*Official Benchmark Thread* - Post it here or it didn't happen :D

[Mr. Fox]

8 hours ago, ssj92 said:

Ah now I am debating on buying another 2x480mm rads and another pump since everything is on sale.....gotta sell some stuff lol

 

 

Also is Firestrike bad at taking advantage of multiple cores? What do people here get in FS physics with their CPU. This is with 3.9Ghz all core, 4.8Ghz single core, and I think 4.6Ghz 6-12 cores multipliers.

 

 

This is my absolute limit with my air cooler. 4.1Ghz all core, thermal throttle for a bit. 

 

Yes, Fire Strike chokes on too many threads. With i9 13900/14900/K/KF/KS your physics and combined scores go way up if you disable 3 of the baby cores. Same for 3DMark 11. They're both too old and don't know how to use them. (From a time when quad core hyperthreading was the norm.) Time Spy and Sky Diver will use them all and give you a score reflecting that a better CPU is installed.

---

While it should not come as a surprise... we already knew this: the talking heads on YouTube that say 9950X is not a major improvement over 7950X are so full of crap their eyes are brown. They clearly don't know how to test CPU performance. (They're probably using games that don't care what CPU is installed and making numerous erroneous assumptions out of ignorance.) The 13th and 14th Gen i9 CPUs rip the head off of 7950X and poop down its neck, but the 9950X more or less levels the playing field against a 13900KS/14900KS. It's really unfortunate that AMD makes it standard practice to dump their trash bin chips on the second CCD. If both CCDs were as good as CCD1 this CPU would be so much better. I mean, it's a great CPU for sure, but why stab one of its nuts with an ice pick? That's just so stupid. They should put those crappy silicon samples in the inexpensive single CCD gamerwuss CPUs and save the cream of the crop chips for both CCDs on 9950X.

[tps3443]

@ssj92

 

Disable HT and run Firestrike Physics again. It’ll probably double lol. This is the only test that cannot run right on all threads for me as well. Although weirdly Windows 22H2 helped a lot with this, but disabling HT helps even more!

 

Firestrike Physics just has no idea what to do with your 3465X. 
 

Another thing is dumb, and this test really makes no sense. My 3175X can run Timespy GT1 at 30% GPU usage with HT on, but GT2 is 100% GPU usage lol. If you turn HT off, it’s full GPU usage for GT1 and GT2. Really just poor optimization I guess. 

[ssj92]

19 minutes ago, tps3443 said:

@ssj92

 

Disable HT and run Firestrike Physics again. It’ll probably double lol. This is the only test that cannot run right on all threads for me as well. Although weirdly Windows 22H2 helped a lot with this, but disabling HT helps even more!

 

Firestrike Physics just has no idea what to do with your 3465X. 
 

Another thing is dumb, and this test really makes no sense. My 3175X can run Timespy GT1 at 30% GPU usage with HT on, but GT2 is 100% GPU usage lol. If you turn HT off, it’s full GPU usage for GT1 and GT2. Really just poor optimization I guess. 

I've been wondering why my FS test 1 has such low utilization but FS test 2 is good on every GPU I tested. I think this might be why, I am going to try this and hopefully that's it 

[tps3443]

40 minutes ago, ssj92 said:

I've been wondering why my FS test 1 has such low utilization but FS test 2 is good on every GPU I tested. I think this might be why, I am going to try this and hopefully that's it 

Yep! They are wonky with the 1st graphics test, and the physics tests, interesting you’re having the same issues with your 28/56 chip. My firestrike was just like that though. And it’s all due to Hyper Threading. The good news is, real world gaming works really really well leaving HT on. So this is really only needed for Firestrike and Timespy. All of my games I play push my RTX4090 perfectly! Coming from a 14900KS, I don’t really notice a difference real world. 

[Mr. Fox]

23 hours ago, Clamibot said:

Well guys, I benchmarked my 2019 LTSC install vs my new WindowsXLite 22H2 install in Shadow Of The Tomb Raider again (this time using the built in benchmark), and I can confirm the 22H2 install does indeed perform better, even with all my programs installed (which seemed to have no effect on performance at all). Looks like WindowsXLite 22H2 is the to upgrade from LTSC 2019! I'm currently installing it on my Legion Go and will be installing it on my X170 next.

