[tcclaviger]

Scroll Down For Gaming Results If Not Interested In Details

TLDR: 5800X3D is a monster at gaming, is easily OCd, and runs hot and beats even a tuned 5950x in most games; for general computing the 5800x, and more importantly now, the 5700x trounce it when OCd as well but cannot match it in gaming. Buy whichever fits your use case better.


My TLDR recommendations for optimum X3D performance:

• AMD CBS > NBIO > SMU > CPPC Enabled​
• AMD CBS > NBIO > SMU > CPPC Preferred Cores Disabled​
• AMD CBS > CPU > Global C-State Control Enabled​
• Use "the tool" or PBO2 Tuner to set CO curve verified with Corecycler AVX2 test.​
• RAM bandwidth seems to be more beneficial than tight latency, obviously fast and tight is best.​
• Keep SOC, CLDO VDDP, CCD, IOD voltages as low as you can for stability to preserve power envelope.​
• PLL of 1.85-1.95 seem to work well for higher FCLK. 2.0+ can cause USB dropout. 1.75 Observed stable at 3800/1900.​

//BELOW THIS ARE EARLY BETA BIOS IMPRESSIONS//
Initial thoughts on an early sample 5800X3D and some overclocking info/results.
​

• Skatter Bencher #29 covers the method, all Crosshair boards with the BCLK overclocking support settings can do it.
• AGESA 1206b is the earliest found that operates correctly. 1203 - 1205 confirmed to boot and work, but don't have functional performance boost so it sits at base clock of 3.4 ghz.
• Avoid upgrading AGESA without research beyond 1206b for now if everything is working as intended, AMD can make permanent changes to the CPU and we all know how useful the notes are...."Improved system performance".
• VID is way under the Robert Hallock mentioned 1.3 to 1.35 vcore limit, low to mid 1.2s mutli core and high 1.1s single core, I verified with an accurate DMM and read points were ~20mv over reported SVI2, inline with previous CPUs that have been in this C6E. My 105.4 BCLK has a +0.0125 offset applied, just enough to get it to boost properly again. VID has never exceed 1.288, SVI2 has never exceed 1.269 using that small offset. No crashes, no CO used, no changes to PBO limits, LLC Auto.
• PCIE will revert at some point as you climb BCLK, eg 4 to 3, or 3 to 2. I got away with manually setting PCIE 16_1 to 3.0 and the m.2 drive to 3.0 YMMV, particularly with PCIE 4.
• It's crazy hot, be ready, OCing on air or aio and keeping it cool is going to be a challenge.
• PBO2 Tuner (link in post 13 by Lionvibes)allows CO and PBO limits to be manipulated (lower only), do so carefully if you it, it doesn't behave quite the same as other zen3s do. For those who used a Zen 2 x cpu and a Zen 2 xt cpu, it boosts more like the xt variant. Flatlining target speed if temp headroom is available instead of floating like Zen 2x and Zen 3 Dual CCD chips.
• Aida 64 memory testing is not a forte of the X3D It doesn't do well there. The extra L3 incurs an L3 cache latency increase, and the relatively low clocks (for Zen 3) don't do any favors to bandwidth or latency. 5800x latencies of ~+4 ns are normal, L3 cache latency of ~12.5 is normal.
• Outside gaming, other apps are equal with 5800x if clock speed is equal or equal to 5800x at a 200mhz deficit, depending on the specific application. In a few cases it covers up to a 400mhz speed defecit.
• I forgot how nice it is to have to not deal with CO, Scalar, Multiplier, DOCS, per loaded core count multiplier, avx offsets, ring buses etc etc etc etc... Just find max stable BCLK, min voltage required, tune memory and send it. So much less work than fully tuning a 5950x.
• Max CPU speed is constrained by tolerance to BCLK increases of the GPU. The 3080ti simply doesn't like anything over 102.8, where the 2080ti would tolerate much higher, 110 was bootable but I didn't try past that. Will put effort into finding ways we can all benefit by increasing GPU tolerance to BCLK increases in the future.

• Overclocked testing being done with this memory configuration, TM5 stable, no WHEAs. Stockish testing done at 1900/3800, otherwise identical timings:
  

Spoiler: Early Testing

The below early testing results were done using a chiller on a C6E

ASUS System Product Name - Geekbench Browser

Benchmark results for an ASUS System Product Name with an AMD Ryzen 7 5800X3D 8-Core Processor processor.

browser.geekbench.com

ASUS System Product Name - Geekbench Browser - Updated, new fastest 8 core Zen 3 GB5 multi on Hwbot.
AMD Ryzen 7 5800X3D @ 4689.49 MHz - CPU-Z VALIDATOR This is 4795 Max boost single core.

