Sign in with
Sign up | Sign in
Your question

Counting transistors??

Last response: in CPUs
Share
April 4, 2008 3:28:07 AM

Many of you may have already come across this chart http://www.intel.com/pressroom/kits/quickreffam.htm and are aware of its many mistakes. My questions is does anyone have confirmation of exactly how many transistors the QX9650 has opposed to the Q9450?....yes I did try emailing Intel, by I think the janitor replied.

More about : counting transistors

April 4, 2008 6:20:36 AM

They are supposed to have identical number of transistors. I believe it is a typo on Intel's part. As you can see on the "intro date", the Q9xxx series still bear the launch date of Jan-08, while its realistic launch date is somewhere in March.

What was Intel's response?
April 4, 2008 6:42:09 AM

Well Intel actually confirmed the 820 million transistors on the QX9650 and only 410 million for the Q9450. The Duo's also have 410 million, how does that work considering the 4 cores of the Q9450 and 12Mb cache? Sorry for anyone who just bought a Q9450 thinking the multiplier was the only compromise.
Related resources
April 4, 2008 6:43:01 AM

What difference does it make? If that is true (link?), then it does seem like a lot. But the QX9650 is $1030 and the q9450 is only $380. I would like to see benches with both chips at the same clock speed before I get in a tizzy about any transistor discrepancy.
April 4, 2008 7:11:53 AM

Sure maybe it doesn't matter if your running notepad, but another 410 million transistors is a god send with 4 cores loaded to the hilt in a complex 3D enviroment.
April 4, 2008 7:26:33 AM

QX9650
Q9550
Q9450
X3360
X3350

Have approximately 820 million transistors.

Q9300
E8200
E8400
E8500
E3110

Have approximately 410 million transistors.

April 4, 2008 8:04:24 AM

Hmm Intel are saying only 410 million for the Q9450 and Q9550.
April 4, 2008 8:19:24 AM

Like I said, I'll believe it when I see the benches.
April 4, 2008 8:24:13 AM

Vertigon said:
Sure maybe it doesn't matter if your running notepad, but another 410 million transistors is a god send with 4 cores loaded to the hilt in a complex 3D enviroment.


Actually it doesn't matter that much, if Intel has streamlined its transistor budget.

Also, Q9300 would definitely have less working transistor than Q9650 due to smaller L2 cache.
April 4, 2008 10:25:20 AM

I would like to see benchmarks with the CPUs set to the same, or near same, frequencies, to take the frequency differential out of the equation.

To illustrate my reasoning, I set my Q6600 to the specs in the CPU-Z pic.



The 3DMark06 bench in your link were as follows

Q6600 @ 2400 1066 FSB 3528
Q9450 @ 2660 1333 FSB 3951

My Score
Q6600 @ 2664 1333 FSB 4215



I ran the 3DMark06 bench with my Q6600 at 3G/1333 and got a score of 4660.

That puts it about halfway between these two CPUs.

QX9650 3.0G/1333 4445
QX9770 3.2G/1600 4853

How is this possible?

April 4, 2008 11:12:49 AM

Well considering your oc'd Q6600 is near the stock Q9450 clock and that your oc'd cpu rates 300 points higher might suggest transistors are significant. Hard to determine since the Q6600 has 172m more transistors on it than the Q9450. Refer to my original post http://www.intel.com/pressroom/kits/quickreffam.htm
April 4, 2008 11:24:26 AM

Well something is screwy. I guess there is really no reason to spend any more money than the price of the Q6600.
April 4, 2008 11:42:31 AM

Depends how you look at it, you might be a well paid engineer, physicist etc, dropping an extra $800 on a machine your not going to need to upgrade for 2 years might be a bargain and a dam convenient idea. If your a kid in school working at the local convenience store and using his comp to play Halo and download porn, keeping the $800 in your pocket might be a better idea. The tragedy is in wanting to need it but not needing it. Also in an era of multiple cores every improvement is an inprovement x2, or 4, soon x6 and x8. I love you world!!!! Huggggsss!!!!!
April 4, 2008 12:00:13 PM

All Q9xxx have 2x 410M transistors as they are made from two 410M dies (a Q9450 is made from two E8200 dies (with 410M transistors each)).
I'm not sure about the Q9300 but it's still two of those dies.
It has half the cache (either disabled due to defects or it's different dies but I don't think they make 2 different dies).

