Gamers - quad core/hex core dilemma

[SPBHM]

some of their CPU load data looks very different to what I see here with the same model of CPU on task manager, like 0 HT load on their graph while here I'm seeing the core and HT "core" with the same load.

 

[Headfoot]

I want cores equal to the number of cores used by any game I will play on it +1. I remember back in the day the gains you could get by running 2 Pentium IIIs on a server board in the single threaded games of the time due to the second core taking over background tasks. So, I'm always going to want more cores than the game can use so background tasks will not take away CPU cycles from my primary cores. These days for cost reasons, I'll probably settle for an equal number of cores as can be used, so long as they have HT. So, 6-core with HT is probably enough given the new consoles are limited to 6 cores for games for the time being.

But... if 8-core Haswell-E can overlock as high as a 4790k provided good enough cooling (high-end water) I may give in.

(Sidenote: I'd bet money that halfway through the PS4/Xbone's life they will reduce the reserved for OS cores to 1 and give the 7th core to games as a sort of life-extending "refresh," probably only accessible via certain API's for certain purposes as to ensure original games running normally)

 

[biostud]

When I upgrade I want to be able to feel a difference, so X99 + hex core is next for me.

 

[DominionSeraph]

You're being ridiculous right now. Just read your comments a few more times.

It needs to be done when talking to people who don't know what they're talking about.

 

[crashtech]

It needs to be done when talking to people who don't know what they're talking about.

It only needs to be done because you want to be rude and don't care enough to explain the point you were trying to make directly.

 

[TurtleCrusher]

It needs to be done when talking to people who don't know what they're talking about.

I'm just going to assume you're having a bad day.

 

[Lepton87]

We are on the verge of transition past 4 cores for games, we're not there quite yet but there are games that can use more then four cores and gain performance in doing so.
I know that one swallow does not make a summer but it's time to abandon 4 threaded CPUs for gaming and at least think about 4c/8t. When I bought my SB HT was a detriment to performance but times are changing. Still, I have limited trust in windows scheduler so right now I would go for a hex for sure.

 

[tidy]

.

 

[SPBHM]

It still is a detriment to performance in many cases. As you say, we're on the verge of transitioning past four cores, meaning a lot of modern games like Far Cry 3 actually using all four (instead of the 2-3 that's been standard for a long time), so right now HT's negative scaling probably does more damage than ever. I think it probably is a prudent time to abandon 4 threaded CPUs, but abandon them for more real cores, not more virtual cores.

With an i7-4771, there was a good 10fps difference in Far cry 3 between HT on and off, in favour of HT being off.

10FPS from 20 to 10 or from 200 to 190?

anyway, this is a new test using FC3, probably with all the patches

the 4690K vs 4770k run at the same clock and one have HT on the other off,

the game doesn't look to benefit from HT on quad cores, but it hardly looks like it's causing a problem,

and if you only have 2 cores HT is working beautifully

if you look at the AMD CPUs, this game is not using well multiple cores/threads over 4 anyway, so it's no surprise the i7 HT is useless here

 

[Galatian]

I think one also needs to look at performance per dollar. 6 cores might be slightly faster, but at what cost? Intels Socket 2011 platform has a much higher cost of entry, so it's not really suited for everybody, especially since you can get 90-95% of the performance for like 60% of the cost (CPU+Mainboard).

 

[moonbogg]

Dang, look at those hex cores catch up in the poll. I didn't expect that.

 

[tidy]

.

 

[avx81]

I think one also needs to look at performance per dollar. 6 cores might be slightly faster, but at what cost? Intels Socket 2011 platform has a much higher cost of entry, so it's not really suited for everybody, especially since you can get 90-95% of the performance for like 60% of the cost (CPU+Mainboard).

And that's exactly why at this current time that going with 6 or more cores is not worth it if all you do is play games. There's only a small handful of games that would benefit from it and you will easily be spending $300 more just for the CPU alone for that very minor benefit. For some people that $300 could be the difference between going with a 4790k or 5930k vs just picking up a 780ti over a 770 that would offer a much better bang for the buck. I realize that there are some enthusiasts on these forums that are always going to go with 2xTitans or 2x780ti and want the latest most expensive CPU to go with it but most people are not interested in spending that much on a gaming pc.