 

I had my 10900K running at 3.7 GHz to induce a CPU side bottleneck. My GPU is a Radeon RX 6950 XT.

 

2019 LTSC:

 

WindowsXLite22H2 (minimalist gamer only installation):

 

WindowsXLite22H2 (all my programs installed + some extra services running):

 

WindowsXLite22H2 wins by about 2%. I did not expect this at all. I was expecting a performance downgrade, but I am very happy I got a slight performance upgrade instead. You don't see that very often with installing newer versions of windows.

 

I like that Shadow Of The Tomb Raider is useful both for CPU and GPU benchmarks. This makes it an easy all in one benchmark that saves me time as it will give me a general idea of performance differences between different machines and windows installs.

 

6 hours ago, Mr. Fox said:

The 13th and 14th Gen i9 CPUs rip the head off of 7950X and poop down its neck, but the 9950X more or less levels the playing field against a 13900KS/14900KS.

I think you might appreciate this...

[tps3443]

28 minutes ago, Mr. Fox said:

 

I think you might appreciate this...

 

 

Maybe you can run this. I remember whooping “Fully Tuned” 7800X3D’s and 7950X3D’s  pretty good with my last setup. Honestly the 9950X/9950XD might just take it now though. 

 

 

[Mr. Fox]

26 minutes ago, tps3443 said:

Maybe you can run this. I remember whooping “Fully Tuned” 7800X3D’s and 7950X3D’s  pretty good with my last setup. Honestly the 9950X/9950XD might just take it now though. 

 

Running it as 720P low would basically make it mostly a CPU benchmark. That makes sense in a way and I see some of the YouTube dudes doing it, but what it doesn't do when one does that is actually capture how a CPU truly performs when being used "properly" (as in a real-world scenario) in a gaming setup. I think the conclusion could end up being misleading if the behavior doesn't always hold true to what the "real" experience is like.

 

I am pretty sure the 14900KS is still going to beat the 9950X in that kind of test, especially with an all core fixed clock on the 9950X like I am running. It can't boost to 57x in light loads like it does stock. I have it fixed at 54x on all cores, both CCDs.

[tps3443]

3 minutes ago, Mr. Fox said:

Running it as 720P low would basically make it mostly a CPU benchmark. That makes sense in a way and I see some of the YouTube dudes doing it, but what it doesn't do when one does that is actually capture how a CPU truly performs when being used "properly" (as in a real-world scenario) in a gaming setup. I think the conclusion could end up being misleading if the behavior doesn't always hold true to what the "real" experience is like.

Yeah CPU only benchmark. Thats how people do it I guess lol. 

[Clamibot]

Shadow Of The Tomb Raider Benchmark on my Legion Go (also using a new WindowsXLite 22H2 install):

 

 

I set the power limit of the APU to 54w and overclocked the iGPU. All 8 CPU cores were active. I also set the temperature limit of the CPU and the skin temperature sensor to 105°C to stop the APU from throttling because it thinks my hands will get burned. The results are a definite improvement over stock settings, but still not up to par with my standards for raw framerate. However, the input lag is not very noticeable when using a controller and setting my raw framerate to 72 fps, then interpolating to 144 fps, thus giving me my high refresh rate experience on this device.

 

Despite raising the power limit to 54w, the system was maxing out at 40w sustained. I'm not sure if this is because the APU was using 40w, with the remaining 14w of the power budget being used for everything else. I thought the power limit I set using the Smokeless tool was for the APU only. I'm definitely pushing the power circuitry pretty hard here as the APU is rated for 28w only.

 

I was able to perform a static overclock on the iGPU to 2400 Mhz and, performance did improve. Interestingly, performance dropped if I tried pushing further as it seems there is either a power limit, voltage limit, or both. UXTU allows me to overclock the iGPU up to 4000 Mhz, but I didn't try going that far. I was however able to overclock my Legion Go's iGPU to 3200 Mhz without the display drivers crashing, so looks like this iGPU has a lot of overclocking headroom left, but is held back by power/voltage limits.

 

I can't wait to see a handheld with a Strix Halo APU and a 240 Hz screen.