As for the 3D Mark scores ... they are strange (old code I guess).
a c 127 à CPUs
April 4, 2008 1:56:37 PM

Um guys.... Am I the only one who finds it weird that they also list it a s a "Core 2 Extreme Quad Processor Q9450/Q9550/Q9300"?

And the Q6600 is showing that it has a 1333MHz FSB when it really has a 1066MHz FSB.

Obviously this is either old or just all wrong.
April 4, 2008 2:05:34 PM

The Pentium D 9xx had 376 million transistors for two cores. Wow, that must have been one beast. Close to the 410 million on full fledged Wolfdales.
April 4, 2008 5:08:12 PM

trassistor number is not important for compare two generation cpu, single current transistor may get job down which may required two or more old transistors.
April 4, 2008 5:31:40 PM

Andrius said:
All Q9xxx have 2x 410M transistors as they are made from two 410M dies (a Q9450 is made from two E8200 dies (with 410M transistors each)).
I'm not sure about the Q9300 but it's still two of those dies.
It has half the cache (either disabled due to defects or it's different dies but I don't think they make 2 different dies).

As for the 3D Mark scores ... they are strange (old code I guess).

Do they use some double cheesburger glue to put them together?
Quad cores are defective cpu's?
:whistle: 
April 4, 2008 5:51:00 PM

The Pentium D 9xx had HyperThreading support (basicly two cores in one) even if HT was only enabled for Extreme Editions I think. It also had a lot of L2 Cache for those times and it had a silly long pipeline so all of that accounts for the extra transistors (190M per die).
April 4, 2008 6:05:44 PM

ctbaars said:
Do they use some double cheesburger glue to put them together?
Quad cores are defective cpu's?
:whistle: 


The dies Intel makes Q9300s out of are probably the same dies they make other Q9x50 quads out off but they are binned (for whatever reason, most likely they failed the L2 cache test or they have to high a TDP) as only having half the L2 cache.

Cheeseburger glue might work for roaches and brainlobes ... but firing up two hot chips dipped into it presents a serious fire hazard :non:  :pt1cable: 
April 4, 2008 8:50:16 PM

No comments on my bench results? Anyone have any Ideas?
April 4, 2008 9:39:24 PM

Zorg said:
No comments on my bench results? Anyone have any Ideas?

Are you using the same gfx card? ram? (including latency) mobo?
Even a different psu might give a change.
There are too many variables to make any kind of reasonable conclusion.
April 4, 2008 10:08:48 PM

endyen said:
Are you using the same gfx card? ram? (including latency) mobo?
Even a different psu might give a change.
There are too many variables to make any kind of reasonable conclusion.
Well I don't believe that. I ran the same bench with the settings in the pic and got a score of 4207, which is lower but in the wind, so it's not RAM or FSB speed. My VGA is a X1950 Pro, nothing special there, and this is supposed to be a CPU test not a VGA/CPU test. As far as the PSU, if it is sufficient then it will have absolutely no affect on the bench. I would hope that the mobo wouldn't cause the massive differences. I don't see the mobo used, but I hope they didn't use a POS. Either way I would like to know, so that I can avoid it.

Any other variables that might cause my Q6600 to show so well? I even have the CPU under volted, C1E/EIST is disabled. What you see in the CPU-Z screenshot is what you get.

I just noticed that he was running :lol:  Vista :lol:  If it is the OS that is causing the disparity, then it really sucks worse than I thought. :pfff:  :pfff: 


April 4, 2008 10:15:05 PM

In 3D Mark 2006 CPU test there is no real difference between the QX9650 and QX6850 (both 3.0GHz so the architectural changes and the increased L2 cache are not important, likely due to it being an older benchmark). Both score around 4350 on a P35 with 800MHz RAM.

Your test is "an anomaly in unbalanced equation" and without proper means of controlling the variables we can't say anything conclusive.

EDIT: Maybe it's the fact that your FSB is 443MHz insted of 333MHz.
I've ran some Sandra benchmarks (CPU INT+FLOAT) a few days back and got scores above the QX6850 and QX9650 which seemed unlikely.
All three CPUs have the same clock, so a QX9650 should be around 10% better in arithmetics.
April 4, 2008 11:00:45 PM

Andrius said:
In 3D Mark 2006 CPU test there is no real difference between the QX9650 and QX6850 (both 3.0GHz so the architectural changes and the increased L2 cache are not important, likely due to it being an older benchmark). Both score around 4350 on a P35 with 800MHz RAM.

Your test is "an anomaly in unbalanced equation" and without proper means of controlling the variables we can't say anything conclusive.