High res benchmarks have been linked before with high end cards running comparing 4 and 6 core cpu's and the difference being little to none. Someone will always come along and link the 1 of 10 games that actually benefit from it running at some low resolution showing a 10-20fps difference when both are already running like 300fps or something ridiculous. Realistically at high resolution there will at best be a few fps difference which wont be noticeable in actual game play and does not justify $300 difference. I doubt many people can sit here on the forums with sli/crossfire during actual game play notice a difference between 80 and 85fps if there even was a benefit. Jayz2cents on youtube had a video that's not terribly old that shows just what cpu bottle necking really is. I know linus also did a comparison on cpu cores for gaming but that video is a little old.

 

[jj109]

HT reduces single-threaded performance when two threads are scheduled on the same core, so DX11 games that are bottlenecked at the submit thread can see some performance regressions. Theoretically, OS schedulers should be smart enough to know which cores share hardware and try to avoid that situation.

 

[BrightCandle]

HT reduces single-threaded performance when two threads are scheduled on the same core, so DX11 games that are bottlenecked at the submit thread can see some performance regressions. Theoretically, OS schedulers should be smart enough to know which cores share hardware and try to avoid that situation.

They are. Windows chooses real cores in preference to HT ones until they run out, has done since Windows Vista. From my gamegpu.ru data I found just 2 games where the 2600k was outperformed by the 2500k at all. 3% slower in Red Orchestra 2 Rising storm and 5% slower in Warframe. In 55 games that is it, just those two games. On average HT was worth 5% extra performance, the third quartile was just 7% faster and the best improvement between those two CPUs was 22% (Medal of honour warfighter).

HT is a net positive in todays games, the problems of the past are done with.

 

[tidy]

.

 

[Galatian]

And that's exactly why at this current time that going with 6 or more cores is not worth it if all you do is play games. There's only a small handful of games that would benefit from it and you will easily be spending $300 more just for the CPU alone for that very minor benefit. For some people that $300 could be the difference between going with a 4790k or 5930k vs just picking up a 780ti over a 770 that would offer a much better bang for the buck. I realize that there are some enthusiasts on these forums that are always going to go with 2xTitans or 2x780ti and want the latest most expensive CPU to go with it but most people are not interested in spending that much on a gaming pc.



High res benchmarks have been linked before with high end cards running comparing 4 and 6 core cpu's and the difference being little to none. Someone will always come along and link the 1 of 10 games that actually benefit from it running at some low resolution showing a 10-20fps difference when both are already running like 300fps or something ridiculous. Realistically at high resolution there will at best be a few fps difference which wont be noticeable in actual game play and does not justify $300 difference. I doubt many people can sit here on the forums with sli/crossfire during actual game play notice a difference between 80 and 85fps if there even was a benefit. Jayz2cents on youtube had a video that's not terribly old that shows just what cpu bottle necking really is. I know linus also did a comparison on cpu cores for gaming but that video is a little old.

Completely agree. Also I'm one of those few who use Quicksync regularly, so it really would be a hard sell for me (more expensive, questionable performance in crease and no Quicksync).

I'm not saying there are no differences, but it's always the same with the high end: a huge mark up for a small increase in performance. It's the same with GPUs: the GTX 670 was the smarter buy compared to the GTX 680 and so is the R9 290 compared to the R9 290X. Yes there is more performance to be had, but you are purchasing those 5 - 10 % with a 50% increase in price.

 

[moonbogg]

HT reduces single-threaded performance when two threads are scheduled on the same core, so DX11 games that are bottlenecked at the submit thread can see some performance regressions. Theoretically, OS schedulers should be smart enough to know which cores share hardware and try to avoid that situation.

I don't think that's the case. What I understand about HT is this: When a CPU core is working and it is waiting for some information, it will do some other task as it waits, thus improving overall multithreading efficiency. It does not take away from the strength of the core and its main task that it is performing work on. The HT "core" is simply the CPU core putting its unused cycles to work on some other task as it waits. The non HT CPUs simply wait and do nothing while they wait. This may explain why a HT CPU produces more heat under heavy thread load, because its performing more work rather than sit around and waste potential.
This also explains why 6 core CPUs do better in new games than 4 core, HT CPUS. The threads of the game are not being expected to run on a CPU core's left over potential, rather the tasks are run on a full core, with full potential.

 

[tidy]

.

 

[skipsneeky2]

Made the move from a i5 2500 non k to a 3770 recently, very well worth it for BF4. Other games will most likely take advantage of it sooner or later when they catch up.