Cinebench - R23 Multi Core with BenchMate overclocking records @ HWBOT

Overclocking records

hwbot.org

Low Effort 2080ti 3dMark Results:
TS- I scored 16 189 in Time Spy
TSE- I scored 7 622 in Time Spy Extreme
FS- I scored 36 051 in Fire Strike
FSE- I scored 19 565 in Fire Strike Extreme
FS-U- I scored 10 136 in Fire Strike Ultra
PR- I scored 10 863 in Port Royal
CPU- I scored 0 in CPU Profile

Low Effort 3080ti 3dMark Results (all updated with C8E motherboard):
TS- I scored 20 458 in Time Spy +26.3%
TSE- I scored 10 136 in Time Spy Extreme +33%
FS- I scored 41 994 in Fire Strike +16.5%
FSE- I scored 26 040 in Fire Strike Extreme +33%
FS-U- I scored 14 473 in Fire Strike Ultra +42.8%
PR- I scored 15 192 in Port Royal +39.6%


- for overclocking results, max expected gains are easy, since any increase over 100 bclk is the same % gain. There are cases where it both over and under delivers expected results.

- Mostly an effort to validate when and where it's showing improvements, plus provide some stock X3D numbers that don't use gimped 3200 or 3600 memory configurations like many of the results in media.

Only games with built in benchmarks will be tested, no time for or interest in custom run testing and online games are far too variable IMHO. The one exception will be Star Citizen, a game that is CPU bottlenecked on every CPU in existence at every resolution and setting currently.
​

Star Citizen, basically defies benchmarking, but here are my impressions and observations.
-Objectively it is a typical 10-30 fps better than 5950x in CPU constrained areas, new Babbage, loreville, space. The HUD no longer drags frame rate down in ships, so you don't get a big fps boost looking away from HUD. Long distance QT showed 130-163 fps range, traversing around a planet in QT was 120sih, New Babbage 70-80 typical with spikes over 110 and min around 60. Loreville 80-95 typical with min around 60 in New Deal. Grim Hex is basically a gain of 20 fps almost everywhere.
- Subjectively it feels night and day better. Frame pacing feels improved and average rates being higher makes everything much nicer. The X3D+2080ti runs it better, in all areas than 5950x+3080ti, to include space.

SotTR: reran with v-sync off, it changed nothing, +/- 1 fps average, gains constrained by GPU bottleneck, mind blown a Ryzen is bottlenecked by a shunted 2080ti at 1080p Lowest. SotTR is showing typical love for all improvements in setup and shows gains as expected:

2080ti results:
Overclocked
720 - Lowest(6% Gain) 1080-Lowest (4.8% Gain) 1080-Medium(1.3% Gain) 1440-Medium(0.5% Gain)

Stockish:
720 - Lowest 1080 - Lowest 1080 - Medium 1440 - Medium

3080ti results:
Overclocked:

Horizon Zero Dawn: Original Quality Setting, V-sync off, no scaling:

Cleary HZD is more GPU constrained than SotTR is, in fact, at 1440p, the GPU and CPU performed slightly better while running the stock test, and the results are better than OC results. I consider this as likely test variance, one may have been done at the top of my coolant temp window (55f) and one at the bottom(48f) as the chiller cycled on and off or some windows foolery going on in the background.

2080ti results:
Overclocked
720 (2.1% gain) - 1080 (1% gain) - 1440 (regression/variance)

Stockish:
720 - 1080 - 1440

3080ti results:
Overclocked:

CB 2077: Default High Preset, Vsync off, Fullscreen, No DLSS:

2080ti results:
Overclocked
720(2.2% gain) - 1080 (1.1% gain) - 1440 (% gain)

Stockish:
720 - 1080 - 1440

3080ti results:
Overclocked:
720 (-5%) - 1080 (+19.4%) - 1440 (+29.6%)

FF14 Shadowbringers: Desktop(High) Preset (reBAR On/Off = no change here)
Results here are astounding. The stockish 5800x3d and 2080ti at 2250mhz+1250mem kills my 5950x when it's running 5250mhz effective boost clocks even when I keep the coolant at 35f + 3080ti at 2350mhz+1330mem. Shows just how CPU bound the FF14 engine is.

2080ti results:
Overclocked:
720(5% gain) - 1080 (4.6% gain) - 1440 (1.7% gain)

Stockish:
720 - 1080 - 1440

3080ti results:
Overclocked:
720 (-2.4%) - 1080 (+3%) - 1440 (+15%)

Assassins Creed Odyssey: Very High Preset
Did not correct strange behavior prior to swapping 3080ti in for 2080ti OC results. Highly volatile results, it's honestly a trash benchmark, but here it is.