EDIT: Maybe it's the fact that your FSB is 443MHz insted of 333MHz.
I've ran some Sandra benchmarks (CPU INT+FLOAT) a few days back and got scores above the QX6850 and QX9650 which seemed unlikely.
All three CPUs have the same clock, so a QX9650 should be around 10% better in arithmetics.
My test is "an anomaly in unbalanced equation"? :lol: 

I ran the Bench at Q6600 @ 2400 1066 FSB and got 3786 compared to their 3528.
If you look again you will see that the test at 443 in my second bench shows a lower score by 7 points 4215 ->4207. That's going the wrong way and is in the wind.

At any rate, I should have run the test above to get a base line, oops. As I said earlier it's not the RAM or FSB. It's not the PSU and not the VGA. It's probably not, or at least shouldn't be, the mobo. I guess that leaves Vista.

Vista, what a piece if DRM laden pigsh!t.

All hail the king...XP that is. :lol: 
April 4, 2008 11:08:25 PM

A Matrix quote a day makes 7 a week. :lol: 

If your Q6600 scores 4350ish at 3.0GHz (9x333MHz) the other result is an anomaly. I don't have 3D Mark as I think it's bloatware as is s(h)i(t)soft Sandra.

The only benchmark that matters to me is "the perception of waiting for the system to complete it's task".
April 5, 2008 5:49:43 AM

Andrius said:
A Matrix quote a day makes 7 a week. :lol: 
Which quote is that? I liked Matrix 1 & 2, but 3 left me high and dry.
Andrius said:
If your Q6600 scores 4350ish at 3.0GHz (9x333MHz) the other result is an anomaly. I don't have 3D Mark as I think it's bloatware as is s(h)i(t)soft Sandra.
My result at that frequency was 4660 not 4350ish but it's nice to see that you are paying attention. What other result is an anomaly and why, be specific.?
Andrius said:
The only benchmark that matters to me is "the perception of waiting for the system to complete it's task".
Certainly that would be important, but not very objective.

Like I said, I blame Vista for the reasons addressed above.

Don't matta.
April 5, 2008 7:51:15 PM

Zorg said:
Which quote is that? I liked Matrix 1 & 2, but 3 left me high and dry.My result at that frequency was 4660 not 4350ish but it's nice to see that you are paying attention. What other result is an anomaly and why, be specific.? Certainly that would be important, but not very objective.

Like I said, I blame Vista for the reasons addressed above.

Don't matta.


The Matrix quote is from Revolutions, from Neo's dialogue with "the Architect".

Your 3D Mark 06 Results are around 300 points over those in Toms CPU Charts (the 4350ish @ 3.0GHz from the QX6850 and QX9650). I didn't really catch what OS they used. If they used Vista it would explain the difference. I never used 3D Mark so it's scores are meaningless to me.

I know that my Q6600 @ 3GHz scores 50kish MIPS and 38kish MFLOPS in SiSoft Sandra which is consistent with a QX6850.
But a few days ago it scored almost 60k MIPS and 45k MFLOPS as I was checking the results against a friends QX6850.
It freaked him out and it took 15 tries to get 3 accurate benchmarks.
Why his QX6850 scores lower than the QX6800 in Sisoft Sandra is still a mistery.

EDIT : My Q6600 @ 3GHz scores around 4380 (repeated the test several times) on default 1280x1024 settings and Windows XP so your result remains a mistery to me (unless it's EIST (I have it enabled and you had it disabled) but I'll check that tomorrow).
April 5, 2008 8:20:39 PM

Vertigon said:
Many of you may have already come across this chart http://www.intel.com/pressroom/kits/quickreffam.htm and are aware of its many mistakes. My questions is does anyone have confirmation of exactly how many transistors the QX9650 has opposed to the Q9450?....yes I did try emailing Intel, by I think the janitor replied.

I'm 99% sure that they have the same number of transistors. They are in fact the same chip, but the QX9650 is clocked higher and has an unlocked multiplier.

I also believe that the Q9300 has the same transistor count as the rest of the Q9XXX series processors, but half of the L2 cache is disabled on each die, E6300 style.
April 5, 2008 11:24:45 PM

Zorg said:
No comments on my bench results? Anyone have any Ideas?


