Giga Flops, lol

Nov 26, 2005
15,194
403
126
So I recently hit 3.81Ghz with my UT3 rig with HT off. I ran LinX and it reported 55.3926 GFlops

I turned on HT. Ran LinX and it dropped to 48.xxxx GFlops

Why did it flop so much?

EDIT: does it need a little Viagra, lol. No, but seriously, why, what, how, ... ???
 
Last edited:

Idontcare

Elite Member
Oct 10, 1999
21,110
64
91
It's reported correctly, it really is lower. Resource contention FTL.

Corei79204GHzwithHT.png
 

Idontcare

Elite Member
Oct 10, 1999
21,110
64
91
Basically what it looses in GFlops it makes up for it in threads?

Yeah this is one of those rare apps where hyperthreading actually causes enough resource contention that performance suffers.

It will be interesting to see how BD performs in this app.
 

Cerb

Elite Member
Aug 26, 2000
17,484
33
86
Edit: IDC: AT graph collector and reply ninja.
 
Last edited:
Nov 26, 2005
15,194
403
126
Just noticed. With HT off, on a full 20 player server of UT3 Greed, there is a noticeable lag compared to being at 3.49Ghz and HT on.
 

Idontcare

Elite Member
Oct 10, 1999
21,110
64
91
Just noticed. With HT off, on a full 20 player server of UT3 Greed, there is a noticeable lag compared to being at 3.49Ghz and HT on.

Most apps you are going to use as a desktop consumer are going to see a performance benefit from hyperthreading.

(That is why they price the models w/HT higher than the models w/o HT. ;) :D)
 
Dec 30, 2004
12,553
2
76
Just noticed. With HT off, on a full 20 player server of UT3 Greed, there is a noticeable lag compared to being at 3.49Ghz and HT on.

math apps are ones that make extremely efficient use of transistors. Any context switching is thus directly taking from "performance" aka work that could have been done.
 

toyota

Lifer
Apr 15, 2001
12,957
1
0
let me know your gamer name... i'll be on tomorrow.. pm me if you want
well I am horrible player with little interest in multi but I used to get on there just to play around. the only reason I even have UT 3 was because it was on sale and it uses bots so I could kill some time when I did not have a good single player game to play. I cant even remember my gamer name but once I get it installed I will figure it out and PM you with it.
 

toyota

Lifer
Apr 15, 2001
12,957
1
0
I was finally able to get my original "trek554" account working in UT 3 again after over an hour of nonsense with gamespyid. BTRY B 529th FA BN, that greed sever was not up but I played around on a couple of other servers. one was 14 people and the other was 20. I had ZERO lag of any type. I even disabled one core of my E8500 and stayed pegged at the framerate cap basically the whole time and it was smooth as can be. so if you are lagging it is 100% not because of HT being off on your i7. and really if the game required 8 threads to be lag free, 90% of people would be lagging their butts off.
 
Last edited:
Nov 26, 2005
15,194
403
126
If you had an option to run my clocks with HT On or Off, which would you choose?

With HT on, i get 48GFlops
With HT off, I get 55GFlops

or do the flops even flip floppin count? lol
 

Cerb

Elite Member
Aug 26, 2000
17,484
33
86
You should learn about what HT is and does, and not worry about synthetic benchmarks. The more efficiently a thread can utilize the CPU, the worse it will perform with HT.
 
Nov 26, 2005
15,194
403
126
You should learn about what HT is and does, and not worry about synthetic benchmarks. The more efficiently a thread can utilize the CPU, the worse it will perform with HT.

is that based on a single thread app? if its a multi threaded optimized app, then the answer would be with HT on, right

4 physical
4 logical
my old P4 3.06 Northwood 533 RDR system used hyper-threading.. so i kinda am familiar with what it is, just not how programers write their code, from app to app
 

Syzygies

Senior member
Mar 7, 2008
229
0
0
It's reported correctly, it really is lower. Resource contention FTL.

That graph is wrong.

The Core-i7 920 has four physical cores, eight virtual cores. The dots fit exactly what I see, making parallel computations on a Core-i7 2600K.

I use the programming language Haskell for parallel programming on various machines. The language itself has a bit of a learning curve, but one can parallelize a program with a handful of extra lines of code. It is particularly sensitive to resource contention, known as the "last core slowdown." On a machine with eight physical cores, performance scales nicely up to seven cores, and craters if one uses all eight cores.

With four physical cores, eight virtual cores, one still gets worse performance using eight threads over seven. Performance scales nicely up to four threads, with a bit of a hill for 5, 6, or 7 threads.

In other words, virtual cores are a scheduling convenience, that allows one to use all four cores of a four core processor. It's a second order effect, even if a rather significant one.

So I believe the dots; I recognize their pattern. LinX looks like it's implemented in a less efficient language for parallel computations, but the appearance of the graph is qualitatively the same as I see with other problems.

I question the two lines. They don't apply to virtual cores.