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

[Mr. Fox]

20 minutes ago, tps3443 said:

So I’m pushing 1:1 speeds to check IMC limit and fabric limits. 
 

I can run 2,167 fclk which seems to work well, and DDR5 6600c32 1:1. Right now I’m trying to dial down to C26 or maybe c28. 

Did you try BIOS 9961 yet to see if that helps? Here are my older and tighter timings. I just reapplied them and ran a couple of benchmarks. They need to be loosened a bit to keep stability on AGESA 1.3.0.1 or newer. tRCDRD, tRP and tRC are too tight with newer AGESA, at least for my CPU. But, the read/write/copy and latency are better with the older BIOS even with the same timings and using "Normal" Bank Refresh. With "Mixed" Bank Refresh, performance degrades measurably. (Mixed is only a workaround for a security concern and I refuse to use it due to the performance hit.)

 

If you are able to run 6600 C32 1:1 that is better than what most people can accomplish with their CPU, mobo and memory kits. 

[Mr. Fox]

12 hours ago, tps3443 said:

What about the faster 1:1 speeds? Im testing 6600c28 right now. How would those compare VS. like DDR5 8000+ in 1:2 speeds? 

8000 2:1 with CL34 and other tight timings is slightly better than 6400 C30 and other tight timings. Bandwidth is higher and latency a bit lower. They are close enough to probably make no meaningful difference in experience, but based on my testing the idea that slower is better on AMD is a myth and running 6400 1:1 is pointless if your system is capable of running 8000 or higher with tight timings. Testing done by Blackbird PC Tech also confirms what I found: 8000 is slightly better (smoother and more consistent FPS) than 6400 for gaming, even with an X3D CPU, but not by a lot.

[tps3443]

15 minutes ago, Mr. Fox said:

Did you try BIOS 9961 yet to see if that helps? Here are my older and tighter timings. I just reapplied them and ran a couple of benchmarks. They need to be loosened a bit to keep stability on AGESA 1.3.0.1 or newer. tRCDRD, tRP and tRC are too tight with newer AGESA, at least for my CPU. But, the read/write/copy and latency are better with the older BIOS even with the same timings and using "Normal" Bank Refresh. With "Mixed" Bank Refresh, performance degrades measurably. (Mixed is only a workaround for a security concern and I refuse to use it due to the performance hit.)

I almost bought that Vcolor kit! I thought it was suspiciously priced low. I suppose that’s why? Because it’s dual rank 16gigs? 
 

Have not tried the bios yet. But, I will soon. Being lazy. I am testing a 6600c28 though! 

[Mr. Fox]

28 minutes ago, tps3443 said:

I am testing a 6600c28 though

If you can run 6600 C28 in 1:1 mode that is already more than what most people can accomplish with the CPU, mobo and RAM kit they have. Many can't even run 6400 1:1 with the crummy silicon they were blessed with. Your CPU would have a very good IMC to do that, so I am not sure why you are having issues with 8000, 8200 or 8400 unless you are not using enough VDD/VDDQ/VDDIO. If you are able to run 6600 C28 1:1 then I doubt there is anything wrong with your motherboard.

28 minutes ago, tps3443 said:

I almost bought that Vcolor kit! I thought it was suspiciously priced low. I suppose that’s why? Because it’s dual rank 16gigs?

I'm glad you did not buy it. I think it would be fine for an average Joe type of gamerboy that just sets XMP/EXPO and moves on without bothering to actually tune anything for maximum performance. The plastic covers that look like a heatsink are a joke though. Gupsterg (oc.net member) discovered a V-COLOR kit he had purchased had plastic covers as well. It kind of pisses me off that they would even sell garbage like that. What a ripoff. I think the marketing info and specs not disclosing the dual rank design and plastic covers qualifies as a scam and I think they are very conveniently not disclosing what they know would make it a hard to sell product.

 

I debated whether or not to just go ahead and install it anyway in my son-in-law's build with the other upgrade parts. The performance would have been fine for him, but I think the high temps could have been an issue with the plastic covers. I decided he was better off with the generic green 2x8GB Hynix A-die with nice aftermarket heatsinks for stability and reliability. Things get kind of toasty in his SFF build with an air-cooled GPU and air-cooled space-heater mounted on top of his CPU, both working together to create a toastly little hot box. He doesn't need 32GB of RAM. 16GB will handle anything he would throw at it just fine.

[electrosoft]

7 hours ago, tps3443 said:

What about the faster 1:1 speeds? Im testing 6600c28 right now. How would those compare VS. like DDR5 8000+ in 1:2 speeds? 

 

If you're able to do 6600 anything 1:1 that is the golden level for IMC  compatibility. Most chips can't do 6600 with proper stress testing. For example, I can post 6600 but can't pass anything meaningful but 6400 is no problem. If you're able to somehow dial in 2200/6600 that is literally the golden standard of G2.

 

7 hours ago, Mr. Fox said:

I did not notice any difference, but I did not do much testing other than Karhu and AIDA64. They seemed about the same as single rank to me with the minimal testing. They perform as expected up to 6400. But, otherwise this kit is pure junk. The chips on both sides doubles their heat production and I am pretty sure the covers on them are plastic, not metal. They were cheap by current overpriced standards, and I can see why. I am going to keep them only to use in case of emergency (DIMM failure). Below is a photo of the label. Avoid these.

 

You *might* have a not so great 8000+ kit on your hands but a really good 1:1 6400 dual rank set in your possession @Mr. Fox

 

If you're able to do 2133/6400 on dual rank that is something major for my low end use case/testing and I would love to see some gaming and bandwidth/latency benchmarks for testing. I've had several dual rank 6000 kits that were trash for 8000 and couldn't hit  2133/6400 for any meaningful stress testing. Do they pass 2133/6400 dual rank Karhu and Aida64?

 

 

[Mr. Fox]

4 hours ago, electrosoft said:

You *might* have a not so great 8000+ kit on your hands but a really good 1:1 6400 dual rank set in your possession @Mr. Fox

 

If you're able to do 2133/6400 on dual rank that is something major for my low end use case/testing and I would love to see some gaming and bandwidth/latency benchmarks for testing. I've had several dual rank 6000 kits that were trash for 8000 and couldn't hit  2133/6400 for any meaningful stress testing. Do they pass 2133/6400 dual rank Karhu and Aida64?

I will have to do more testing once I decide where the Z890 Apex is going to land and if decide that I am going to disturb the AMD builds. Depending on how the Intel testing goes, I anticipate the X870E taking the 4585PX from the B850MPOWER and going into the O11 Mini with the 5080, the Z890 Apex going into the O11 XL with the 5090 and the B850MPOWER taking the 9950X and going onto the test bench with my spare AIO and air-cooled memory. If the move back to Intel doesn't impress me the Z890 Apex will remain on the test bench, but that seems unlikely. I anticipate the change will be welcomed since I have plateaued on my current hardware and my interest is beginning to wane. If I like it, I will end up having to purchase a delid tool and heater any Mycro block from TG because it is extremely unlikely that I will find contentment with anything less than bare die.