CPU Hyperthreading and Gaming: Real-World Benchmarks

[guskline]

Termie: Once again outstanding post.

BTW was just in your fair city of Washington DC for the weekend with my wife. We watched our 11 week old grandson while our youngest daughter and our son-in-law were at a conference at the Washington Hilton. We will be back down there this upcoming Friday thru the Weekend as our oldest daughter and our other son-in-law attend a conference at the Omni. We get to watch our 16 mo old grandson. Love it.

Now if GrandPa could teach them how to build computers!

 

[toyota]

FRAPS run for Deus Ex at 1024x768 with oced gtx560 se and 2500k at 4.4. there is no way HT can 4 cores help when at just 1024, even 4 cores offers nothing more than 3 cores. minimums can vary from run to run and 4 cores does not offer ANY advantage over 3 so again how could HT help on top of 4 cores?

2 cores
Frames, Time (ms), Min, Max, Avg
5533, 48937, 72, 125, 113.064

3 cores
Frames, Time (ms), Min, Max, Avg
5958, 49546, 88, 125, 120.252

4 cores
Frames, Time (ms), Min, Max, Avg
5792, 48516, 83, 125, 119.383


and here is Just Cause 2 at just 1024x768. I can run these over and over and over and its only within a tiny margin that 3 cores has any advantage over 2 cores and 4 cores does NOTHING more than 3 cores. yet you somehow say that HT is helping 4 cores? that makes no sense when 4 cores is not doing anything more in the first place.

2 cores
Frames, Time (ms), Min, Max, Avg
7696, 90637, 67, 109, 84.910

3 cores
Frames, Time (ms), Min, Max, Avg
8039, 90745, 71, 113, 88.589

4 cores
Frames, Time (ms), Min, Max, Avg
7746, 88312, 71, 110, 87.712

even at a silly low resolution it shows that 4 cores of an i5 at 4.4 are doing nothing more than 3 or even 2 for the most part. not to mention that back when I had my E8500 that I got over a 50% gain going from gtx260 to a gtx470 which is how much faster a 470 is so there was no limitation at real world resolution even from the old E8500 in Just Cause 2.

 

[Cerb]

Hyperthreading would most likely shine with multitasking. Such as streaming your game to Twitch.tv, which would mean the processor needs to compress/encode the video in real time.

I'd bet that would also be a good case to use a Xeon, for more cache, too (sans overclocking much).

Games don't support HT or am I missing something

Yes, you're missing something. SMT presents the program with many cores, just like having many cores. The OS is generally the one responsible for scheduling them correctly. Only a handful of games can use more than 4 cores (most can't do that), leaving an i3 gaining more than an i7. While using more will get to be a bit more common, I doubt it will be common enough for expensive CPUs like i7s to worthwhile in general until HTM from Haswell starts being supported (lock elision could be a godsend to multithreaded games). Synchronization costs make expanding out to more CPUs a careful situation to consider. That's also something to consider with i7 CPUs with HT: nobody thinks they are worthless; just that they aren't worth getting a worse video card instead of, if the budget is relatively fixed.

Good stuff. Very good thread.

Yes HT is treated like cores but the advantage is 30 to 40 percent avg speed boost in HT apps.

Having 8 threads will be nice for DAW and Video Rendering. But if the software really isn't designed to take full advantage then that 30 to 40 percent can become 15 to 25 percent ,, estimate... gl

The problem is when it can be unpredictable. HT works well by assuming (usually right) that there is additional on-chip bandwidth to be used, and not using it is being wasteful (may be right or wrong). However, when you want constant very small latencies, even when it does increase throughput (it's rare that it doesn't, today), it can still cause higher maximum latencies.

So, you would definitely want to check with the vendor of your DAW software, and/or other users of it. Even if it increases maximum latencies, they only need to be less than some expected value, so with a fast enough CPU with SMT, it could end up better even with measurably much worse response times. FI, if you must stay within 50us, and HT takes you from always meeting 20us to always meeting 35us, then the throughput gains will probably outweigh anything else.

So I'm wondering if in cases where from a purely "gaming centric" perspective we might be led to conclude "older game engines...appear to respond negatively to hyperthreading" but if we load up that computer with the kinds of background processing tasks that we do end up loading up then do those older engines actually benefit from the presence of HT and CMT cores because they aren't competing for CPU cycles with Norton Antivirus or Windows Update service?

