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

[electrosoft]

5 hours ago, jaybee83 said:

 

yeah thats the tachyon though, seems like. barely available anywhere, but good to know u have more than one 😄 

i for one just crapped out on the cpu imc and/or the mobo, cant get 6400 or 8000 stable for the life of me. so now im tightening things up for 6200 and 7800 instead.

i do see significant improvements in timings tho with the new 8200-48gb kit vs. my previous 6600-32gb kit.

 

3 hours ago, Raiderman said:

If you can run PBO voltage curve at -30 all cores with CB 23, I would say it's your board and not imc. I am bench stable at 8000 xmp on, and 6400. I have benched 6600, but not stable. Of course that's only with very limited time playing with it.

 

3 hours ago, Mr. Fox said:

 

If both of your motherboards are 4-DIMM and the Tachyon is 2-DIMM that more than likely accounts for most, if not all, of your limited memory overclock capacity if the scenario is the same as Intel. I'm not aware of any amount of skill that allows a 4-DIMM Intel mobo to match 2-DIMM models. I think it's totally illogical that they even manufacture 4-DIMM enthusiast mobos. I would avoid 4-DIMM boards like a plague. This is why the Tachyon, Unify-X, Dark and Apex boards rule, at least on the Team Blue side of the house. They should stop manufacturing 4-DIMM boards for gaming and enthusiast applications. They are a waste of money.

 

My 12th Gen Celeron 2C/2T netbook CPU runs 8000 stable on the Dark and Apex. IMC and mobo quality certainly matters, but not as much as topology.

 

100% spot on.

 

This can not be overstated enough when you're really pushing your memory regardless of CPU platform.

 

When the new gear settings were announced, I posted quite awhile back even calling into the equation motherboard quality as a factor now with frequencies so high along with what real world results we would actually see which so far outside of benchmarks seems to be low/limited for AMD but hey take every bit you can get!

 

 

[chew]

6 hours ago, jaybee83 said:

 

yeah thats the tachyon though, seems like. barely available anywhere, but good to know u have more than one 😄 

i for one just crapped out on the cpu imc and/or the mobo, cant get 6400 or 8000 stable for the life of me. so now im tightening things up for 6200 and 7800 instead.

i do see significant improvements in timings tho with the new 8200-48gb kit vs. my previous 6600-32gb kit.

 

24x2M is going to be slower than a good 2x16gb kit.

 

M die shines with Gdm on but gdm on is slow. You can get away with alot more gdm  on.

 

GDM off the main things are trtp needs to be 16 or higher. you might have to back off twr. almost guaranteed trrd ls tfaw needs be 8/12/32. You will lose some trcd.  you will lose some CL.

 

I can run 8000 36/47/47 170ns gdm on  8000 np.....

 

When I run karhu though mb/s is in the 300 range. 

 

GDM off that gets knocked to 38/48/48 but gets me 319mbs in karhu

 

On A die with a good tune typically  you want to be and should be in the 330mbs range on a 7950x and 260-270mbs on single ccd.

 

4 hours ago, Mr. Fox said:

 

If both of your motherboards are 4-DIMM and the Tachyon is 2-DIMM that more than likely accounts for most, if not all, of your limited memory overclock capacity if the scenario is the same as Intel. I'm not aware of any amount of skill that allows a 4-DIMM Intel mobo to match 2-DIMM models. I think it's totally illogical that they even manufacture 4-DIMM enthusiast mobos. I would avoid 4-DIMM boards like a plague. This is why the Tachyon, Unify-X, Dark and Apex boards rule, at least on the Team Blue side of the house. They should stop manufacturing 4-DIMM boards for gaming and enthusiast applications. They are a waste of money.

 

My 12th Gen Celeron 2C/2T netbook CPU runs 8000 stable on the Dark and Apex. IMC and mobo quality certainly matters, but not as much as topology.

 

Depends on how the board is tuned.....depends on who is tuning the board also depends on who is "working" on the boards deep level tuning at the bios level......sacrifices can be made elsewhere bios level if your willing to make them.

Extreme is weak on the DS dimms front atm.....I wonder why 😉

 

What i can tell you is AMD is in a better place than intel with there 4 dimm boards currently.....however they just did refreshes so subject to change quickly.

[chew]

Here's some super low bin non binary  M die on intel. Typically this translates to GDM  on capability on AMD.

 

Intel is useful as a template to figure out what you can pull off on AMD.

 

Can't be a true  mem guy unless your versatile and since intel has  been doing gear 2 longer can learn alot of things to carry over and test on AMD.