[jaybee83]

On 11/22/2024 at 2:17 PM, Mr. Fox said:

The latency killer feature MSI added to the UEFI seems worthless to me. It seems to only offer value if a person is using an XMP/EXPO profile, which seldom ever produces impressive memory performance. This EXPO Trident Neo definitely works better (for some reason) than the equivalent XMP kit. Maybe the IC bin quality is better, or maybe the SPD firmware is different or AMD-specific JEDEC. But, I am not using the EXPO profile, only manual settings.

 

This would be so much easier and better with a 2-DIMM motherboard. It's so sad the only great option for that on AMD is getting screwed over paying new retail price for a used X670 Gene motherboard that might have something wrong with it.

I finally received the RAM jackets from China, so I will be pulling off the trash stock G.SKILL heatsinks heating blankets and getting this RAM under water like my Intel systems. The memory temperatures with the trash G.SKILL covers are just so lousy (actually cooler even naked, but ugly to look at).
 

I received the parts ordered from Thermal Grizzly, so maybe tonight or tomorrow I will decapitate the 9950X, test the TG heatspreader, then remove that and install the Mycro. I ordered a heatspreader to compare the thermal difference between that and bare die. Just for curiosity, in case I am ever asked I will know the answer.

 

funny timing, Jufes just did a comparison of all three (stock IHS, HP IHS, direct die):

 

 

delid + stock IHS = no difference

HP IHS + paste = -5C

HP IHS + LM = -10C

direct die = -20C

[Mr. Fox]

6 hours ago, jaybee83 said:

delid + stock IHS = no difference

HP IHS + paste = -5C

HP IHS + LM = -10C

direct die = -20C

Cool (pun intended). Will probably use the HP IHS on my son's system with an AM5 CPU. I guess I won't bother comparing now. Just straight to direct die. No point in wasting my time thanks to Jufes. (That may sound sarcastic, but I don't mean for it to. I was not looking forward to the testing.) 

 

Delidding the 5950X did absolutely nothing with an IHS.

[Talon]

On 11/23/2024 at 8:40 PM, tps3443 said:

 

 

Maybe you can run this. I remember whooping “Fully Tuned” 7800X3D’s and 7950X3D’s  pretty good with my last setup. Honestly the 9950X/9950XD might just take it now though. 

 

 

 

https://benchmark.unigine.com/results/rid_77e3607cb53d4f128d156fb9ceb6dae5

 

My 9800X3D with a lowly 4070 Super. While mostly CPU limited, there are times where I can see a 4090 would help. Maybe this week I'll find time to swap GPUs for a few runs. Currently #4 spot with this GPU though. 

[Mr. Fox]

  

12 hours ago, jaybee83 said:

 

funny timing, Jufes just did a comparison of all three (stock IHS, HP IHS, direct die):

 

 

delid + stock IHS = no difference

HP IHS + paste = -5C

HP IHS + LM = -10C

direct die = -20C

 

6 hours ago, Mr. Fox said:

Cool (pun intended). Will probably use the HP IHS on my son's system with an AM5 CPU. I guess I won't bother comparing now. Just straight to direct die. No point in wasting my time thanks to Jufes. (That may sound sarcastic, but I don't mean for it to. I was not looking forward to the testing.) 

 

Delidding the 5950X did absolutely nothing with an IHS.

Success! https://imgur.com/a/HAftuqu

Before Decapitation ^^^

IHS Removed ^^^

Liquid Metal Application ^^^

Micro Direct Die Block ^^^

CPU in socket (Kapton tape to prevent CPU from lifting out of socket with MDD block) ^^^

Installed... ^^^

Finished Product ^^^

Before Delid ^^^

After Delid ^^^

Before Delid ^^^

After Delid ^^^

AIDA64 5-Minute Stress Test (ambient 79°F / 26°C) - normalized at 72°C

 

Edit: I just noticed that with the latest BIOS that Benchmate is not reading the core voltage correctly. The core voltage is fixed at 1.250V and other monitoring software confirms the same. Clearly not 2.00V+ vCore or CPU would be melting, LOL...

Edited November 25 by Mr. Fox
added comments

[tps3443]

13 hours ago, Talon said:

 

https://benchmark.unigine.com/results/rid_77e3607cb53d4f128d156fb9ceb6dae5

 

My 9800X3D with a lowly 4070 Super. While mostly CPU limited, there are times where I can see a 4090 would help. Maybe this week I'll find time to swap GPUs for a few runs. Currently #4 spot with this GPU though. 

