Discussion Cinebench 2026

[Makaveli]

Created a little table with GPU scores. Trying to find scores at stock speeds. If you have one of the missing cells let me know!

Also, I used AI to help me fill in some of the specs so if you see something that is just wrong, that is why.

View attachment 136191

This application doesn't run on RDNA 3 where is that 60k score coming from?

 

[Hulk]

This application doesn't run on RDNA 3 where is that 60k score coming from?

Somewhere in this thread I found it?

 

[Sunaiac]

It does run on rdna3.
Here is an asrock phantom gaming out of the box clocks.

 

[lightmanek]

Here is my 7900XTX with slight OC on water:


10min Burn-in test.

EDIT: 9950X3D with -20CO and 220W PPT on water

 

[Geminiui]

What will be the result at 5.0 GHz on all cores?

 

[lightmanek]

What will be the result at 5.0 GHz on all cores?

I can try when I'm back from work.

5GHz all cores - fixed clock (real clock 5025MHz)



Normal Windows 11 Pro install with a ton of services in the background.

 

[Makaveli]

It does run on rdna3.
Here is an asrock phantom gaming out of the box clocks.

Hmm then I guess its just me maybe because i'm still on windows 10?

 

[Sunaiac]

Hmm then I guess its just me maybe because i'm still on windows 10?

I cannot test this theory anymore sorry.
All I can add is I'm on driver 25.8.1

 

[Makaveli]

I cannot test this theory anymore sorry.
All I can add is I'm on driver 25.8.1

You are on win 11?

 

[Sunaiac]

Yes, you can see the OS used in the captures.

 

[Aeonsim]

Windows 11 - Ryzen 9 7900 (@120PPT and 6000MT) & 9070 XT
GPU - 49009
CPU MT - 6646
CPU SC - 629
CPU ST - 478

MacOS 15.7.3 - M2 Pro 14"
GPU - 13491
CPU MT - 4028
CPU ST - 473

It seems to very much like NVIDIA and Apple GPU's compared to AMD GPUs.

AMD Z1 Extreme (Asus Rog Ally @ 30W) - Win 11
GPU - 4362 (ran in auto mode using 8GB of shared memory).
CPU MT - 3049 (~3.7Ghz all core)
CPU SC - 519
CPU ST - 403

 

[therealmongo]

Here is another 9950x3d
PBO == enabled no CO adjustments

 

[Z O X]

Here is another 9950x3d
PBO == enabled no CO adjustments
View attachment 136747

As a long time C4D user (since ver 9.5), it worries me a little that 9799/562 should be 17,43

 

[digitaldreamer]

As a long time C4D user (since ver 9.5), it worries me a little that 9799/562 should be 17,43

I see your point.
I had though it was doing internal rounding. But, even 9799.9/562.0 won't be more than 17.44 (17.4359), to say nothing about why it isn't rounding up the points.

 

[therealmongo]

good catch! never paid attention to the ratio. will need to run it on my 9800x3d to see if the ratio is "borked" also

 

[gregory_x86]

Cinebench 2026 doesn't work on chinese KX-7000 , the program show message that lack AVX2 instructions.




But benchmark works on my AMD 9800 (not Ryzen 9800X3D) It took 75 min to finish.

 

[burninatortech4]

Cinebench 2026 doesn't work on chinese KX-7000 , the program show message that lack AVX2 instructions.

View attachment 136889


But benchmark works on my AMD 9800 (not Ryzen 9800X3D) It took 75 min to finish.

View attachment 136890

Be careful posting about Bristol. You'll summon Nosta from his cave.

 

[Abwx]

Be careful posting about Bristol. You'll summon Nosta from his cave.

Not much risk, this CPU use bulk 28nm not FDSOI.

Beside that s the first AM4 compatible CPU, it s a first that this 2016 plateform allow roughly for one order of magnitude better throughput upgrade with a 5950X.

 

[burninatortech4]

Please add these to the spreadsheets:

7800XT (25.12.1 driver): 41271



5950X (-10 CO)

 

[Sunaiac]

I tried my PBO2 CO setup on my 5800x3D (-30 all cores, 95/60/90 PPT/PMH/EDC), my score didn't move in any useful manner (+3 pts lol)

Anyone tried nehalem with intel's AVX2 emulator for the lulz ?

 

[gregory_x86]

Be careful posting about Bristol. You'll summon Nosta from his cave.