Had a 3930k setup back in January of 2013 to replace that i5 2500 non k. Never had a extreme chip before but found out it was a waste of money. Sold it off quickly but I think today some games like BF4 would beast on it better then even on my 3770.

 

[moonbogg]

What causes the 'negative scaling' then?

I don't know.

 

[TurtleCrusher]

I don't know.

I've never seen someone so honest here.

 

[jj109]

I don't think that's the case. What I understand about HT is this: When a CPU core is working and it is waiting for some information, it will do some other task as it waits, thus improving overall multithreading efficiency. It does not take away from the strength of the core and its main task that it is performing work on. The HT "core" is simply the CPU core putting its unused cycles to work on some other task as it waits. The non HT CPUs simply wait and do nothing while they wait. This may explain why a HT CPU produces more heat under heavy thread load, because its performing more work rather than sit around and waste potential.
This also explains why 6 core CPUs do better in new games than 4 core, HT CPUS. The threads of the game are not being expected to run on a CPU core's left over potential, rather the tasks are run on a full core, with full potential.

Speaking from a hardware perspective, Hyperthreading time-shares the core fetch/decode and presents it to the OS as two cores. The execution units of the out of order core are not aware that the frontend has been Hyperthreaded, so they always pick the oldest ready to execute instruction in the queue

So if both virtual cores are being scheduled to, it means that the second thread can use the execution units that the 'main' thread does not use. Two threads can run simultaneously on the core so throughput increases. But HT is not thread aware! To the execution engine there is no 'main' or 'secondary' thread. The execution order is first ready, first serve.

In the worst case (both threads require the same resources every cycle), single thread performance at the macro level is almost halved. The only saving grace here is that hyperthreading a core is preferrable to time-sharing the core through the OS b/c each context switch is around 100k CPU cycles.

The best case is that each thread never uses the same resources on the same cycle. In this case each virtual core performs like a physical core minus any bottlenecking at the decode or retire stage.

In my experience, Windows 8.1 has been much better at not stomping on an critical thread

 

[Ylurien]

The only posts that are really worthwhile throughout this entire thread are those made by people with the understanding that nobody should be buying CPUs based on how well they perform solely with today's games. All these recommendations for the i5 seem, to me, absolutely ridiculous given what we know about next-generation console specifications. How shortsighted to simply say, "Go with a 4670k (as general advice) —most games hardly even use that nowadays".

We have already seen evidence of games like Battlefield 4 (multiplayer), Watch Dogs (badly optimized probably, but still), and even the almost exactly 4-year old Starcraft 2 performing appreciably better with four or even more cores.

If all you ever play is minesweeper, no you won't need more cores. If you don't plan to play upcoming games, no you won't need more cores. But let me ask you: do you want to play upcoming big sandbox games? How about popular MMOs or multiplayer shooters with lots of players in the area at once? Realtime strategy games? The next installment in the Elder Scrolls series (OK, you might have time to upgrade again before that sees the light of day )? You are kidding yourself if you think four cores is not going to lose hard to six similarly-clocked cores in these scenarios.

This same argument came up around two years ago when I was contemplating what video card to buy. Almost everybody seemed to think 2 GB of VRAM would be sufficient, even overkill for games of that time. I ended up buying a 4 GB because I don't like to replace hardware every year. Do you know how happy I am now that I did that? Forget about playing at 2560x1440 with good looking textures in most games with anything less than 3 GB of VRAM. I feel sorry for people who paid in the ballpark of $750 for a 780 Ti card and they have to play at 1920x1080 even though they have a 2560x1440 because they start suffering horrible stuttering when maxing out graphics options.

It should definitely be said that the hardware a person buys should match what they intend to use it for. But let me just tell you - are you interested in the AAA games shown at E3? Games like the next Assassin's Creed or Destiny? Better give some serious thought to at least an i7 from the last few years if you want to get the most out of them.

The new consoles are finally here. Though it took eight long years, progress is finally being made. So let's stop pretending it isn't. Otherwise, you'll end up with a hardware upgrade you'll regret.

Having said all of this, this advice still stands: if you are rocking a Sandy Bridge or newer i7, the only thing you need to be thinking about for games is graphics cards. And start saving those pennies: people are saying Nvidia will drop its prices for the next round, but even a significant price drop still won't make them affordable! ;-b

 

[tidy]

.