 

[tps3443]

I ordered my Super Cool DDR5 waterblock kit (Real memory watercooling)! 😁 

 

My 2x24GB M-die’s will be on chilled water now 😂

 

Man oh man. I’ll be able to pass just about any memory stability test for however long so easily lol. 🤣

 

I am seriously excited for some DDR5 running like 15C lol. This is gonna be so stupidly glorious 😭

[Mr. Fox]

1 hour ago, tps3443 said:

I ordered my Super Cool DDR5 waterblock kit (Real memory watercooling)! 😁 

 

My 2x24GB M-die’s will be on chilled water now 😂

 

Man oh man. I’ll be able to pass just about any memory stability test for however long so easily lol. 🤣

 

I am seriously excited for some DDR5 running like 15C lol. This is gonna be so stupidly glorious 😭

It looks like all they have is the jackets for the modules, but not the block that goes on top. Were you able to snag the block that mounts on top of the RAM jackets as well?

[tps3443]

1 hour ago, Mr. Fox said:

It looks like all they have is the jackets for the modules, but not the block that goes on top. Were you able to snag the block that mounts on top of the RAM jackets as well?

it includes everything. I’d order asap but only 2 left. I’ve been waiting on these for 1.5 years. Finding them used is close to impossible. (It’s actually impossible lol) people apparently asked him to make another batch of these for ram, and he finally had enough request(s) and he actually did it. His stuff is amazing though.
 

The ram kit includes the blocks for (2) ram sticks for both sides, and the top manifold that directs water and mounts to the both ram blocks. He has a cheaper kit also which is a full kit too, but includes the less expensive clear polymer manifold instead of the solid copper manifold.

 

And @Mr. Fox you only need (1) and it’s an entire kit, so either the one for $79, or the one for $93.

This is the $79 dollar kit.

 

[Mr. Fox]

28 minutes ago, tps3443 said:

it includes everything. I’d order asap but only 2 left. I’ve been waiting on these for 1.5 years. Finding them used is close to impossible. (It’s actually impossible lol) people apparently asked him to make another batch of these for ram, and he finally had enough request(s) and he actually did it. His stuff is amazing though.
 

The ram kit includes the blocks for (2) ram sticks for both sides, and the top manifold that directs water and mounts to the both ram blocks. He has a cheaper kit also which is a full kit too, but includes the less expensive clear polymer manifold instead of the solid copper manifold.

Ordered. Thanks for the heads up. It's about twice as much ($112 USD including shipping) as my other liquid cooled memory components cost me, but it will be interesting to see if keeping the memory below 30°C improves overclocking. If not, then I will know not to spend extra next time. Also nice that now I can get rid of the nasty G.SKILL RGB heating blankets and rainbow puke for good. I really hate RGB memory. Such a stupid thing, LOL. Finding good memory options without RGB garbage should be much easier. 

[tps3443]

3 minutes ago, Mr. Fox said:

Ordered. Thanks for the heads up. Now I get get rid of the nasty G.SKILL RGB heating blankets and rainbow puke for good.

It’s an amazing deal for the kit!! And typically he only does small batches of these products, so when they are gone that’s it. I’m so ready to dump mine as well though. I’ve been really enjoying the 2x24 (7200) but during summer time I had to reduce MEM overclocks due to ambient heat.

 

This is another pic of the top of the ram with block mounted. These are pretty much built for the Apex and as good as ram cooling gets. 
 

[Mr. Fox]

Just now, tps3443 said:

It’s an amazing deal for the kit!! And typically he only does small batches of these products, so when they are gone that’s it. I’m so ready to dump mine as well though. I’ve been really enjoying the 2x24 (7200) but during summer time I had to reduce MEM overclocks due to ambient heat.

 

This is another pic of the top of the ram with block mounted. These are pretty much built for the Apex and as good as ram cooling gets. 
 

The real benefit will be seen in what, if anything, it does to extend my memory overclocking. At 8200 I seldom see more than 32-34°C under stress testing with the liquid cooled memory on the Z690 Dark. It is definitely not a bad price for what a complex design it is with the water flowing through the RAM jackets. Active cooling directly on the memory ICs is pretty amazing when you think about it. The price is really decent, as long as it yields a benefit on the overclocking capacity. If it doesn't then it will just be an extra-fancy setup that I spent more on than I needed to.

[tps3443]

3 minutes ago, Mr. Fox said:

The real benefit will be seen in what, if anything, it does to extend my memory overclocking. At 8200 I seldom see more than 32-34°C under stress testing with the liquid cooled memory on the Z690 Dark. It is definitely not a bad price for what a complex design it is with the water flowing through the RAM jackets. Active cooling directly on the memory ICs is pretty amazing when you think about it. The price is really decent, as long as it yields a benefit on the overclocking capacity. If it doesn't then it will just be an extra-fancy setup that I spent more on than I needed to.