Not much risk, this CPU use bulk 28nm not FDSOI.

Beside that s the first AM4 compatible CPU, it s a first that this 2016 plateform allow roughly for one order of magnitude better throughput upgrade with a 5950X.

A8-7680 Carrizo on FM2+ platform with DDR3 RAM. It achieve higher score than A12-9800 on AM4 with DDR4. Weird.
And whole platform is more efficent. But single score is higher on AM4 because A12 achive 4,2GHz.

For comparision.
FM2 platform : A8-7680 Stock(3,5-3,8 Turbo) 65W TDP + ASUS FM2 MB + 2x8GB DDR3 2133 + Nvidia T400 + SSD = 115 W Total in CB 2026
AM4 platform: A12-9800 Stock(3,8-4,2 Turbo) 65 W TDP + MSI B350 MB + 2x8GB DDR4 2400 + Nvidia T400 + SSD = 155 W Total in CB 2026

On both platform I use the same cooler AMD wraith prism and second fan to VRM on board.

 

[Abwx]

A8-7680 Carrizo on FM2+ platform with DDR3 RAM. It achieve higher score than A12-9800 on AM4 with DDR4. Weird.
And whole platform is more efficent. But single score is higher on AM4 because A12 achive 4,2GHz.

For comparision.
FM2 platform : A8-7680 Stock(3,5-3,8 Turbo) 65W TDP + ASUS FM2 MB + 2x8GB DDR3 2133 + Nvidia T400 + SSD = 115 W Total in CB 2026
AM4 platform: A12-9800 Stock(3,8-4,2 Turbo) 65 W TDP + MSI B350 MB + 2x8GB DDR4 2400 + Nvidia T400 + SSD = 155 W Total in CB 2026

On both platform I use the same cooler AMD wraith prism and second fan to VRM on board.

View attachment 136969

The 9800 had a cut down cache compared to its Steamroller predecessor, eventualy it became cache latency sensitive and perhaps that DDR3 lower latency vs DDR4 helped compensate for this shortcoming in latency sensitive tests.

Beside the 9800 was pushed with excessive voltage, so much that the 35W 9800E was close in perfs to the regular 9800 and wasnt far behind an Athlon 200GE wich is also 35W.

Anyway nice vintage CPUs here, Excavator had a broad ISA support, with FMA3/FMA4/AVX and even AVX2, also you re lucky as not all FM2+ MBs actualy supported the A-8 7680, that was a hit and miss.

 

[Sunaiac]

I tried my PBO2 CO setup on my 5800x3D (-30 all cores, 95/60/90 PPT/PMH/EDC), my score didn't move in any useful manner (+3 pts lol)

Anyone tried nehalem with intel's AVX2 emulator for the lulz ?

So, this Nehalem question kept me from sleeping.
Therefore, I tried on windows 11 but it never started (no message). I thought it was for some lack of instruction (I'm dumb like that) but no start even with SDE.
Moving to Sequoia 15.7.1, tada, CPU works (but not GPU, error message I can't read because all tiny in render window).

So here is the answer to the burning question : is Nehalem still the best CPU arch ever ?
And the answer is of course still yes.

Ahem, anyway, here's the score :
- Single thread : 145 pts
- Single core : no can do
- Multi core : 2141 pts (14,77x)

Configuration :
- Mac pro 5.1 2012
- 2x x5690 3.46GHz (for a total of 260W of pure raw performance)
- 6x 16GB DDR3 1333 (3 per CPU)
- Radeon VII
- Sequoia 15.7.1 on OCLP 2.4.1
- SSD SATA 3

Windows is on the same machine, tiny 11 with drivers 25.8.1, but alas, no start.
I'll update if I manage to get some windows or Radeon VII scores

 

[Det0x]

GPU bench seems really light, barely use above 600w
196882 points
View attachment 135802

Time to show result of system
Hardware used:

• Delidded 9850X3D
• ASUS x870e ROG crosshair APEX
• TG DD mycro pro waterblock @ mora rad on a different floor (with 4 pumps in series)

Not sure if this benchmark scales with memory tuning or not

 

[Joe NYC]

Time to show result of system
Hardware used:

• Delidded 9850X3D
• ASUS x870e ROG crosshair APEX
• TG DD mycro pro waterblock @ mora rad on a different floor (with 4 pumps in series)

Not sure if this benchmark scales with memory tuning or not
View attachment 137512

Is that the highest Single Core score?