Discussion Cinebench 2026

[SPBHM]

I ran this on my RTX 3050 it got 25510 which I thought was impressive because it's not super far from the 2070 super from the graph,
also on the CPU in MT it's a 4ghz 6c/12t "coffee lake h" (Chinese BGA to LGA thing) running with DDR3 it got 1777, I didn't have enough patience for SC/ST

 

[SPBHM]

My office HP AIO PC with AMD Mendocino with 4 x Zen2 and 16GB 64-bit LPDDR5 running Windows 11 25H2.

My benchmarking is running fine; Zen2 is considered low-end by today standard and yet no issues.



View attachment 135916

I have a comparison!
Skylake e3 1270 v5 (full load is 3.7GHz for MT I think, ST can go up to 4.0) with DDR3 1600
MT 1129
SC 481
ST 268

 

[Markfw]

well, they won't send my return email so I can download it, so no bench's from me.

 

[SPBHM]

I downloaded it on techspot, they didn't ask for my email!

 

[Markfw]

Well, I don't know how long it will run until it stops, but looks like I will get 35,000 out of the CPU and 108,000 out if the GPU. How many passes does it make before it stops ?

 

[511]

Well, I don't know how long it will run until it stops, but looks like I will get 35,000 out of the CPU and 108,000 out if the GPU. How many passes does it make before it stops ?

Have you removed the run 10min option in the advanced settings otherwise it would run for 10min

 

[Markfw]

Have you removed the run 10min option in the advanced settings otherwise it would run for 10min

I have not touched the advanced settings. It runs over 10 minutes for sure.

 

[SPBHM]

the default settings are 10 minutes and over that until it's finished

 

[Markfw]

It finished ! The same as I said above 34.5k CPU and 108K GPU (4080 super)

 

[Fjodor2001]

It finished ! The same as I said above 34.5k CPU and 108K GPU (4080 super)
View attachment 135989

Says "AMD Eng Sample". Just wondering where you got hold of that beast?

 

[Markfw]

Says "AMD Eng Sample". Just wondering where you got hold of that beast?

ebay. Its a 9755 QS

 

[511]

ebay. Its a 9755 QS

Need luck 😅

 

[Hulk]

If anyone want to get some data on some cpus not in the table I posted that might be fun? I would need the singe core and single threads scores along with the average clock during the run. If you watch in HWinfo you can get a good idea of the clockspeed. Or you can lock it down with a simple registry edit and use Power Plan CPU max frequency setting.
We have Zen 5 and Zen 3, Zen 3+ and M1 Pro with a pretty good amount of confidence I think.

Isolating the Intel hybrid P's and E's can be tricky but the results would be interesting...

Again, to compare throughput ("IPC") of these cpus we need average clock during the run.

 

[511]

Again, to compare throughput ("IPC") of these cpus we need average clock during the run.

Going to be a problem in laptop

 

[Hulk]

Going to be a problem in laptop

Setting the max frequency with Power Plan works perfectly when there aren't hybrid cores. But even my HX370 is considered "hybrid" by Windows as it won't lock them down at 3GHz even though I know they can all run that frequency. But I know my old Surface Laptop will stay locked at a given frequency if it's on wall power.

The hybrid cores are always hard to "corner."

But we're the "best of the best" in here. I'm sure some of our experts will figure a way to obtain some good data.

 

[511]

Setting the max frequency with Power Plan works perfectly when there aren't hybrid cores. But even my HX370 is considered "hybrid" by Windows as it won't lock them down at 3GHz even though I know they can all run that frequency. But I know my old Surface Laptop will stay locked at a given frequency if it's on wall power.

The hybrid cores are always hard to "corner."

But we're the "best of the best" in here. I'm sure some of our experts will figure a way to obtain some good data.

I have a way just it's too painful .... and the benchmark can bug out just pin the cores when you see the White square box

 

[Hulk]

I have a way just it's too painful .... and the benchmark can bug out just pin the cores when you see the White square box

Yeah, benching is a time vampire.
The best way to isolate is to shut down cores in the BIOS and fix frequency. But all that rebooting and waiting,...

Wait, I'm not following "pin the cores when you see the white square box"

Are you referring to setting affinity?

 

[Z O X]

Yeah, benching is a time vampire.
The best way to isolate is to shut down cores in the BIOS and fix frequency. But all that rebooting and waiting,...

Wait, I'm not following "pin the cores when you see the white square box"

?

when rendering starts, pin the threads ...

 

[Hulk]

This is kind of interesting. IPC or throughput decreases as frequency increases. If I had to guess I would say that since memory access time remains constant as CPU frequency decreases there is more time for the cores to attain the data they need. As you can see from this chart by 3GHz the cores are all "well fed" and any additional memory capacity is not needed.

The Redshift engine is not like the one from CB23, which would scale in a completely linear (and unrealistic manner). This is still "embarrasingly parallel" but at least a more "real world" simulation in this respect.

 

[Hulk]

Maximum efficiency in this workload (at the cost of silicon) would be lots and lots of cores running at around 3GHz. Or of course just be smart and use a GPU

 

[itsmydamnation]

Maximum efficiency in this workload (at the cost of silicon) would be lots and lots of cores running at around 3GHz. Or of course just be smart and use a GPU

so you mean marks 9755 then

 

[Markfw]

Maximum efficiency in this workload (at the cost of silicon) would be lots and lots of cores running at around 3GHz. Or of course just be smart and use a GPU

My 9755 was running over 3 ghz on all 128 cores.

 

[poke01]

This is kind of interesting. IPC or throughput decreases as frequency increases. If I had to guess I would say that since memory access time remains constant as CPU frequency decreases there is more time for the cores to attain the data they need. As you can see from this chart by 3GHz the cores are all "well fed" and any additional memory capacity is not needed.

The Redshift engine is not like the one from CB23, which would scale in a completely linear (and unrealistic manner). This is still "embarrasingly parallel" but at least a more "real world" simulation in this respect.

View attachment 135992

Imo, I like the R26 test better its boring but way more realistic like you say.

R23 never made sense it to me

Edit: is R26 still bandwidth bound like R24? It doesn’t look it is to me.

 

[Markfw]

Imo, I like the R26 test better its boring but way more realistic like you say.

R23 never made sense it to me

Edit: is R26 still bandwidth bound like R24? It doesn’t look it is to me.

12 channel memory I think proves its not bandwidth bound. My 9755 seems to have shattered not only any score here, but ever the 128 core ampere. The GPU of course was beat by the 5090 in spades.

 

[Hulk]

12 channel memory I think proves its not bandwidth bound. My 9755 seems to have shattered not only any score here, but ever the 128 core ampere. The GPU of course was beat by the 5090 in spades.

Memory subsystem definitely has some effect on performance.