That's going to be a per-AV problem, as well, unfortunately. Avast!, Norton, and McAffee, I know, reduce performance not because they eat up too much CPU, but because they create locking IO on files, force buffer flushes, and in the case of Norton, almost constantly have disk IOs going (Norton and McAffee are in another league from Avast!, but I have confirmed some occasions where Avast! was blocking IO for 10+ms at a time). CPU threading--and cache--effects may be there, but it would be (a) somewhat arduous to properly test, and (b) the VFS and disk effects can be quite obvious, and also somewhat random. I've thought about trying to test it myself, on occasion, but I can find holes in my testing methods much easier than ways to plug them or control for them.

 

[blastingcap]

Nice post, thanks for hte info

 

[Termie]

FRAPS run for Deus Ex at 1024x768 with oced gtx560 se and 2500k at 4.4. there is no way HT can 4 cores help when at just 1024, even 4 cores offers nothing more than 3 cores. minimums can vary from run to run and 4 cores does not offer ANY advantage over 3 so again how could HT help on top of 4 cores?

2 cores
Frames, Time (ms), Min, Max, Avg
5533, 48937, 72, 125, 113.064

3 cores
Frames, Time (ms), Min, Max, Avg
5958, 49546, 88, 125, 120.252

4 cores
Frames, Time (ms), Min, Max, Avg
5792, 48516, 83, 125, 119.383


and here is Just Cause 2 at just 1024x768. I can run these over and over and over and its only within a tiny margin that 3 cores has any advantage over 2 cores and 4 cores does NOTHING more than 3 cores. yet you somehow say that HT is helping 4 cores? that makes no sense when 4 cores is not doing anything more in the first place.

2 cores
Frames, Time (ms), Min, Max, Avg
7696, 90637, 67, 109, 84.910

3 cores
Frames, Time (ms), Min, Max, Avg
8039, 90745, 71, 113, 88.589

4 cores
Frames, Time (ms), Min, Max, Avg
7746, 88312, 71, 110, 87.712

even at a silly low resolution it shows that 4 cores of an i5 at 4.4 are doing nothing more than 3 or even 2 for the most part. not to mention that back when I had my E8500 that I got over a 50% gain going from gtx260 to a gtx470 which is how much faster a 470 is so there was no limitation at real world resolution even from the old E8500 in Just Cause 2.

Thanks for doing this. To be clear, my results for Just Cause 2 were not that significant, and I would toss those out as within the margin of error if not for the other results.

I will retest Deus Ex tomorrow on the slower machine to see if I can repeat my results.

 

[SPBHM]

can any of you share the savegame from deus ex and a description and how you are testing?
I would like to do something similar with my i3,
if I remember well, just walking around the streets would give a good notion of CPU performance in this game...
also I would be curious to see the i7 being tested on witcher 2, the i3 was heavily bottlenecking my VGA, even with HT on (but the CPU usage on task manager was moderate).
not my video, but I tested in this area -spoiler alert-
http://www.youtube.com/watch?v=w_ydZOu2Tk8#t=8m30s
for me it was by far the worst during the game,



as for the test methodology, I did some tests to check the consistency, and the result was good, I saw more variation on Crysis 2 (particularly on max framerate), so I decided to do 3 runs for each configuration;

 

[Termie]

Termie: Once again outstanding post.

BTW was just in your fair city of Washington DC for the weekend with my wife. We watched our 11 week old grandson while our youngest daughter and our son-in-law were at a conference at the Washington Hilton. We will be back down there this upcoming Friday thru the Weekend as our oldest daughter and our other son-in-law attend a conference at the Omni. We get to watch our 16 mo old grandson. Love it.

Now if GrandPa could teach them how to build computers!

I'm sure if grandpa can build computers, he can teach too! Line up those grandkids at the next family gathering and show off your gaming rig! They'll want to grow up just like you!

 

[BTRY B 529th FA BN]

Great post, Termie!

 

[Termie]

Because I tested Deus Ex on both systems and there appeared to be a consistent benefit of HT in the game, I have conducted additional in-depth analysis of the game with 2 cores, 4 cores, with HT and without HT. The dual-core was the i7-860 at 3.65 GHz, which approximates some of the fastest i3 dual-cores on the market today, and the quad-core was again the i7-860 but at 3.2GHz. I ran the dual-core and quad-core tests at different speeds because dual-cores often run at a higher GHz than their quad-core counterparts. The number of active cores and the HT state on any i7 can be set within the BIOS. The GPU was a GTX460-768@850, which is equivalent to a ~$120 HD7770, by no means a high-end GPU today.