I’ve heard they are capable of running the same temp as the water circulating under load. I’ve heard this so many times, no delta in temps. We’ll see though! I’m excited to see what 15C ram temps will provide.
 

 

[Mr. Fox]

1 hour ago, tps3443 said:

I’ve heard they are capable of running the same temp as the water circulating under load. I’ve heard this so many times, no delta in temps. We’ll see though! I’m excited to see what 15C ram temps will provide.
 

 

I will not be surprised to learn that it is accurate. It's not very hard to cool memory chips and I could see having no delta being relatively easy to achieve with water flowing through the jackets. My liquid cooled memory on the Z690 Dark is rarely ever more than a 5°C delta from the water temperature with only the block on top to passively cool the memory jackets. The block on top alone drops the memory load temps about 30°C.

 

The issue is that memory starts having errors at relatively low temperatures. But they don't get terribly hot, which is why the half-assed heating blankets most of the retail kits ship with is so inexcusable. My naked green sticks run cooler with nothing but a fan blowing on them than expensive retail kits do with so-called "heatsinks" on them that actually tend to increase the temperature rather than reduce it. 

[tps3443]

I’ve really cranked the clocks on this 3090 for every frame I can possibly get. I’m kinda of surprised this is even working lol. Let it run cool enough, and this 2,205-2,220Mhz boost in game stays locked in for hours and hours and hours. My CPU is also at 6.1Ghz. Also got my 23Gbps on my GDDR6X. This GPU is really so fantastic for holding up to such abuse for so many hours. Pushing that 50 hour playtime mark in Starfield. I really need to throw on some Liquid Metal and gain another 15-30+MHz out of my GPU. I kind of like the thermal paste due to being able to neglect it though. Since 2021, I’ve had this card pulled apart at least 2 times for maintenance. 
 

 

 

[chew]

2 dpc........

 

And I maintained performance.

 

 

[jaybee83]

19 hours ago, Raiderman said:

If you can run PBO voltage curve at -30 all cores with CB 23, I would say it's your board and not imc. I am bench stable at 8000 xmp on, and 6400. I have benched 6600, but not stable. Of course that's only with very limited time playing with it.

 

havent really started with tuning the CPU yet tbh 😄 might be stupid questions but: how would an undervolt on the cpu cores say anything about the IMC? different dies....

as for stability, sure, benching is no problem with 6400 and 8000, can do that all day long. im testing with anta777 absolut tm5 tho, and thats not passing haha. i dont want to end up with random BSODs down the line popping up in a few weeks / months time...

 

13 hours ago, tps3443 said:

I ordered my Super Cool DDR5 waterblock kit (Real memory watercooling)! 😁 

 

My 2x24GB M-die’s will be on chilled water now 😂

 

Man oh man. I’ll be able to pass just about any memory stability test for however long so easily lol. 🤣

 

I am seriously excited for some DDR5 running like 15C lol. This is gonna be so stupidly glorious 😭

 

9 hours ago, tps3443 said:

it includes everything. I’d order asap but only 2 left. I’ve been waiting on these for 1.5 years. Finding them used is close to impossible. (It’s actually impossible lol) people apparently asked him to make another batch of these for ram, and he finally had enough request(s) and he actually did it. His stuff is amazing though.
 

The ram kit includes the blocks for (2) ram sticks for both sides, and the top manifold that directs water and mounts to the both ram blocks. He has a cheaper kit also which is a full kit too, but includes the less expensive clear polymer manifold instead of the solid copper manifold.

 

And @Mr. Fox you only need (1) and it’s an entire kit, so either the one for $79, or the one for $93.

This is the $79 dollar kit.

 

 

9 hours ago, Mr. Fox said:

Ordered. Thanks for the heads up. It's about twice as much ($112 USD including shipping) as my other liquid cooled memory components cost me, but it will be interesting to see if keeping the memory below 30°C improves overclocking. If not, then I will know not to spend extra next time. Also nice that now I can get rid of the nasty G.SKILL RGB heating blankets and rainbow puke for good. I really hate RGB memory. Such a stupid thing, LOL. Finding good memory options without RGB garbage should be much easier. 

 

lulz, please keep us updated on your results guys. curious to see if going even lower in temp will improve OCability at all. or might behave like current vRAM and lose clocks once u reach certain low temps...

 

1 hour ago, chew said:

2 dpc........

 

And I maintained performance.

