BANANA Benchmark Results Comparison

As seen in one of my screenshots, that's editable, I did 48 and still sat over 100%, though that's not technically fair to represent.

You can also edit vehicle which is kinda cool, though for consistency I stuck to pickup
--- Post updated ---
I did a pigeon run but that was just silly

 

I've basically read every single post from the creation of the thread up to today to make it as updated as possible. I've seen some users removing the limit of 20 vehicles (and they were added to the spreadsheet with the max number they could achieve) and some using other vehicles that were not added, since all the previous tests were done with the pickup. I think it's right to set a benchmark that is as equal as possible between all tha platforms, so if you (referred to the community) want to be added to the sheet, please use the pickup test (you can still remove the vehicle limit tho in my opinion).

 

Decided to re-run the benchmark to see if anything changed from last time... Gained only 4 Mbeams/s

 

I

Nah more my point is I've screenshotted a 3900X doing 48 pickups and still being at 105% realtime, but that's on the sheet as 20
--- Post updated ---
But you'd then need a way of showing this test was capped at 20 and so only achieved 20 versus those that were capped at more but only achieved realtime at 20, cell background maybe
--- Post updated ---
I'd like to build a modified bananabench that keeps going until 100% realtime is not met

 

Here was my findings for CCX usage scenarios, each picture is the result.

TEST 1: Both CCX in use for 8 cars


TEST 2: Single CCX in use for 8 cars


TEST 3: 2 cores on each CCX in use (2 per CCX disabled) for 8 cars


In all the results didn't show much noticeable differences but it does seem that test 3 yielded slightly higher performance then test 2 unless the timing of the screen shot was at a spike in perforance

 

Yes, this is interesting, I'd still like to investigate further, of course I'm on the 3900X so it's even more complex as I have 2 discrete CPU dies each with 2 CCX's of 3 cores each.

When that 3950X comes along in September, holy moly that's gonna annihilate things

 

And the spreadsheet is back online, sorry for any inconvenience, I had to fix the Sharing settings, now it should work as expected.

Alright, I got the point . I've fixed your result (I didn't see the second run you've had) and I've highlighted with a different background colour the cells where the test was capped at 20 vehicles.

 

I didn't even know that individual CPU cores could conk out. What do you think caused that to happen?

 

Can u include my i9 9900k in the list, or didn't I meet a requirement

 

You're already in the list.

 

Thx sry

 

Ok, I added a wait function to the banana bench lua so it stays open long enough for me to get a screenshot. Running my i5-3570k @ 4.5 GHz, I can consistently get a score of 112 M/Beams/s.

@Alex [ITA] can I be on the spreadsheet now?

 

I'm going a bit off-topic here, but doing it for a staff member's benefit.
"that behavior" in the cpu cores conking out is usually a result of 'electro-migration' or basically, the voltage / energy going through the circuit fatigues the normal circuit paths, breaking a path or shorting two paths, causing an error of sorts which can result in:
*Not POSTing, or a reboot loop where the machine turns on and off repeatedly as if it is training RAM like 3xx~5xx AMD chipsets (and some intel chipsets) do during 'first boot' with new / modified hardware or a cleared CMOS (or a dead CMOS battery in a machine that's been off for more than a few minutes - like mine)
*Random reboots during initial post (after the ram training phase when it's handing off BIOS to various boot-ROMs).
*BSOD 'sometimes' when Windows loads, other times not, random crashing in Windows such as 'Your computer has encountered an error'
*MEMtest86 errors abound as a computational error inside the processor will be picked up as RAM errors sometimes, depending on the error
*all of the above, too.

Those processors really need over 1.45v to degrade... they're an older process (28nm or 32nm or something like that), so they can eat a lot of volts, but if your VRM gets pretty toasty (it can with 8-core FX chips + tower-style CPU coolers) and doesn't have the cooling of the OEM down-draft style coolers (where the fan points airflow against the motherboard/cpu forcing it through the heatsink, out the sides, to cool the VRM), your VRM can cause issues. Also, cheaper analog VRM or analog + digital VRM's can sometimes be suspect (as can voltage regulators on the board, etc).

Can even be something as silly as fan vibrations making one or more pins on the CPU socket not contact properly.
Or it could just be a manufacturing defect that did not rear it's ugly head until now.

That being said, an old 3xx or 4xx AMD board (aside of a300 chipset boards) can go with a new 3xxx or used 2xxx Ryzen quite nicely if you don't wish to pay the intel tax.
https://forums.guru3d.com/threads/help-me-choose-which-amd-board-for-ryzen-3000.427485/ (this link is generally SAFE FOR WORK)
Just look up the resource I have handy (of course) to judge which AMD board to pickup, and if going with a 3xxx chip make sure it can be flashed without an older cpu laying around (BIOS Flashback).
It is prudent that you review the limitations of the VRM on which board you wish to choose, to pair it with the best processor for your needs and it's ability.
Will be building a new Ryzen machine sometime this year. This one is almost 5 years old so... for the folks handing in their 3xxx Ryzen series benchmarks THANK YOU!
My old 4790k here, 5 years of age, on Power Supply #2, 4.4ghz-all-cores, w/2400mhz 32gb ddr3 cas11 RAM, Phanteks TC14PE air cooler, delidded with liquid metal above and below the IHS.

Brought to you by Windows 7 Professional, because it still works! Oh, and STAYS WORKING and doesn't commit suicide with updates while I'm out, or worse, while I'm working.

 

Oh thank you, I felt quite lonely as I still use Windows 7 on my secondary rig (i5-3470/8Gb of 1333MHz ram) and on my laptop which had that horendous OS known as Windows 8 (yes, the original one, not 8.1). Unfortunately I had to ditch it for my main rig since it was not officially supported by the CPU. I'll keep it till Micro$oft kills the security updates on January

 

That's still at stock speeds? Holy Bananas.

 

Yep, we're not even fully boosting as we've hit the power limit, also not thermal throttling though, I think we hit about 4.3ghz on all cores

 

Looking through table three, it appears that hyperthreading doesn't really make all that much difference to the results.

 

That's because (but I might be wrong) BeamNG uses one physical core per vehicle more or less. You may be able to squeeze one/two more vehicles over your core count with the recent optimizations.