We don't know their methodology, the hardware used, as well as other variables. Unless we can obtain their testing machine specification, we can never duplicate the result.
April 6, 2008 2:55:57 AM

yomamafor1 said:
We don't know their methodology, the hardware used, as well as other variables. Unless we can obtain their testing machine specification, we can never duplicate the result.
I understand that, but a CPU test, if it is of any value, should not be dependent on these other variables. It should run like prime95 small FFTs. I understand that technically you should only compare CPUs that are benched with the same HW, but the results should not be that far off. I don't have anything special in my machine. I'm guessing Vista is the problem, and that is very disappointing.
April 6, 2008 3:28:25 AM

Andrius said:
The Matrix quote is from Revolutions, from Neo's dialogue with "the Architect".

Your 3D Mark 06 Results are around 300 points over those in Toms CPU Charts (the 4350ish @ 3.0GHz from the QX6850 and QX9650). I didn't really catch what OS they used. If they used Vista it would explain the difference. I never used 3D Mark so it's scores are meaningless to me.

I know that my Q6600 @ 3GHz scores 50kish MIPS and 38kish MFLOPS in SiSoft Sandra which is consistent with a QX6850.
But a few days ago it scored almost 60k MIPS and 45k MFLOPS as I was checking the results against a friends QX6850.
It freaked him out and it took 15 tries to get 3 accurate benchmarks.
Why his QX6850 scores lower than the QX6800 in Sisoft Sandra is still a mistery.

EDIT : My Q6600 @ 3GHz scores around 4380 (repeated the test several times) on default 1280x1024 settings and Windows XP so your result remains a mistery to me (unless it's EIST (I have it enabled and you had it disabled) but I'll check that tomorrow).
I guess these benchmarking programs don't mean that much. It's kind of frustrating. I don't have any problems with the speed of my machine. As you said, "the perception of waiting for the system to complete it's task", it works for me. Screw the meaningless benches.
April 6, 2008 4:29:45 AM

Zorg said:
I understand that, but a CPU test, if it is of any value, should not be dependent on these other variables. It should run like prime95 small FFTs. I understand that technically you should only compare CPUs that are benched with the same HW, but the results should not be that far off. I don't have anything special in my machine. I'm guessing Vista is the problem, and that is very disappointing.


Yeh. I found another benchmark (Q9300) from Xbitlabs. They also have lower 3DMark06 in Vista than yours in Windows XP, and on par with Hothardware's result.

So I guess it is Vista's problem.
April 6, 2008 2:29:00 PM

Zorg said:
I guess these benchmarking programs don't mean that much. It's kind of frustrating. I don't have any problems with the speed of my machine. As you said, "the perception of waiting for the system to complete it's task", it works for me. Screw the meaningless benches.


So I retested with EIST and Enhanced Halt State (C1E) Disabled today.
My CPU score was 4683 (10 run average) (vs 4380 with EIST disabled and C1E Enabled).

I guess we now have the culprit. C1E is to blame for the 6% performance reduction :whistle: 
As for the "perception of performance" it still only scores 2fps in the 3DMark CPU test :lol: 
I've also just noticed that my passivly cooled 8600GT appears to "whistle" while switching from 2D to 3D mode.
I don't game much so I must say I never noticed it before (Supreme Commander doesn't make it "whistle").

WAY TO GO 3D Mark :pfff:  ... A new FEATURE for my 6 month old hardware !!!
April 6, 2008 8:57:12 PM

There shouldn't be any difference with EIST and C1E enabled or disabled. I thought C1E was enabled on the last test because I usually have it enabled, but I think I might have forgotten to enable it. I will set it to stock with C1E enabled and see what I get.
April 6, 2008 10:36:06 PM

I ran the bench with C1E enabled @ 2400 1066 FSB and got 3770 instead of 3786, 16 points are in the wind IMO.

With C1E and EIST enabled @ 2400 1066 FSB I got a score of 3789, which is higher than my initial score of 3786.

It appears that C1E and EIST have absolutely no affect on the CPU score.


Go figure. :pt1cable: 
a b à CPUs
April 6, 2008 11:08:36 PM

Hmm... Intresting. I would assume that the extra transistors might have some thing to do with the unlocked multi on the QX9650, it is weird though, since both are based on the same Yorkfield core design.
April 6, 2008 11:22:14 PM

Zorg said:
I ran the bench with C1E enabled @ 2400 1066 FSB and got 3770 instead of 3786, 16 points are in the wind IMO.

With C1E and EIST enabled @ 2400 1066 FSB I got a score of 3789, which is higher than my initial score of 3786.

It appears that C1E and EIST have absolutely no affect on the CPU score.