 That’s what I was curious about. So I suppose about 8% faster for the 9800X3D in seriously CPU bottlenecked 720P situations AKA: “1080P Quality DLSS”. My 14900KS was stock, other than cache and memory overclocking. 
 

Not sure if a 4090 might help that. I think my 4090 was only around 25-35% GPU usage average. With a max of 48% in the hardest areas. But definitely give it a test and see for sure. Maybe it can do even 70K. 

[Mr. Fox]

9 hours ago, Mr. Fox said:

  

 

Success! https://imgur.com/a/HAftuqu

Before Decapitation ^^^

IHS Removed ^^^

Liquid Metal Application ^^^

Micro Direct Die Block ^^^

CPU in socket (Kapton tape to prevent CPU from lifting out of socket with MDD block) ^^^

Installed... ^^^

Finished Product ^^^

Before Delid ^^^

After Delid ^^^

Before Delid ^^^

After Delid ^^^

AIDA64 5-Minute Stress Test (ambient 79°F / 26°C) - normalized at 72°C

 

Edit: I just noticed that with the latest BIOS that Benchmate is not reading the core voltage correctly. The core voltage is fixed at 1.250V and other monitoring software confirms the same. Clearly not 2.00V+ vCore or CPU would be melting, LOL...

Next projects...

• Delid 14900KS and install bare die block
• Replace trash stock heatsinks on G.SKILL Neo 8000 RAM kit with water cooling

[Mr. Fox]

On 11/23/2024 at 7:40 PM, tps3443 said:

Maybe you can run this. I remember whooping “Fully Tuned” 7800X3D’s and 7950X3D’s  pretty good with my last setup. Honestly the 9950X/9950XD might just take it now though. 

Interesting that the GPU utilization is higher on the 9950X in spite of the preset being the same. Not sure why the difference there. But, basically a toss-up and within a margin of error. I didn't do a second run (which is often higher) just grabbed the first for each system.

[Mr. Fox]

[electrosoft]

21 hours ago, Talon said:

 

https://benchmark.unigine.com/results/rid_77e3607cb53d4f128d156fb9ceb6dae5

 

My 9800X3D with a lowly 4070 Super. While mostly CPU limited, there are times where I can see a 4090 would help. Maybe this week I'll find time to swap GPUs for a few runs. Currently #4 spot with this GPU though. 

 

Awesome score! True story, I ran this before on my 7950X3D and was bummed over the low score but looking back at my screen shots I now realize I was running it at 1080p. I totally missed the 720p low setting!!!

 

----

 

Meanwhile in my neck of the woods....

 

Asus B650 Plus rock solid 6000 and 6200 on older M-die sticks with the 7600x and older BIOS but needed to flash to newer BIOS for 9800X3D support.

 

New BIOS can not be downgraded. Fine, need 9800X3D support.

 

Initial tests at 4800 with 9800X3D pass everything as expected. Time to use my old profiles for some testing.

 

9800X3D...only runs up to ~5400 on memory. Anything after that is flaky and sometimes will boot and other times not. Same settings.

 

Figure it may be a weak chip, swap back in 7600x now even it won't run at its old settings and fails at ~5400. Exact same settings.  7600x before could do 2200 fclk for testing. Now it can't do anything past 2000. Same for 9800X3D.

 

I've been fighting with it all weekend. Leave everything stock, fine. As soon as I press pass ~5400, non boots and crashes or if I make it to the desktop TM5 bombs out within minutes if not seconds. This is between two cpus, three sets of memory.

 

Finger kept hovering over the Newegg Combo deal for a Hero + 9800X3D for $60 off and a free Corsair memory kit 2x16GB 6400 but then sanity kicked in and that would be another $1200.

 

Time to search deals for something competent...and cheap. I still can not stomach paying more than $350 for a motherboard. I'll accept it sooner than later.

 

Finally gave up the ghost and lucked up and found an open box ASUS ROG Crosshair X670E Hero for $299.99 on Amazon Warehouse/Reseller (whatever they're calling themselves these days). They had 1 only so I snagged it. Hopefully it won't be a miss.

 

Bonus, I guess I get to now fret over SP ratings again along with everything else. 🤣

 

 

 

 

[tps3443]

How are the AMD SP ratings comparing? Tested any that are much better than other chips? Or does it not matter much? 