I ran two identical runs through Hengsha's streets immediately after being dropped off by the transport. I ran down one set of stairs, went all the way around the lower streets, and came up the other set of stairs. I did not interact with NPCs or engage enemies. I simply walked, and yes, I felt a bit like a rat in a maze, but I did it for science. Because there was absolutely no variation between any of the sets of two runs, I did not repeat any run for a third time. Interestingly, there was a significant CPU-intensive spot right below the landing platform, as I passed a chef frying food. Each time the GPU usage would come crashing down, except on the quad with HT enabled. More information is available in the spreadsheet below.

My results speak for themselves as to the benefits of HT:

Here is the data for Deus Ex:

The following screenshot demonstrates that the quad-core with HT was able to keep the GPU fully-loaded for the entirety of the test. This was not true in any of the other configurations I tested.

I realize there are some people in this thread, and more broadly in the computer enthusiast community, who do not believe that hyperthreading can help a quad-core in any current game. I have conducted all of this testing not because I had an agenda, but because I had a question that I wanted answered and believed there was a lack of information in the enthusiast community and on professional revivew sites. I've already shown that HT does not help in many games, particularly with a single GTX670 in BF3 multi-player, where going in I was sure I'd detect a measurable difference. I had no idea that Deus Ex of all games would be one that would conclusively show that HT has a benefit not just for dual-cores, but for quad-cores, even in a situation that should be GPU-limited. I invite others to try to support or disprove my findings, but unless you are running at real-world settings with and without HT, you are not running the same test, and thus your findings cannot be used to support or disprove the data above.

My findings are as follows: all things being equal, in modern games, HT will either have no negative impact or will have a significant positive impact, especially as to minimums, which determine the smoothness of the gameplay. The benefit of HT is more pronounced on slower CPUs than on faster CPUs and on dual-cores than on quad-cores. When building a computer, I believe HT is a good investment when taking a long-term perspective on the usage of the system, and in fact, as between a fast dual-core with HT and a slower quad-core without HT, I might suggest the faster dual-core as the more cost-effective solution. Finally, a dual-core without HT simply isn't equipped to handle modern games with even a ~$100 GPU, let alone a higher-end GPU. I would not recommend, for instance, the current Pentiums for a gaming system.

 

[Maximilian]

Nice thread, grats Termie!

For a long time the only thing ive seen showing HT doing anything for gaming was this table that gets thrown around various forums.

Its nice to see somthing more in depth and conclusive :thumbsup:

 

[Termie]

Nice thread, grats Termie!

For a long time the only thing ive seen showing HT doing anything for gaming was this table that gets thrown around various forums.

Its nice to see somthing more in depth and conclusive :thumbsup:

Thanks, Maximilian. Actually, the other graph that has been thrown around (including by me), is the following:

Comparing those two graphs, and reviewing the data that I've collected, I'd say that BF3 is only cpu-limited and thus only shows the benefits of HT when testing at lower settings that don't stress the GPU. On a modern system with a high-end CPU and GPU, HT doesn't matter. But once you throw in two modern GPUs (e.g., as shown here: http://forums.anandtech.com/showpost.php?p=33814087&postcount=174), or consider a future game like BF4, HT will likely again make a difference, just as it does at BF3 at lower settings or in a number of games on my i7-860 system.

 

[toyota]

I am still not seeing it anywhere else but from you. both the toms and techspot reviews indicate HT should make no difference with 4 cores in Deus Ex and my own testing shows that 3 cores delivers the same performance as 4 cores even at just 1024x768. heck 2 cores were giving 99% of what 3 or 4 cores could do in that game. same goes for Just Cause 2 from testing where really even more 2 cores there makes no difference in any test but yours. the fact that you even claim HT helped your quad in GTA 4 says something is off because again it's a fact that GTA 4 only uses 4 threads.

 

[Termie]

I am still not seeing it anywhere else but from you. both the toms and techspot reviews indicate HT should make no difference with 4 cores in Deus Ex and my own testing shows that 3 cores delivers the same performance as 4 cores even at just 1024x768. heck 2 cores were giving 99% of what 3 or 4 cores could do in that game. same goes for Just Cause 2 from testing where really even more 2 cores there makes no difference in any test but yours. the fact that you even claim HT helped your quad in GTA 4 says something is off because again it's a fact that GTA 4 only uses 4 threads.

I haven't seen any tests on Deus Ex quite like the ones I conducted. I test for averages and minimums, and obviously test for the effect of HT.

