Processor GFlops Compilation

[trickson]

Originally Posted by newtekie1

Table Updated.

Very nice score on that Q9650 Trickson! Amazing what 775 is still capable of

Thank you .
I can push a tad more out of it but need get get below ambient temps to do it . 4.5GHz was the best but not stable with current cooling .

[Mussels]

Originally Posted by newtekie1

The one you are using that scored 50.60 is 4 cores and 4 threads. With 4 cores and 8 threads I got 49.39.

... why would i care about 8 threads on 4 cores? i was only concerned with the overall effect of HT... and i was expecting a boost, just not a large one.

a speed DROP makes me wonder wtf HT is even good for

[newtekie1]

Originally Posted by Mussels

... why would i care about 8 threads on 4 cores? i was only concerned with the overall effect of HT... and i was expecting a boost, just not a large one.

a speed DROP makes me wonder wtf HT is even good for

That is my point, you said 60.60 / 8 = 15.15 per thread, it should be 50.60 / 4 = 12.65. The correct score for the 4 threads w/out HT is 50.60, you used 60.60, and you are dividing by 4 threads but you put 8.

When you said you were dividing by threads, since you put 8 in your equation, I assumed you meant to use the 8 threads w/out HT score.

I think HT is better for applications that handle a lot of different data. But in an application that purely crunches large numbers, HT hurts more than helps. I wonder if this would lead to better F@H numbers with HT off...

[Mussels]

Originally Posted by newtekie1

That is my point, you said 60.60 / 8 = 15.15 per thread, it should be 50.60 / 4 = 12.65. The correct score for the 4 threads w/out HT is 50.60, you used 60.60, and you are dividing by 4 threads but you put 8.

When you said you were dividing by threads, since you put 8 in your equation, I assumed you meant to use the 8 threads w/out HT score.

I think HT is better for applications that handle a lot of different data. But in an application that purely crunches large numbers, HT hurts more than helps. I wonder if this would lead to better F@H numbers with HT off...

it was 2am dude :P

my concern is that stupid games/apps may not know the difference between the cores, and a dual threaded app may run on a regular core and its HT core... and give worse performance than running with HT off, without the user/folder/cruncher knowing.

[newtekie1]

Originally Posted by Mussels

it was 2am dude :P

my concern is that stupid games/apps may not know the difference between the cores, and a dual threaded app may run on a regular core and its HT core... and give worse performance than running with HT off, without the user/folder/cruncher knowing.

Yeah, I know, I'm not blaming you, just clarifying.

I believe, in theory, no core is more real than the other as it appears to Windows. It is just that each physical core can handle 2 threads, so each thread isn't more real than the other. So work sent to each thread is automatically handled by the physical core's scheduler.

[Mussels]

Originally Posted by newtekie1

Yeah, I know, I'm not blaming you, just clarifying.

I believe, in theory, no core is more real than the other as it appears to Windows. It is just that each physical core can handle 2 threads, so each thread isn't more real than the other. So work sent to each thread is automatically handled by the physical core's scheduler.

game needs/uses two threads worth - and they end up on the one physical core, giving you less performance than if they had two physical cores each. that is my concern.

[newtekie1]

Originally Posted by Mussels

game needs/uses two threads worth - and they end up on the one physical core, giving you less performance than if they had two physical cores each. that is my concern.

Yeah, I can see that as a concern for sure.

Though I think the scheduler on the CPU is smart enough to assign threads to unused cores first, which is why when I run it with only 4 threads w/ HT enabled I get pretty close to the same score as I get with HT disabled.

[Athlonite]

from what i understood about HT is that each single thread called by software gets divided by the thread dispatcher on core into two if HT is enabled so if software calls for 2 treads it gets divided by four

[somebody]

Originally Posted by newtekie1

Though I think the scheduler on the CPU is smart enough to assign threads to unused cores first, which is why when I run it with only 4 threads w/ HT enabled I get pretty close to the same score as I get with HT disabled.

Just FYI I ran my 860 at the same clock as your 875k to make a comparison.



The first run was with HT and 8 hardware threads with 4 software threads selected in iBurn / OS. It seems clear that 2 software threads are running on one core at times judging by the results.

The second run was with HT and 8 hardware threads with 4 software threads selected by me so as to only have one software thread per core. Results speak for themselves.

Mussels makes a very valid point about HT, that is poor or random hardware thread selection can possibly result in a performance degrade with HT.

Originally Posted by Athlonite

from what i understood about HT is that each single thread called by software gets divided by the thread dispatcher on core into two if HT is enabled so if software calls for 2 treads it gets divided by four

A single software thread may run on 1 hardware thread and then after a context switch run on another hardware thread but not both hardware threads at the same time.

[Mussels]

thanks somebody, you made me feel better about going AMD over i7... i wanted more real cores, and now it seems i'm justified.

[DuDu]

Hello guys.This is my first post on TPU

DuDu | Intel | Q9450 3,70ghz | 46.92734 | 12.68306486 | 64-bit | Ram 1734mhz

[Athlonite]

Originally Posted by Mussels

thanks somebody, you made me feel better about going AMD over i7... i wanted more real cores, and now it seems i'm justified.

it would seem you have an justified complaint about HT then mussels maybe the results suggest then to just turn HT off and do without it as it looks like it's faster without than with

[Athlonite]

well i just did a little test to see what would happen if i forced 4 threads on my Athlon x2 7750BE vs forcing only 2 threads

4 threads


2 threads


and yes I know AMD's don't have HT but look at the 4 thread results it starts out with an hiss and a roar then slows down considerably vs the 2 thread result which stay relatively static

[newtekie1]

Table Updated.

Originally Posted by Athlonite

it would seem you have an justified complaint about HT then mussels maybe the results suggest then to just turn HT off and do without it as it looks like it's faster without than with

Yes, in this single benchmark only, but in many other real world uses HT does help significantly.

[Athlonite]

I'd test it in other places if I could but not being an Intel quad or better with HT owner/user I can't

[Pembo210]

Pembo210 | Intel | Core i5-750 @ 3.6GHz | 050.44122 | 14.01145 | 64-bit



These are my new results. I recently upgraded to a Corsair H50 cooler and moved my i5 from 3.3GHz to 3.6GHz. My old numbers were 045.21908 gflops and 13.70275 gflop/GHz

[newtekie1]

Table updated.

[gja822]

gja822 | Intel | Celeron 723 (Penryn) @ 1.20GHz | 003.28018 | 02.7401 | 32-bit

Some mobile Intel processor to "slow-end" of table. And yes, IntelBurn is for Intel optimized (at least compared to my AMD ).

[newtekie1]

Table Updated

[Frick]

Frick | AMD | AMD Sempron 140 @ 3.5GHz | 10,09238 | 2,883537142857143 | 32-bit | Single core

The CPU is not unlocked.

[MoonPig]

MoonPig | Intel | Core2 Quad Q9550 @ 3.399 | 046.13596 | 13.5705 | 64bit

http://img.techpowerup.org/100731/3400.png

MoonPig | Intel | Core2 Quad Q9550 @ 4.037 | 052.16536 | 12.9218 | 64bit

http://img.techpowerup.org/100731/4037.png

[Jstn7477]

Jstn7477 | AMD | Phenom II X4 955BE @ 3.210 | 038.4461 | 12.0144 | 64-bit

[newtekie1]

Table Updated

[Gabkicks]

here is what I am running @ now Hyperthreading is off!!

[Frick]

I think ya need to add some tables there mon. Like this:

Jstn7477 | AMD | Phenom II X4 955BE @ 3.210 | 038.4461 | 12.0144 | 64-bit