39. The dual channel Athlon 64 3800 is starting to distance itself with a 4% lead over the single channel Athlon 64 3400 - still far below the PR difference!

UT2K4 - 1600*1200
ATHLON 64 3800 NEWCASTLE - SOCKET 939
2026-05-19 20:46:37 - UT2004
Frames: 7522 - Time: 60000ms - Avg: 125.367 - Min: 90 - Max: 175

ATHLON 64 3400 NEWCASTLE - SOCKET 754
2025-09-22 04:42:49 - UT2004
Frames: 6353 - Time: 60000ms - Avg: 105.883 - Min: 64 - Max: 156

Only at the extreme 1600*1200 resolution does the socket 939 system manage to leave the 754 one behind by a significant margin! Time to move on to the first Directx9 game to be tested - FARCRY!

40 & 41 & 42. FARCRY - I used hardware oc farcry benchmark tool that works perfectly and its results can be easily reproduced by anyone!
ATHLON 64 3800 - SOCKET 939

FAR CRY
ATHLON 64 3400 - SOCKET 754

Farcry performs slightly better on Athlon 64 3800 - the keyword here is "slightly"! Until now there is no result to justify the PR of socket 939 CPU'S with the exception of the memory scores in PCMARK!

43. Next game is Half life 2, one of the best games of all times. I tested the original version and not the modern Steam version which is far more tougher on a system of this era.
Because HardwareOC half life 2 benchmark tool refused to work I used again Fraps 3.4.7 in order to test a 60 seconds interval in the first level in front of the train station where Dr Breen is making his infamous speech. I waited 1 minute before pressing F11 and I walked around the monument; after fraps finished i changed the resolution and waited 1 minute before testing again and so on. Repeated 3 times, the last runs are the actual results that I kept. Tested with sound!
Again tested at max settings - this game is very optimized and poses no problem for the X1950PRO, not even at the extreme 1600*1200 we are not GPU-bound and we still get an average of 60+ FPS.

1024*768
ATHLON 64 3800
2026-05-22 12:53:55 - hl2
Frames: 5331 - Time: 60000ms - Avg: 88.850 - Min: 66 - Max: 122

ATHLON 64 3400
2026-01-05 06:13:50 - hl2
Frames: 5142 - Time: 60000ms - Avg: 85.700 - Min: 62 - Max: 123

44. HALF LIFE 2 - 1280*1024

ATHLON 64 3800
2026-05-22 12:52:10 - hl2
Frames: 5268 - Time: 60000ms - Avg: 87,800 - Min: 64 - Max: 115

ATHLON 64 3400
2026-01-05 06:12:05 - hl2
Frames: 4993 - Time: 60000ms - Avg: 83.217 - Min: 58 - Max: 107

45. HALF LIFE 2 - 1600*1200

ATHLON 64 3800
2026-05-22 12:50:25 - hl2
Frames: 4464 - Time: 60000ms - Avg: 74.400 - Min: 56 - Max: 96

ATHLON 64 3400
2026-01-05 06:10:25 - hl2
Frames: 4246 - Time: 60000ms - Avg: 70.767 - Min: 46 - Max: 95

HAL LIFE 2 perform around 5% better on the dual channel system. With a stronger CPU such as FX-55 or FX-57 you would get even better minimums and averages would exceed 100fps at 1024*768 and 1280*1024!

46. And we reached the only OPENGL game tested: DOOM 3. I recently started the game again with a creative x-fi card installed in my socket A go-to system and selected nightmare - of course I became stuck pretty fast!
I tested at ultra quality but with AA disabled using the builtin benchmark:

A. Athlon 64 3800 - socket 939
B. Athlon 64 3400 - socket 754

47. DOOM 3 - 1280*1024
A. Athlon 64 3800 - socket 939
B. Athlon 64 3400 - socket 754

47. DOOM 3 - 1600*1200
A. Athlon 64 3800 - socket 939
B. Athlon 64 3400 - socket 754

3-4% better than the single channel counterpart does not bode well for the Newcastle 3800!

49. SUPERPI - where a shorter time constitutes a better result is next!
A. Athlon 64 3800 - socket 939
B. Athlon 64 3400 - socket 754

And finally we got a difference of 9% - much closer to the PR difference of these 2 processors than the measly 4-5% that we got until now!

50. After the welcome boost provided by SUPERPI it is time for the CINEBENCH 2003 and to take a look at the CPU scores!
A. Athlon 64 3800 - socket 939
B. Athlon 64 3400 - socket 754

Disappointing performance by socket 939's Athlon 64 3800! Cinebench cares about frequency and IPC, nothing else!

51. And we reached the final test: TOM2D! Venice did not provide the correct combination for TOM2D to "work" in the previous round of tests, let's see if Newcastle is up to it!
A. Athlon 64 3800 - socket 939
B. Athlon 64 3400 - socket 754

Again dual channel makes no difference; actually socket 939 scores worse than its cheaper counterpart!

Time to get the results into perspective! Athlon 64 3400 Newcastle for socket 754 gets a 100% base rating and Newcastle 3800 for socket 939 is compared to it!

1. ATHLON 64 3800 - NFORCE4 - SOCKET 939: 104.44%
2. ATHLON 64 3400 - NFORCE4 - SOCKET 754: 100%

The difference between these 2 CPU's make the PR of the 3800 unrealistic! There should be only 200 PR points between the 2 Newcastle for socket 939, respectively 754! So AMD did inflate the ratings! Taking into account the previous tests where Venice 3500 for socket 939 was compared to Venice 3200 for socket 754 and the results showed the exactly same thing,it stands to reason to conclude that: DUAL CHANNEL = 200 PR points!

We can also compare the results between the 2 socket 939 CPU's tested so far: Athlon 64 3800 and Athlon 64 3500. 3800 is 4.83% faster than 3500 which also indicates a 200 PR points difference between these 2 models and NOT the 300 points that AMD suggested! Please note than even with the high performance nforce4 compared to the K8T800 PRO used for the 3500, the 3800 can not distance itself enough to justify those 300 points that exist between them!
However the above comparison is between 2 different Athlon 64 as 3500 has the Venice core and 3800 has the Newcastle one! This begs the question: are all revisions equal? Does a later released core have superior performance to an early one? Is, for example, San Diego better than Clawhammer?

Lancaster provided about 4% boost in CPU benchmark over Clawhammer at the same frequency in Re: Battle of the platforms: socket 754! . San Diego should be the same.

Have you tested with msi afterburner, fraps and everest/aida running/recording in the background? I got around 5% better score when I disable all of them: Re: Battle of the platforms: socket 754!
I started testing 6 of the 7 revisions and they perform neck in neck!

I didn't have any background apps.

On screenshots of PC Mark I mostly see Irfanview, Total commander, Everest, those should not influence the score much. There are many system processes normally running. It depends on what they do.