Here are the tests you mentioned:
(1) Toms Hardware (http://www.tomshardware.com/reviews/deus-ex-human-revolution-performance-benchmark,3012-7.html) - they test at low resolution, and don't even test a quad with HT. Furthermore, no minimums are provided. They make an inference about HT based on the poor performance of the Phenom six-core, but this is dubious at best, given how poorly that architecture performs generally.
(2) TechSpot (http://www.techspot.com/review/436-deus-ex-human-revolution-performance-test/page7.html) - they test at almost the exact same settings I used, but do not provide minimums, which is the area I saw improvement over and over again on two vastly different systems.

 

[know of fence]

So this topic caught my interest because I did similar tests, back with bf2, I chose to reply because you had the due diligence or perhaps levity to post your data. It's a lofty goal you set for yourself to disprove the internet. You should brace yourself for some disappointment just in case.

Well you did the testing and made the graphs, which is why people are encouraging you, I guess. Whether the results pan out or not, is not up to the person running the experiment. What is up to you is how you approach and present this stuff. The tech sites have an advantage here, because they try to keep and present methodology, observation and conclusions separate. Certain sites do better than others in that regard, while in your case:
I'm still in the dark about how you disabled HT.
Do you run tests on different PCs and compare incomparable results, now?
Why don't you just disable cores on one PC, it should be possible as far as I know.
You write: "The dual-core was the i7-860 at 3.65 GHz", does this mean you figured out how to disable cores, the i7 is obviously not a dual core. In case you figured it out why not do all testing on a single testing platform following the example set by review sites.

There is a certain meticulousness required in both testing and presenting your data, I like the graphs and tables, but I'm confused by your writing which jumbles expectations, observation, argumentation and mixes test platforms. You have to wait with conclusions as much as possible, this is where people can easily nail you, by simply pointing out an error in your logic.

Lastly you write: "Because there was absolutely no variation between any of the sets of two runs, I did not repeat any run for a third time." Picking results is cheating, besides you need at least 4 results (the more the better obviously) to make any kind of statement about variation, but I would hate to ask for more maze-runs before it's clear how you get to 2/2, 2/4 HT, 4/4, 4/8 HT and why.

 

[toyota]

I haven't seen any tests on Deus Ex quite like the ones I conducted. I test for averages and minimums, and obviously test for the effect of HT.

Here are the tests I've found:
(1) Toms Hardware (http://www.tomshardware.com/reviews/deus-ex-human-revolution-performance-benchmark,3012-7.html) - they test at low resolution, and don't even test a quad with HT. Furthermore, no minimums are provided.
(2) TechSpot (http://www.techspot.com/review/436-deus-ex-human-revolution-performance-test/page7.html) - they test at almost the exact same settings I used, but do not provide minimums, which is the area I saw improvement over and over again on two vastly different systems.

I didn't find anything on TechReport, which you mentioned.

sorry I meant techspot. well I tested minimums and with my cpu at 4.4 there was again NO advantage at all to having 4 cores over 3 or even 2 for the most part. if 4 cores provides no improvement over 3 cores even at a silly low resolution then how could HT possibly be of any benefit on a 4 core cpu in those cases?

 

[Termie]

So this topic caught my interest because I did similar tests, back with bf2, I chose to reply because you had the due diligence or perhaps levity to post your data. It's a lofty goal you set for yourself to disprove the internet. You should brace yourself for some disappointment just in case.

Well you did the testing and made the graphs, which is why people are encouraging you, I guess. Whether the results pan out or not, is not up to the person running the experiment. What is up to you is how you approach and present this stuff. The tech sites have an advantage here, because they try to keep and present methodology, observation and conclusions separate. Certain sites do better than others in that regard, while in your case:
I'm still in the dark about how you disabled HT.
Do you run tests on different PCs and compare incomparable results, now?
Why don't you just disable cores on one PC, it should be possible as far as I know.
You write: "The dual-core was the i7-860 at 3.65 GHz", does this mean you figured out how to disable cores, the i7 is obviously not a dual core. In case you figured it out why not do all testing on a single testing platform following the example set by review sites.

There is a certain meticulousness required in both testing and presenting your data, I like the graphs and tables, but I'm confused by your writing which jumbles expectations, observation, argumentation and mixes test platforms. You have to wait with conclusions as much as possible, this is where people can easily nail you, by simply pointing out an error in your logic.

Lastly you write: "Because there was absolutely no variation between any of the sets of two runs, I did not repeat any run for a third time." Picking results is cheating, besides you need at least 4 results (the more the better obviously) to make any kind of statement about variation, but I would hate to ask for more maze-runs before it's clear how you get to 2/2, 2/4 HT, 4/4, 4/8 HT and why.

