Ryzen Owners Zen Garden

[Nordic]

It's not that I don't believe you, but I just really have a hard time believing that either of these situations (tuned 6000 vs 6200, tuned 6000 vs 6400) will result in more than 3% of real performance. And on X3D, no less. Maybe if there is a specific well-controlled in-game benchmark that you can run.

Not even tuned vs XMP, usually results in massive double digit real world gains.

As to Star Citizen, not a space simmer but only thing I've ever heard people say about performance is that it is optimized like garbage.

My synthetic testing was specifically done with Microbenchmarks. I can share my data table with setting and results when I get home. I did no in game testing myself. I do not know how much in game or other real work loads performance has improved.

While I cannot speak to the specifics of the memory performance testing with Star Citizen that showed a 20% improvement I can confirm that it is a very unoptimized game. Star Citizen is an alpha and until the last year has been minimal performance optimization. The game requires an NVME SSD and 32gb memory to not have a stuttery slide show. I don't know the technical reasons why but Star Citizen is a memory intensive game. Memory bandwidth is very impactful.

There is no in game bench mark. Everyone I know of doing extensive in game performance testing does many runs in the same locations and aggregates the data using CapFrameX. This is the best we can do to test performance right now.

 

[AusWolf]

Perhaps it's time to tried a direct die solution and better thermal interface like Kryosheet/LM. Stock IHS is a big part of the thermal problem on AM5.

Or you could try the TG High Performance heatspreader, since you've already delidded, @dgianstefani has one

I'm not one to complain about the IHS, but why do we have the compute die, the hottest part of the package squished into a tiny corner? The IO die has no business being in the middle, imo.

 

[tabascosauz]

I'm not one to complain about the IHS, but why do we have the compute die, the hottest part of the package squished into a tiny corner? The IO die has no business being in the middle, imo.

IOD has always been pretty big so maybe they couldnt figure out a layout that way

I don't think it really matters where the die is unless you have a direct heatpipe cooler. CCD is still going to be that small.

 

[Space Lynx]

i get 2000 less points than other people in 3dmark timespy, and lower scores than everyone. I think I just got a bad bin. its a 4th week of 2024 made cpu, so eh i don't know. i am a little disappointed, but i guess this doesn't really translate to gaming so it doesn't matter just annoying

 

[Super Firm Tofu]

i get 2000 less points than other people in 3dmark timespy, and lower scores than everyone. I think I just got a bad bin. its a 4th week of 2024 made cpu, so eh i don't know. i am a little disappointed, but i guess this doesn't really translate to gaming so it doesn't matter just annoying

You can't compare to the person who posted below you in the TS thread. His CPU and memory are juiced much further than most people could ever get to.

 

[tabascosauz]

i get 2000 less points than other people in 3dmark timespy, and lower scores than everyone. I think I just got a bad bin. its a 4th week of 2024 made cpu, so eh i don't know. i am a little disappointed, but i guess this doesn't really translate to gaming so it doesn't matter just annoying

I'd sooner blame your board than your CPU. Clearly it's a case of "won't" rather than "can't" meet a certain clock.

It has also been pretty disheartening seeing Hydra humble my previously "stable" CO settings, but stability > all.

Just get back to gaming and be happy.

 

[Space Lynx]

You can't compare to the person who posted below you in the TS thread. His CPU and memory are juiced much further than most people could ever get to.

I just looked again and you are right, he has an extreme oc and insane ram timings.

Thanks for pointing that out, I don't feel so bad now. my scores are where they should be then, its like 0.5% less than other 7800x3d that i can see, so yeah not end of world. im going to run aida64 latency test today to see what my ns, just to make sure that all looks good, then i am done.

I agree with you both though, time to game and forget the rest, its not like its going to matter for casual 1440p backlog gaming. lol

 

[tabascosauz]

"Gold", it says my ass

At least it works......most of the time, at least. Enough to be semi-automated at least. Couldn't say the same for its predecessor CTR.

 

[Space Lynx]

is 70ns normal for this setup on the ram? its just expo no other changes. numbers look good?

 

[Nordic]

It's not that I don't believe you, but I just really have a hard time believing that either of these situations (tuned 6000 vs 6200, tuned 6000 vs 6400) will result in more than 3% of real performance. And on X3D, no less. Maybe if there is a specific well-controlled in-game benchmark that you can run.

My synthetic testing was specifically done with Microbenchmarks. I can share my data table with setting and results when I get home. I did no in game testing myself. I do not know how much in game or other real work loads performance has improved.

I believe Micro Benchmarks should be the standard for memory benchmarking. It is the only memory benchmark I have found that has low variance in results. Aida is almost useless as a benchmark. I learned about this benchmark from Buildzoid who also says it is the best memory benchmark.

Release Add PCIe BW Test · clamchowder/MicrobenchmarksGui

Test copy bandwidth between the CPU and GPUs Allow test size editing and starting at a certain size Better window resize handling Fix an issue with CPU bandwidth calculation

github.com

Each of the following scores are 10 run averages.

 

[tabascosauz]

I believe Micro Benchmarks should be the standard for memory benchmarking. It is the only memory benchmark I have found that has low variance in results. Aida is almost useless as a benchmark. I learned about this benchmark from Buildzoid who also says it is the best memory benchmark.