2080ti results:
Stockish:
720 - 1080 - 1440 - 3440

3080ti results:
Overclocked:

5950x head 2 head testing canceled, simply don't care to use my time for it. X3D is faster in nearly every game and nearly imperceptiblly slower in most non-gaming general use loads. Productivity obviously the 5950x crushes the X3D.

Don't need tests and graphs for that info lol.

 

[Farih]

Subbed

 

[kairi_zeroblade]

I am curious on where to get that PBO2 Tuner app, I know its from someone here over on OCN, but I also think its not publicly available.

 

[tcclaviger]

I am curious on where to get that PBO2 Tuner app, I know its from someone here over on OCN, but I also think its not publicly available.

No one told me that, I'll post it tomorrow

To not clutter the gaming results since it will be a lot of screen shots when complete, here is a shot of hwinfo during 16 thread run of cpuz bench to show VIDs etc being requested.
When interpreting temps, consider my coolant is 51f while this is running, room ambient is 84f, so on a ambient radiator based loop I would be at least +18.6c hotter on the cores, but more realistically about 24c hotter using a 10f air/water delta for ambient loop math (sorry for interchanging units, it's how I think):

Capture of the same during 1t boosting:

 

[Taraquin]

Nice info and testing! I wish they had allowed for negative CO, would have gotten a nice multicore boost (up to 6% if -30 CO is possible), also would liwered temps in SC scenarios.

 

[Mach3.2]

I am curious on where to get that PBO2 Tuner app, I know its from someone here over on OCN, but I also think its not publicly available.

[Official] AMD Ryzen DDR4 24/7 Memory Stability Thread

This is the limit you can raise to in rm if you leave them on auto What's the point in raising it then, if it will never apply? I mean (if I get it right from your post) your BIOS allowed EDC limit to be manually set to max 220 amps, and that it's way below the actual mb limit? This is the...

www.overclock.net

 

[OCmember]

Will probably be joining the club on release day.

 

[Luggage]

My gb5 as comparison.

Micro-Star International Co., Ltd. MS-7C35 vs ASUS System Product Name - Geekbench Browser

the difference for aes, speech and machine learning are huge.

 

[jonathandc1984]

Awesome initial write up. I’ll be jumping on the bandwagon as soon as it is released.

Quick question and hoping somebody can assist -
Does anybody know if the crosshair viii dark hero has an external clock generator and also supports the voltage suspension feature available in the crosshair viii extreme?

thank you kindly.

 

[tcclaviger]

Awesome initial write up. I’ll be jumping on the bandwagon as soon as it is released.

Quick question and hoping somebody can assist -
Does anybody know if the crosshair viii dark hero has an external clock generator and also supports the voltage suspension feature available in the crosshair viii extreme?

thank you kindly.

it does have an external generator it can do what I'm doing, it is missing voltage suspension but voltage suspension is the least important of all the technologies to help overclocking using the Crosshair and x3d CPU. I'm using a Crosshair 6 extreme at the moment which is a 5-year-old Zen One release and that's what all the scores posted above were obtained with. I chose this over my Crosshair 7 Hero because, frankly, the C6E is a better board in almost every regard (boot times being the main exception).

Nice info and testing! I wish they had allowed for negative CO, would have gotten a nice multicore boost (up to 6% if -30 CO is possible), also would liwered temps in SC scenarios.

Curve Optimizer can be used using the pbo2 tuner but it doesn't need to be. That's why I mentioned it behaves differently than other Zen 3 CPUs, simply dropping the Curve for each core does not necessarily produce a better results like it does on all the other Zen 3s.
It's so low voltage stock that it ends up dropping voltage to a point where it no longer boosts properly and instead gets stuck at like 4 ghz when using more than -10 CO values for my sample. It's also not TDC/EDC/PPT limited through virtue of not hitting the limiters, so less power doesn't help anything.

The only time I've observed similar behavior on normal Zen3 is when PPT is cut so far down the CPU now doesn't have headroom to boost right, like PPT 50 on a 5950x.

 

[yzonker]

Some good info for sure. I may grab one of these if I can manage to get it. Anyone know what time of day they will go up Wednesday?

 

[tcclaviger]

Some good info for sure. I may grab one of these if I can manage to get it. Anyone know what time of day they will go up Wednesday?

I was shocked they shipped mine early, had no idea it was going to happen (might have been a mistake tbh), so I was unprepared and still waiting on the Crosshair 8 and second memory waterblock to arrive to do the 5950x comparison data hah.