If you knew anything at all about the BIOS on an Intel system, you would know that it's as simple as choosing the number of active cores and enabling or disabling HT. I'm writing for a sophisticated enthusiast audience. I expected that this audience would understand what I was doing, but I've added a note above to explain how I did this.

You're right that review sites have the benefit of a dozen systems to test on, but I haven't found them conducting in-depth analyses of hyperthreading. So I am adding to the body of available knowledge with the equipment I have available.

To continuously criticize my methodology without so much as the slightest clue as to how these CPUs work just baffles me. And now you want me to conduct four runs of each test even though there isn't even a 2% variation between my two runs? You're being ridiculous. Run your own tests if you care so much.

I know some people here find my results informative. If you don't, just unsubscribe.

 

[Termie]

sorry I meant techspot. well I tested minimums and with my cpu at 4.4 there was again NO advantage at all to having 4 cores over 3 or even 2 for the most part. if 4 cores provides no improvement over 3 cores even at a silly low resolution then how could HT possibly be of any benefit on a 4 core cpu in those cases?

I really do appreciate your input, Toyota, because you're taking the time to actually benchmark and add your results. Would you mind testing at the exact same settings I used (including resolution) and lowering your CPU multiplier to stock? This will almost exactly duplicate my test configuration. Then just compare two core versus four core.

As you can see in my more in-depth analysis, my initial finding of a much-improved average was not repeatable, but the minimums were. This may be very level-dependent, so perhaps I need to test a different map in Deus Ex. Again, I only present my findings to add to the amount of available data on core and hyperthreading use. As others post their findings, a clearer picture of what's really going on.

 

[dac7nco]

Hyperthreading helps a LOT if you're gaming and running background apps. I ran an i5-2400 in a gaming HTPC for a bit, and BF3 frame rates suffered when I was recording from my Ceton tuner. Games were on an SSD, the tuner recorded to it's own HD. When I replaced the i5 with a Xeon E3 (Sandy Bridge) with hyperthreading, doing both was no sweat. On a 3930K I have yet to find a limit (to what I need). I can run Handbrake, record a stream and game pretty well. I initially had Handbrake locked-down to using four cores, but even using six, W7 thread scheduling seems to manage the system well. I'm running two 32GB sets of Geil RAM, a Thermaltake 240mm AIO water cooler and a mix of Plextor, Seagate and Hitachi SATA and SAS drives.

Daimon

Edit: I don't understand how people get acceptable frame rates on BF3 using Sandy/Ivy dual-cores. Single player maybe? on the i5-2400 I would dip into the low 20s @ 1080p on big maps with a GTX 580.

 

[kernelc]

@Termie: Very good, interesting post.

The problem is when it can be unpredictable. HT works well by assuming (usually right) that there is additional on-chip bandwidth to be used, and not using it is being wasteful (may be right or wrong). However, when you want constant very small latencies, even when it does increase throughput (it's rare that it doesn't, today), it can still cause higher maximum latencies.

So, you would definitely want to check with the vendor of your DAW software, and/or other users of it. Even if it increases maximum latencies, they only need to be less than some expected value, so with a fast enough CPU with SMT, it could end up better even with measurably much worse response times. FI, if you must stay within 50us, and HT takes you from always meeting 20us to always meeting 35us, then the throughput gains will probably outweigh anything else.

This is right, but an ever worse problem is when processing two thread produce resource trashing on the shared structures (eg: cache, uop cache, ecc.): http://ixbtlabs.com/articles3/cpu/ci7-turbo-ht-p1.html

To reduce this problem, with any new generation Intel is trying to reduce the statically shared structures, using dynamically shared ones instead.

Regards.

 

[kernelc]

Other very interesting test can be read hear:
http://www.overclock.net/t/671977/hyperthreading-in-games

The bottom line probably is that, according to Termie's tests, on dual core processor Hyperthreading do a measurable positive difference in games performance.

On the other hand, with 4 real cores it has little positive/negative impact, depending on the title.

However, start doing something a little complex in the background, and any additional hardware thread become very useful. See here: http://techreport.com/review/23662/amd-a10-5800k-and-a8-5600k-trinity-apus-reviewed/12

Regards.

 

[Termie]

Other very interesting test can be read hear:
http://www.overclock.net/t/671977/hyperthreading-in-games

The bottom line probably is that, according to Termie's tests, on dual core processor Hyperthreading do a measurable positive difference in games performance.