Go figure. :pt1cable: 


Well EIST was disabled in both my tests.
I tested at 3.0Ghz (10 repeats, CPU tests 1&2 only).
It appeared to be the anwser as my score was 300 points over those from yesterday
(and within a percent of your 4660 score) and all I did was disable the C1E setting in BIOS.
Maybe it was the fact that I rebooted the system after 3 days :lol:  .
The anomaly from the unbalanced equation strikes again.

I'll retest everything tomorrow.
April 6, 2008 11:35:42 PM

Shadow703793 said:
Hmm... Intresting. I would assume that the extra transistors might have some thing to do with the unlocked multi on the QX9650, it is weird though, since both are based on the same Yorkfield core design.


The unlocked multiplyer is probably a limit register (ROM) value of 25 (or 99.9) insted of the actual number for the other models. If we think back to the days of the original Athlon the same function was performed by a jumper (contact pins) array with some connections missing.

The transistor count of all Qx9xxx should be the same and twice that of all E8xxx.


a b à CPUs
April 7, 2008 12:44:19 AM

^ Ok, now I see it.
April 7, 2008 3:18:07 AM

Andrius said:
The transistor count of all Qx9xxx should be the same and twice that of all E8xxx.

Yes
a c 127 à CPUs
April 7, 2008 2:28:10 PM

Andrius said:
So I retested with EIST and Enhanced Halt State (C1E) Disabled today.
My CPU score was 4683 (10 run average) (vs 4380 with EIST disabled and C1E Enabled).

I guess we now have the culprit. C1E is to blame for the 6% performance reduction :whistle: 
As for the "perception of performance" it still only scores 2fps in the 3DMark CPU test :lol: 
I've also just noticed that my passivly cooled 8600GT appears to "whistle" while switching from 2D to 3D mode.
I don't game much so I must say I never noticed it before (Supreme Commander doesn't make it "whistle").

WAY TO GO 3D Mark :pfff:  ... A new FEATURE for my 6 month old hardware !!!


I had a feeling that since 3DMark06 goes in and out of the game it doesn't fully utilize the CPU when C1E and EIST are enabled. Thats why when I disabled that and OC'ed my GPU I got like 12320. Its interesting how when using Vista people automatically attribute it to Vista when Vista is quite fast and I have yet to see a performance drop yet in TF2, C&C3, Hell even Crysis runs near as fast as XP now.
April 7, 2008 2:39:47 PM

Andrius said:
So I retested with EIST and Enhanced Halt State (C1E) Disabled today.
My CPU score was 4683 (10 run average) (vs 4380 with EIST disabled and C1E Enabled).

I guess we now have the culprit. C1E is to blame for the 6% performance reduction :whistle: 


I've retested both settings (both around 5 minutes after rebooting).
Q6600 @ 3.0GHz (9x333MHz, 1.24375V) 10 repeats
C1E Enabled : I get a score of 4711
C1E Disabled: I get a score of 4714
(Everytime I test I get a higher score :lol: ).
Indeed the C1E setting has no effect (as should be the case). :bounce: 

I attribute my previous lower result and hence conclusion (or error) to several days of system uptime.



a b à CPUs
April 7, 2008 3:36:17 PM

Im calling Intel Conspiracy on this, now! :lol: 
April 7, 2008 5:35:54 PM

Shadow703793 said:
Im calling Intel Conspiracy on this, now! :lol: 


I think my Q6600 mutated into QAI.
It's getting better at benchmarks every time I run them :lol: 
Intel conspiracy!
April 7, 2008 6:04:46 PM

I got 4660, so 4714 isn't that much higher, about 1%. Pretty much in the wind. Are you using Vista or XP? If you already said, sorry I forgot.

I see you have XP SP2, So we still haven't ruled out Vista as the cause.

It must be a Microsoft conspiracy.
a c 127 à CPUs
April 7, 2008 6:38:12 PM

Zorg said:
I got 4660, so 4714 isn't that much higher, about 1%. Pretty much in the wind. Are you using Vista or XP? If you already said, sorry I forgot.

I see you have XP SP2, So we still haven't ruled out Vista as the cause.

It must be a Microsoft conspiracy.


I have Vista. I will pull my results pre SP1 and Cata 8.3 when I get home and post them then do a run with SP1 and Cata 8.3 just to see what happens.
April 7, 2008 6:52:19 PM

Cool, I was hoping someone would do that.
a c 127 à CPUs
April 7, 2008 6:54:42 PM

Zorg said:
Cool, I was hoping someone would do that.


Luckily I also have a Q6600 G0 @ 3GHz running off of the 333MHz FSB.
!