Dual vs. Quad-Core + Hyperthreading Analysis *(TBG)*

[wand3r3r]

There's an interesting analysis regarding dual vs. quad core and HyperThreading! Another solid analysis by The Tech Buyers Guru! http://www.techbuyersguru.com/CPUgaming.php

3DMark Fire strike clearly loves hyperthreading in its Physics score. +1 for hyperthreading
What I find interesting is that correlation between threads and cores is practicly linear! 3DMark clearly knows how to best utilize threads on these Intel processors.

Quite unlike the Graphics Score, the Physics Score is a nearly ideal test of CPU processing power - the 3770K scores almost exactly twice as high as the 3220 at the same clock speed (6287 vs. 3097 without Hyperthreading), showing that there is perfect core scaling in this benchmark. Interestingly, Hyperthreading appears to add the equivalent of one additional core to both processors, increasing the score in this benchmark by 49 percent on the i3-3220 and 42 percent on the 3770K.

For the graphics however, the cores and threads appear to make no difference (from 2-4 cores with/without HT) and clockspeed is likely the only factor. Perhaps it's single or dual threaded and that's it.

BF3 single player

HyperThreading has a clear effect on the dual core, strange results on the quad core.

BF3 Multiplayer

HyperThreading on the dual core is sure spreading it's wings. The 3770k is clearly better having true cores vs. extra thread capabilities.


Far Cry 3

Here's the article, there's more games and better analysis, it's definitely worth the read and is very well written imo (even worth the read for referring friends and family who are trying to get the barebones and cheap setups ).
http://www.techbuyersguru.com/CPUgaming.php

----
So, have you guys tried disabling HT? Thoughts?
I wonder what the effect of the next gen consoles will have on this, the situation could change soon?

I certainly will make sure to recommend dual cores with hyperthreading as an absolute minimum. I never knew the results were that dramatic. It's funny it's so helpful for a dual core, they are clearly underpowered.

 

[alcoholbob]

HT slows down most games because most games are designed to run on 3 cores--the lowest common denominator of multiplatform games, aka the xbox360.

Next gen console games will be running on 6+ cores standard so a quad core will be better off with HT on in future products.

 

[moonbogg]

That BF3 test was GPU bound...just sayin.

 

[BrightCandle]

http://forums.anandtech.com/showthread.php?p=35080439

Take a look at the 2600k verse 2500k results. We can clearly see that on average HT has a decent positive impact. It does so on all but 1 game.

 

[Endymion FRS]

That BF3 test was GPU bound...just sayin.

Single player is, multiplayer is a much different story. A strong CPU helps a ton in BF3 multi.

 

[toyota]

never mind

 

[2is]

http://forums.anandtech.com/showthread.php?p=35080439

Take a look at the 2600k verse 2500k results. We can clearly see that on average HT has a decent positive impact. It does so on all but 1 game.

Not so clear when you also factor in the 2600K is 100MHz faster and more cache. Not saying HT does or doesn't help, just that whatever improvement you see, at least some of it comes from the i7's other advantages. The method in this thread is a more accurate way to isolate the effects of HT.

 

[wand3r3r]

Not so clear when you also factor in the 2600K is 100MHz faster and more cache. Not saying HT does or doesn't help, just that whatever improvement you see, at least some of it comes from the i7's other advantages. The method in this thread is a more accurate way to isolate the effects of HT.

Agreed, the methodology is quite good in the OP link. It's pretty accurately demonstrating the dual vs quad and hyper threading with interesting results!

 

[BrightCandle]

If what you do is take a 3770 and turn off the HT you do get to compare the HT in isolation, but it doesn't really help inform a buying decision on processors, for that you need a comparison of a 3570 verses a 3770 because their are other changes apparent in the CPUs. The 3770 without HT is academic only because its not a CPU you can actually buy.

In the gamegpu results (which has a LOT more games) we see an average of 5% benefit, and the extra clock speed is worth about 3%. Comparatively just one game is slower on HT and all others are at least as quick if not faster. The mass of data is what makes the gamegpu.ru data interesting, because it doesn't focus on a very small subset of games, it looks at every major release in the last 6 months. Overwhelmingly the data shows HT is a net positive beyond the clock speed in all but 1 game.

So while this review might be in depth what it misses is suitable breadth to inform a buying decision well, its misleading for showing a data point that is unique and not represented in any other games. You can't extrapolate that HT is bad from their results, because that doesn't agree with the overall data at all.

 

[chimaxi83]

BF3 single player to test CPU performance? I guess that's for the lulz

Are there any games that actually make use of HT?

 

[toyota]

BF3 single player to test CPU performance? I guess that's for the lulz

Are there any games that actually make use of HT?

a few but Crysis 3 is the only shooter I am aware of. http://maldotex.blogspot.nl/2013/02/hyperthreading-and-real-custom-graphics.html

 

[BrightCandle]

Are there any games that actually make use of HT?