On the other hand, with 4 real cores it has little positive/negative impact, depending on the title.

However, start doing something a little complex in the background, and any additional hardware thread become very useful. See here: http://techreport.com/review/23662/amd-a10-5800k-and-a8-5600k-trinity-apus-reviewed/12

Regards.

kernelc - thanks so much for the link to overclock.net. Looks like another user did a comprehensive review very much like mine. Ironically, he found that HT helped in DA:O and hurt in GTA:IV, while I found the opposite for each game. And that's despite the fact that in GTA:IV we used the exact same scene (the first drive to the cousin's house). I'm not sure what to make of this - hyperthreading's impact is certainly complicated. His results in Fallout 3 were almost identical to mine in Fallout:NV (slight negative impact of HT), and in every other game the effects of HT were essentially zero, except for Oblivion, which showed a significant positive impact of HT. I'd say this is very consistent with my findings. There are a few games where HT helps noticeably, a few where it hurts slightly (usually older games), and in most games, it has little impact.

One thing I'll add is that he conducted his test over 2 years ago. In that time, game engines have certainly advanced, and I've tested several new games here. They tend to show a slightly greater benefit of HT, and none of the games I tested that were released in the past two years showed negative scaling such as was found in the overclock.net forum thread.

 

[Fanatical Meat]

Termie any chance you could run a guild wars 2 test?

 

[Termie]

Termie any chance you could run a guild wars 2 test?

Unfortunately I don't have that game, and it looks like there is no demo available. The best I can do is point you to this test at Tom's Hardware: http://www.tomshardware.com/reviews/guild-wars-2-performance-benchmark,3268-7.html

The dual-core with HT is doing little better than the dual-core without HT, which is pretty surprising given that the quad does so much better.

 

[know of fence]

Please don't take my criticism personally. If it helps I will add to your data a column with mean deviation and standard deviation for every average result you get, If you share your spreadsheet on any filehoster. If the HT difference is bigger then the latter, then you got meaningful results. There is little to be learned from min and max, they can only be used as an indication.

And yes 4 is the accepted sample minimum, simply because getting 2 similar results is relatively likely e.g. 50% for a coin flip experiment. The chance is 50% that you won't get the full range of possible results if you only toss a coin twice, which may lead someone to the false conclusion that the coin always shows 'tails'.) While the risk of getting 4 similar results in a row is already pretty unlikely (6.25%).

I have some experience with experimentation and analysis, but not as much with disabling cores/HT. Surely our goal should be to share the things we know to inspire others, rather than let our egos get in the way, while throwing wild theories and graphs at each others heads.
I'm certain that randomness is the key to understanding why HT seems to "help" in some games and doesn't "help" in others, especially in games that are essentially designed for a single core console.

Also just because you started this topic, doesn't mean I have to let things slide and unsubscribe. I tried to be polite, but that probably made my replies sound more dick'ish than they would have been straight-up.

 

[Termie]

Please don't take my criticism personally. If it helps I will add to your data a column with mean deviation and standard deviation for every average result you get, If you share your spreadsheet on any filehoster. If the HT difference is bigger then the latter, then you got meaningful results. There is little to be learned from min and max, they can only be used as an indication.

And yes 4 is the accepted sample minimum, simply because getting 2 similar results is relatively likely e.g. 50% for a coin flip experiment. The chance is 50% that you won't get the full range of possible results if you only toss a coin twice, which may lead someone to the false conclusion that the coin always shows 'tails'.) While the risk of getting 4 similar results in a row is already pretty unlikely (6.25%).

I have some experience with experimentation and analysis, but not as much with disabling cores/HT. Surely our goal should be to share the things we know to inspire others, rather than let our egos get in the way, while throwing wild theories and graphs at each others heads.
I'm certain that randomness is the key to understanding why HT seems to "help" in some games and doesn't "help" in others, especially in games that are essentially designed for a single core console.

Also just because you started this topic, doesn't mean I have to let things slide and unsubscribe. I tried to be polite, but that probably made my replies sound more dick'ish than they would have been straight-up.

Thanks for your input. If I conduct additional tests, I will do more than 2 runs. My Deus Ex data is consistent enough, however, that I truly doubt a third run would have changed anything.

While I admit I'm not following the strictest of testing methodologies, I have been open about my data, which is more than I can say for most professional reviews. I certainly wouldn't call my theories or graphs "wild." If they've given others something to think about, I think I've done the job I set out to do.