Release Add PCIe BW Test · clamchowder/MicrobenchmarksGui

Test copy bandwidth between the CPU and GPUs Allow test size editing and starting at a certain size Better window resize handling Fix an issue with CPU bandwidth calculation

github.com

Each of the following scores are 10 run averages.

I don't disagree, we've all known a long time that AIDA is not representative of actual memory performance. My point was that this is a pretty apples to oranges comparison to compare 6000 XMP to 6200 tuned, and even worse if your FCLK was different between the two. I was just saying that there is not nearly as much difference to running 6400 if you run a proper 6000 profile.

My 6000 profile is somewhere in the mid 60.5ns region and it's still on 50k tREFI and XMP primaries.

 

[AusWolf]

IOD has always been pretty big so maybe they couldnt figure out a layout that way

I don't think it really matters where the die is unless you have a direct heatpipe cooler. CCD is still going to be that small.

Yeah, but being in the middle, it could spread its heat more evenly over the IHS, methinks.

 

[Nordic]

I was just saying that there is not nearly as much difference to running 6400 if you run a proper 6000 profile.

I don't disagree here. I am hoping for more than 6400mhz from the 9800x3d unlikely as it may be.

 

[tabascosauz]

Yeah, but being in the middle, it could spread its heat more evenly over the IHS, methinks.

Maybe. But probably not enough to matter with the thick AM5 IHS in the way. That thing is truly the biggest pest of all time.

I have been tempted by the TG high performance IHS recently

 

[Space Lynx]

I turned off memory integrity through settings and hyper-v through control panel.

My scores are better now across the board.

Sigh, I hate modern computers, so many variables it just annoys me at this point. I think that is why I love Steam Deck so much, gave me that console experience back and I loved it.

Even my latency from 70-71ns to 67-68ns on a few runs.

I hope there are no downsides to leaving memory integrity off for a casual gamer, cause I am going to leave it off lol

 

[lexluthermiester]

I hope there are no downsides to leaving memory integrity off for a casual gamer, cause I am going to leave it off lol

Shouldn't be. A lot of things that are not needed are enabled by default in Windows when they should only be running when needed.

 

[Cifu]

How were your temps originally?

I agree on the direct die cost, but you've already delidded. It would honestly be a waste to do all that just to go back to the original crappy IHS. Those aren't exactly made to very exacting physical standards. I only know of the TG IHS, idk if companies like rockitcool make anything for AM5.

Settings:
PPT Limit: 115W
CPU RPM: 51%
PBO CO: -10 all core

20240503 -> pre-delid
20240510 -> after delid

 

[TheLostSwede]

is 70ns normal for this setup on the ram? its just expo no other changes. numbers look good?

View attachment 347549

Tweak your tRAS, tRC, tFAW and tRFC, it'll make a huge difference.

 

[Chomiq]

Settings:
PPT Limit: 115W
CPU RPM: 51%
PBO CO: -10 all core

20240503 -> pre-delid
20240510 -> after delid

So your temps went up after delid, great!

 

[Cifu]

So your temps went up after delid, great!

That's my problem. I even recorded the room temps, but less than 1°C change noted - that's can't be the reason of the positive difference.

 

[Veseleil]

That's my problem. I even recorded the room temps, but less than 1°C change noted - that's can't be the reason of the positive difference.

Why not direct die, since you've already messing with this? Sorry if I missed something.

 

[Cifu]

Why not direct die, since you've already messing with this? Sorry if I missed something.

A guy who doing delid for Intel CPU's mostly gave me an opportunity to check it, because most 7800X3D which delided won't show too much improvement.
I say why not, I won't expect too much, if 5-7°C thermal drops happens that's not a bad thing.
Actually the delid won't help but worsen my temps.

The guy itself not an amateur but only have few opportunity for AM5 delid, I don't really think he messed up anything.
I don't want anything originally my CPU, I was a happy user in this regard with 115W PPT limit and 80-85°C top temperature.

So what I don't really want now to began experiencing with direct die.
If I want better gaming experience I change my CPU for a 7800X3D (or so).
But @4K gaming with an RTX 4080 I won't have any CPU limit even with my current 7700X.

 

[lexluthermiester]

That's my problem. I even recorded the room temps, but less than 1°C change noted - that's can't be the reason of the positive difference.

I don't want to seem like I'm raining on your parade, however, delidding these days is more or less useless as both Intel and AMD are soldering the IHS to the dies, as a general rule. The problem with the Ryzen 7000 series is a combination of the thickness of the IHS as well as the alloys being used to make them. Neither are optimal.

 

[Space Lynx]

I don't see why you guys need to delid at all. My 7800x3d in games doesn't even break 55 celsius most of the time, though they are older games like Sleeping Dogs, but even Guardians of the Galaxies completely maxed out 165hz 165 fps 1440p... I didn't even budge past 68 celsius. I absolutely love this rig

 

[tabascosauz]

I don't want to seem like I'm raining on your parade, however, delidding these days is more or less useless as both Intel and AMD are soldering the IHS to the dies, as a general rule. The problem with the Ryzen 7000 series is a combination of the thickness of the IHS as well as the alloys being used to make them. Neither are optimal.

Agreed. It's a good start, but to make the most of a delid on just about any platform these days there really needs to either be a custom heatspreader or direct die.