Yes quite a few (about 25% of games). Going through the gamegpu.ru game data for the last 6 months (49 games in all) produces the following summary:

Hyperthreading improvement seen

Average        102%
Maximum        118%
25% Quartile   101%
75% Quartile   104%
Minimum         92%

The games that seem to show the most improvement are

Medal of honor Warfighter    118%
Far Cry 3 Blood Dragon	     116%
Crysis 3 MP Beta	     116%
Battlefield 3 end game       107%
Need for speed most wanted   107%
Starcraft 2                  105%
Far Cry 3 Aftermath          105%
Project cars                 105%
Guild Wars 2                 105%

gamegpu.ru don't have a 2600k with HT turned off so I have estimated the impact by adjusting the 2500k clock speed up to match (linearly) the 2600k and then compared those results. Its not quite right, in some cases the 2600k cache will be giving a boost more than just the clock speed difference and in other cases the linear increase in performance is way to much and will show too little HT gain attributing some of it to clockspeed. Thus some results will show less impact of HT, some more than they ideally should with a proper test. But knowing the limitations of the data it still shows us some interesting results.

The 2600k v 2500k summary for example is the following comparison which is more useful if the question is "Should I get a 2600k/3770k/4770k processor or the i5 equivalent":

Average			105%
Maximum			122%
25% Quartile		104%
75% Quartile		107%
Minimum			95%

 

[Lepton87]

Your methodology is flawed you compare CPUs and that's it. It's widely known that SC2 uses only 2 cores yet you think it benefits from 8 threads?

8 cores Xeon at 3.1 is faster then 6 cores at 3.5GHz thus it must benefit from more cores. It's the same fallacy. In reality this game is not well threaded at all it is just a cache hog, how do you know that sc2 is different? Either you compare CPUs or the benefit of HT, they did it right you did it wrong. The only good point you made is that they tested too little games. And it's not only academical either, everyone is free to turn HT off on their i7s.

 

[wand3r3r]

If what you do is take a 3770 and turn off the HT you do get to compare the HT in isolation, but it doesn't really help inform a buying decision on processors, for that you need a comparison of a 3570 verses a 3770 because their are other changes apparent in the CPUs. The 3770 without HT is academic only because its not a CPU you can actually buy.

In the gamegpu results (which has a LOT more games) we see an average of 5% benefit, and the extra clock speed is worth about 3%. Comparatively just one game is slower on HT and all others are at least as quick if not faster. The mass of data is what makes the gamegpu.ru data interesting, because it doesn't focus on a very small subset of games, it looks at every major release in the last 6 months. Overwhelmingly the data shows HT is a net positive beyond the clock speed in all but 1 game.

So while this review might be in depth what it misses is suitable breadth to inform a buying decision well, its misleading for showing a data point that is unique and not represented in any other games. You can't extrapolate that HT is bad from their results, because that doesn't agree with the overall data at all.

Well it's clear hyper threading has mixed results. True the 3770k doesn't perfectly emulate a 3570k (I should get the specs for comparison) but its clear that the effect of enabling vs disabling HT is accurate on the same processor at the same clocks. The dual core clearly benefits widely.

Why not take HT as you can disable it if you find a game doesn't benefit from it. I'm actually quite interested in the threading once next gen console ports hit and are used to 8 cores (or however many ms/Sony allocate). Maybe then 4 cores + HT will be ideal or at least far more beneficial.

 

[wand3r3r]

Your methodology is flawed you compare CPUs and that's it. It's widely known that SC2 uses only 2 cores yet you think it benefits from 8 threads?

8 cores Xeon at 3.1 is faster then 6 cores at 3.5GHz thus it must benefit from more cores. It's the same fallacy. In reality this game is not well threaded at all it is just a cache hog, how do you know that sc2 is different? Either you compare CPUs or the benefit of HT, they did it right you does it wrong. The only good point you made is that they tested too little games. And it's not only academical either, everyone is free to turn HT off on their i7s.

Interesting that cache made it appear more cores were helping. Then again the 6 core 3970 was behind the 3770k.

 

[Madrox]

By a very , very narrow margin. Maybe a margin of error.

 

[wand3r3r]

Interesting that cache made it appear more cores were helping. Then again the 6 core 3970 was behind the 3770k.

By a very , very narrow margin. Maybe a margin of error.

Presuming you mean the comment above?

Yeah they are easily within the margin of error. It does indicate the additional cores are not helping to say the least.

 

[BrightCandle]

The way I look at this question is really should someone buy an i3, i5 or i7 for gaming. Isolating the impact of HT is a nice testing mechanism but it doesn't really inform a buying decision because you can't buy a processor without HT that has the cache of the i7 processor, its purely academic. The actual choice is between real CPUs with real differences, HT, cache and clock speed. So while you can't isolate the HT, you don't really need to if the purpose is to determine which CPU is better for gaming.

Now of course I like the next guy am interested in whether a game benefits from cache, clockspeed or HT, that I think is an interesting question. But then what I really want is a review that looks at all three aspects.