[electrosoft]

1 hour ago, tps3443 said:

How are the AMD SP ratings comparing? Tested any that are much better than other chips? Or does it not matter much? 

 

Silicon variance is definitely in play in regards to pull and overclocking. In the end, it is still a CPU.

 

SP111 = meh

SP115+ = golden land

 

If you decapitate them and go DD, things tighten up real quick.

 

I'd trade SP for a 2200+ fclk all day though.

 

 

 

 

[Mr. Fox]

1 hour ago, electrosoft said:

Silicon variance is definitely in play in regards to pull and overclocking. In the end, it is still a CPU.

 

SP111 = meh

SP115+ = golden land

 

If you decapitate them and go DD, things tighten up real quick.

 

I'd trade SP for a 2200+ fclk all day though.

I don't know if it is me and my ignorance of AMD overclocking, or if it is something bugged with the firmware, but I cannot set CCD0 and CCD1 at different core ratios. If I do that they all run 43x max and ignore my settings. Seems like only PBO and All Core work, and PBO sucks as much as eTVB... rubbish. Sometimes I wish I had an SP rating, but other times I think my state of mind is better just exploring where the edge of functionality is for this CPU and not care about SP rating. I used to not care and not look, and we all got sucked into that flawed ASUS gimmick that creates a jaded view before any tuning starts rather than an informed understanding.

 

Anyhoo... personal all-time best score for Cinebench R23. And, I still have thermal headroom if the silicon quality is able to handle more.

 

https://hwbot.org/submission/5702114_

 

[electrosoft]

12 hours ago, Mr. Fox said:

I don't know if it is me and my ignorance of AMD overclocking, or if it is something bugged with the firmware, but I cannot set CCD0 and CCD1 at different core ratios. If I do that they all run 43x max and ignore my settings. Seems like only PBO and All Core work, and PBO sucks as much as eTVB... rubbish. Sometimes I wish I had an SP rating, but other times I think my state of mind is better just exploring where the edge of functionality is for this CPU and not care about SP rating. I used to not care and not look, and we all got sucked into that flawed ASUS gimmick that creates a jaded view before any tuning starts rather than an informed understanding.

 

Anyhoo... personal all-time best score for Cinebench R23. And, I still have thermal headroom if the silicon quality is able to handle more.

 

https://hwbot.org/submission/5702114_

 

 

Whew, pressing it nicely! 285k and 9950x are turning out some decent scores!

 

I never tried to set different ratios between ccds on my 7950x3d on my x670e Carbon so I can not confirm or deny there. See what @Raiderman or @jaybee83 get doing the same since they have a Carbon and Hero respectively.

 

 

[Mr. Fox]

22 minutes ago, electrosoft said:

 

Whew, pressing it nicely! 285k and 9950x are turning out some decent scores!

 

I never tried to set different ratios between ccds on my 7950x3d on my x670e Carbon so I can not confirm or deny there. See what @Raiderman or @jaybee83 get doing the same since they have a Carbon and Hero respectively.

 

 

I reached out to @chew on oc.net (chewonthis) and he says it works, but it absolutely does not work or something I am setting due to lack of understanding of how AMD firmware works is making it not work. (I am thinking the BIOS is bugged.)

 

I am still within my return Windows and seriously debating whether I should RMA this and get the Taichi instead. Overall I really like it, but I expected a little better in terms of BIOS tuning function. I could also be jumping from the frying pan to the fire and end up regretting the switch. I should probably poke around in the oc.net X870E Taichi forums first and see what, if anything, people are complaining about.

 

I need to ping @Raiderman about something else (not computer related).

[tps3443]

I could not resist this. Brand-new Xeon 3175X sealed. $380 bucks. 
 

Snatches! 
 

Planning to test and compare with mine, and sell one of them. 
 

5e dice roller

 

5e dice roller

 

5e dice roller

[Mr. Fox]

18 minutes ago, tps3443 said:

I could not resist this. Brand-new Xeon 3175X sealed. $380 bucks. 
 

Snatches! 
 

Planning to test and compare with mine, and sell one of them. 
 

5e dice roller

 

5e dice roller

 

5e dice roller

That is a great deal. Congrats.

---

These silver Byski RAM sinks and aluminum heatsink on top fit the Edge aesthetic nicely. It looks better than the black ones.