I would see if you can find a place like B&H that has them on "Waiting list only" so you get a notification the instant they drop if you want one. Despite a lot of nay-sayers, I have feeling it's going to sell like Zen 3 originally did at launch...quickly.

 

[Lionvibez]

I am curious on where to get that PBO2 Tuner app, I know its from someone here over on OCN, but I also think its not publicly available.

Direct link here.

Debug.7z

drive.google.com

 

[yzonker]

I was shocked they shipped mine early, had no idea it was going to happen (might have been a mistake tbh), so I was unprepared and still waiting on the Crosshair 8 and second memory waterblock to arrive to do the 5950x comparison data hah.

I would see if you can find a place like B&H that has them on "Waiting list only" so you get a notification the instant they drop if you want one. Despite a lot of nay-sayers, I have feeling it's going to sell like Zen 3 originally did at launch...quickly.

Thanks for the tip. I suspect if I don't have windows open to BH/Newegg/Amazon/AMD/etc... when they drop though, I won't have a chance of getting one. Looks like previous releases have all been the standard 6:00 AM PT, so I'll have to tell my boss I have an "appointment" that morning (thanks to them blocking everything on the internet at work). LOL

 

[Lionvibez]

Hmm also on a side note it seems Ryzen master now also has access to Curve Optimizer

https://www.tomshardware.com/news/curve-optimizer-heads-to-ryzen-master-for-zen-3-cpus

 

[tcclaviger]

Hmm also on a side note it seems Ryzen master now also has access to Curve Optimizer

https://www.tomshardware.com/news/curve-optimizer-heads-to-ryzen-master-for-zen-3-cpus

View attachment 2556621

Good to know for those who use RM. I refuse to ever touch it again, but at least the option is out there.

 

[Panchovix]

Hmm also on a side note it seems Ryzen master now also has access to Curve Optimizer

https://www.tomshardware.com/news/curve-optimizer-heads-to-ryzen-master-for-zen-3-cpus

View attachment 2556621

Not 5800X3D user here (just a 5800X), but testing this and it has been pretty good, finally an "official" AMD Auto curve optimizer.

 

[des2k...]

Not 5800X3D user here (just a 5800X), but testing this and it has been pretty good, finally an "official" AMD Auto curve optimizer.
View attachment 2556638

The curve optimizer is interesting. I didn't even know that was a thing with Ryzen Master.
Started this morning to see what all CO value would be stable for my 5900x.

-27 worked for all cores. Now I'm doing 1h OCCT cycler for each core to see which one can handle -30.
I'll confirm with Prime95, OCCT 8h overnight once I have all my values.

 

[Luggage]

The curve optimizer is interesting. I didn't even know that was a thing with Ryzen Master.
Started this morning to see what all CO value would be stable for my 5900x.

-27 worked for all cores. Now I'm doing 1h OCCT cycler for each core to see which one can handle -30.
I'll confirm with Prime95, OCCT 8h overnight once I have all my values.

Read the corecycler thread.

 

[tcclaviger]

The curve optimizer is interesting. I didn't even know that was a thing with Ryzen Master.
Started this morning to see what all CO value would be stable for my 5900x.

-27 worked for all cores. Now I'm doing 1h OCCT cycler for each core to see which one can handle -30.
I'll confirm with Prime95, OCCT 8h overnight once I have all my values.

Corecycler is better unattended testing imo because OCCT doesn't always let you know which core trigger a restart when CO is too aggressive, cycler does via the log. It's also slightly better at triggering faults than OCCT. OCCT is great for roughing it in, as you have done.


I was curious how the extra cache and powering it impacts the X3D vs normal Zen 3 so I did a few tests.

I disabled CCD1 on my 5950x, set DOCS to 44.5 at 1.263 VID and then capped single core boost to 4550 using PBO2 Tuner, so simulate the 5800X3D and see what, exactly, the cache does outside gaming and check silicon quality between them.

X3D and simulated X3D use identical VID during 4550 limited 1t boosting, and my 5950x has a very, very good CCD0. Hopefully this is the norm for all X3Ds and I didn't get an outlier.

Package Power draw is consistently 5 watts higher for simulated x3D, no explanation other than MB differences maybe.

• Cpuz within a couple of points multi, exactly the same score single.
• R23 simulated x3D wins by multi by 200 points, the nature of DOCS vs PB2 all core loads, manual always wins given frequency parity. Single score is within 2 points.
• GB5: 1645/12153 vs 1619/11419 X3D, Simulated X3D respectively. X3D cache is helping here.
• Passmark below:
  X3D
  Simulated X3D