It's performance impact is extremely variable (ranging from -5% to +60% with an average in the ballpark of +25%).
Originally Posted by ZealotKi11er
HT makes 0 difference in games. I have had HT since P4, Core i7 920 and makes 0 difference.
Not always the case.
There are maybe two or three games in existence where HT can benefit my hex-core Intel parts. However, most remotely recent games will benefit to some degree from HT on my dual-core parts.