 

 

 

welp, im slowly thinking my bottleneck is the classic PEBCAK 😄 now please do the same with the Asus Extreme 😛 

 

so what kinda bandwidth and latencies are we talking here, anyways? have u compared max clocks between 1:1 and 1:2?

[ryan]

You fellas wouldn't happen to know how to get into the advanced bios of the omen 16 11800h version would you. All this talk about ram timing is making me itchy. Everything on my laptop is modified except screen and ram..man this laptop is built like a tank. Been running max fans and overclocking it for 2 years and 0 issues

[jaybee83]

53 minutes ago, ryan said:

You fellas wouldn't happen to know how to get into the advanced bios of the omen 16 11800h version would you. All this talk about ram timing is making me itchy. Everything on my laptop is modified except screen and ram..man this laptop is built like a tank. Been running max fans and overclocking it for 2 years and 0 issues

 

i assume uve already gone through all available bios options? if theyre not in there, they would need to be manually exposed by modifying the bios and flashing it....

[ryan]

I found a way but it looks locked, and I think Talon mentioned it has to be modified. sucks just hoping there was a way around thats a little less risky

 

here is cinebench 2023

 

not bad no overclock

[tps3443]

3 hours ago, ryan said:

I found a way but it looks locked, and I think Talon mentioned it has to be modified. sucks just hoping there was a way around thats a little less risky

 

here is cinebench 2023

 

not bad no overclock

That is a seriously difficult test to run while overclocked.  I messed around with it last week for a bit, and I had to increase CPU voltage by +0.035 just to be able to reliably run it. 

[Talon]

 

Now to get these stable. They game for hours in BF2042 without issue, pass Geekbench, and haven't had any issues using them this way, but will eventually error out TM5. Hard to keep these dual ranks cool even with air. If I push the cache to 50x gets the latency down to 52ns. Not too shabby for 2x32gb kit. 

 

Took that picture hiking with the wife in Yosemite earlier this year. Gotta touch grass every now and then. 😁

[tps3443]

14 minutes ago, Talon said:

 

Now to get these stable. They game for hours in BF2042 without issue, pass Geekbench, and haven't had any issues using them this way, but will eventually error out TM5. Hard to keep these dual ranks cool even with air. If I push the cache to 50x gets the latency down to 52ns. Not too shabby for 2x32gb kit. 

 

Took that picture hiking with the wife in Yosemite earlier this year. Gotta touch grass every now and then. 😁

Yeah that picture is awesome. Looks like an area where a Bigfoot/Sasquatch would be standing lol. 
 

I hear and read about something interesting, and that was about PCH temps on the Z790 Apex. If you can drop those temps it greatly increases memory stability. I believe one of Sugi’s buddies discovered this. And a few others confirmed. They all started water cooling their PCH 😂 Anyways, maybe that’s a start. 

[chew]

10 hours ago, jaybee83 said:

 

havent really started with tuning the CPU yet tbh 😄 might be stupid questions but: how would an undervolt on the cpu cores say anything about the IMC? different dies....

as for stability, sure, benching is no problem with 6400 and 8000, can do that all day long. im testing with anta777 absolut tm5 tho, and thats not passing haha. i dont want to end up with random BSODs down the line popping up in a few weeks / months time...

 

 

 

 

lulz, please keep us updated on your results guys. curious to see if going even lower in temp will improve OCability at all. or might behave like current vRAM and lose clocks once u reach certain low temps...

 

 

welp, im slowly thinking my bottleneck is the classic PEBCAK 😄 now please do the same with the Asus Extreme 😛 

 

so what kinda bandwidth and latencies are we talking here, anyways? have u compared max clocks between 1:1 and 1:2?

Typically speaking If you keep temps under 58c most A die will not produce errors. I just pull rgb difusers off so i can actually cool memory and not a piece of plastic......while the supercool setup is badass........I change memory like women change shoes.....it's just not realistic for the test station. Better suited for having binned many kits and then installing on a known kit and shoving in a daily.

 

To further explain this lets take a look at some testing i did.......

 

Test 1 low speed fan......as you can see max temp hit 58c......which resulted in an error in y cruncher fft on amd this is ( vst to intel ) . Test 2 server grade finger chopper fan......was able to maintain a temp under 58c while at 2.15 vdd.....

 

I ran 10 mins because for a benchmark like 32m I need 5 mins......if i can survive 10 I can "benchmark" anything. Also y cruncher fft tends to run mem hotter than any memtest on AMD i can run so this is fairly worst case.......

 

 

Performance is a trade blows scenario in most cases.....I use good old 32m pi to really see what's what. Until we get working multis at 82x I'm reserving the right to say that typically in everyday workloads 1/1 is just going to be easier and faster........If my math is right cl 32 8200 will be the break even or surpass point unfortunately although I can test it once you introduce eclk into the mix it brings effective clock issues skewing data so I really need a working 82x to do a fair comparison. Can't pull off 8000 c30 and cl32 8000 is actually slower by 1/10th.

 

Please note vcache cpu's are 10 seconds faster cpc......this is specifically for non vcache 5 gig low clock challenge....which I own both best results still to this date for vcache or non vcache......

 

[Mr. Fox]

2 hours ago, tps3443 said:

I hear and read about something interesting, and that was about PCH temps on the Z790 Apex. If you can drop those temps it greatly increases memory stability. I believe one of Sugi’s buddies discovered this. And a few others confirmed. They all started water cooling their PCH 😂 Anyways, maybe that’s a start. 

Yes, that is true. You can lower the PCH temps by enabling power management in the BIOS for ACPI/ASPM/DMI. Same was true on the Unify-X and Crosshair VIII X570. Both of them had a crazy hot PCH and it does hinder memory overclocking stability. So did the Strix Z690-E. Not sure why they don't do a better job with the heat sinks. The Z690 Dark has a very cool PCH, but the heat sink for it weighs almost as much as the entire Apex mobo, LOL. I actually used a big heatsink with fans and a thermal pad on top of the stock heatsink on the Unify-X before I found out about the power management in the BIOS being disabled causing the PCH temps to soar.

 

7 hours ago, tps3443 said:

That is a seriously difficult test to run while overclocked.  I messed around with it last week for a bit, and I had to increase CPU voltage by +0.035 just to be able to reliably run it. 

Yes it is. Not sure why. I think there is something wrong with the benchmark to be honest. I think it is doing something weird with cache or memory. It's not as stressful as Cinebench R23. My CPU doesn't get as hot as R23, but it still takes lots of extra voltage for some reason. It's strange. I haven't tried lowering the cache ratio or decreasing the memory overclock to see if the behavior changes, but something unusual is going on. It seems to waste a bit of time "preparing" for the benchmark before it starts. Not sure what that's about. Sometimes it won't start rendering. Sometimes it stops responding without crashing or freezing. Other times it runs normally. I think it's just buggy.

 

On top of that, I find it really annoying to have to remember to manually change it from a timed stress test to benchmark using the "off" option. I just tried to open it and got a BSOD a minute ago. Yet, I can run any other version of Cinebench, OCCT or AIDA64 stress test without issue, so yeah, it's buggy.

[Mr. Fox]

10 hours ago, tps3443 said:

That is a seriously difficult test to run while overclocked.  I messed around with it last week for a bit, and I had to increase CPU voltage by +0.035 just to be able to reliably run it. 

I can't get it to work correctly except using static voltage. Adaptive I think it gets too much VDROOP even using L8 LLC. If I could disable Guardband on the Apex I think it would do better. I run with that disabled at all times on the Z690 Dark. I will have to run it on the 13900K in the Dark and see how it does. I'm tempted to put the Dark back on the open bench and make the Apex my "work" PC. I can't figure out how to disable Guardband on the Apex. I can adjust it, but I don't see a way to completely disable it.

---

CPU: https://hwbot.org/submission/5351245_ | GPU: https://hwbot.org/submission/5351246_

[electrosoft]

3 hours ago, Mr. Fox said:

Yes it is. Not sure why. I think there is something wrong with the benchmark to be honest. I think it is doing something weird with cache or memory. It's not as stressful as Cinebench R23. My CPU doesn't get as hot as R23, but it still takes lots of extra voltage for some reason. It's strange. I haven't tried lowering the cache ratio or decreasing the memory overclock to see if the behavior changes, but something unusual is going on. It seems to waste a bit of time "preparing" for the benchmark before it starts. Not sure what that's about. Sometimes it won't start rendering. Sometimes it stops responding without crashing or freezing. Other times it runs normally. I think it's just buggy.

 

On top of that, I find it really annoying to have to remember to manually change it from a timed stress test to benchmark using the "off" option. I just tried to open it and got a BSOD a minute ago. Yet, I can run any other version of Cinebench, OCCT or AIDA64 stress test without issue, so yeah, it's buggy.

 

Does it exhibit the same behaviors at complete stock? It might be buggy (wouldn't be the first time) or it might be rooting out mild little instabilities in overclocks somewhere.

[tps3443]

3 hours ago, electrosoft said:

 

Does it exhibit the same behaviors at complete stock? It might be buggy (wouldn't be the first time) or it might be rooting out mild little instabilities in overclocks somewhere.

I can run it at 6.1P/4.6E. But then again, my water is 15C at least.