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

[electrosoft]

18 hours ago, Mr. Fox said:

The normie sheeple and ignoramus population has become so vast and prevalent the OEMs no longer give a rat's butt about enthusiast money because they don't need to. We no longer matter. There are no customers, only victims. Clueless morons pay asinine prices for things they know nothing about, and they do it with a smile. They believe all the ads they read, all the lies their "friends" on Facepoot and Reddtit tell them, and all of the disinformation they watch on the "news" so is it any wonder that clowns like A$$zeus behave as if they have a license for sodomy? We have both to blame: the stupid sheeple and the shifty thugs peddling grossly overpriced, misrepresented and overhyped crap.

 

But we also know when rallied they can make a difference. Look at the de-prioritization of 8GB 5060/9060 cards due to lack of sales and more focus on the 16gb models.

 

Look at prices falling (for now) for sub 5090 cards.

 

Look how the market shifted away from Intel to AMD.

 

Many of those decisions (and more) were consciously made by consumers even normies.

 

If they could stop just shrugging their shoulders and locking onto RGB and aesthetically pleasing things and focus on the hardware and capability/functionality but I don't think we're ever going to be there because most end users don't care about any of that. They just want to plug in and get going and if problems arise they track down peeps that are like us in this thread.

 

The A$$zeus license comment made me lol btw..... 🤣

 

-----------------------

 

@jaybee83 confirmed working on the Suprim Air by two users who have flashed it and it's working as intended.

 

 

 

So far, it appears higher voltage card users that had higher ranges but the 600w limit kept them lower are experiencing increases in voltage to go along with their newfound 200w of power. On the other hand, if your card is mid or lower and already bouncing off that vmax as is at 600w, you won't suddenly open up more voltage but that 200w will still give you a nice, healthy kick.

 

List so far of cards not working with it:

 

Asus Astral 5090 Black

Asus Astral 5090 LC Black

Asus TUF

Suprim Liquid

5090 FE (duh)

 

Ironically, the revamped design of the Astral white 5090 does work with it.....so clearly not done out of maliciousness but a byproduct of the fan design.

 

Haven't seen the Zotac 5090 AIO tested along with a few other cards yet but the vast majority are able to run it.

 

Someone in Asus won't walk right after they are separated from their keister for letting this slip out as is and not differentiating power detection properly.

 

 

[Mr. Fox]

6 hours ago, electrosoft said:

Haven't seen the Zotac 5090 AIO tested along with a few other cards yet but the vast majority are able to run it.

Have you identified anyone that owns a Zotac 5090 AIO version? I haven't gone looking but I don't recall seeing anyone in the forum at oc.net that has one. In general, I do not want an AIO GPU, (or CPU AIO,) but definitely think it is better than air. I do not care for the appearance of most of the AIO cooled GPUs. I think most of them are pretty garish-looking and I do not like the short and stubby look. At least the water blocked GPUs are not nearly as short and stubby as the OEM AIO models.

[electrosoft]

1 hour ago, Mr. Fox said:

Have you identified anyone that owns a Zotac 5090 AIO version? I haven't gone looking but I don't recall seeing anyone in the forum at oc.net that has one. In general, I do not want an AIO GPU, (or CPU AIO,) but definitely think it is better than air. I do not care for the appearance of most of the AIO cooled GPUs. I think most of them are pretty garish-looking and I do not like the short and stubby look. At least the water blocked GPUs are not nearly as short and stubby as the OEM AIO models.

 

Actually, it is one of the few cards I have never seen owned by anyone out in the wild. It may be quite some time before we know if it works with the 800w vbios or not.....

 

 

[Papusan]

17 hours ago, Papusan said:

 

Not sure I even will buy next gen graphics cards. Probably more locked down and we have Nvidia surfing on an AI bubble. Everything is messed up in the tech world. More broken trash and all too expensive for my likings.

 

Post #24,578

https://www.overclock.net/posts/29530054/

 

 

Btw. Nvidia have done a great job for repair shops

 

Fixed thousands of melted connectors. NorthridgeFix should send Nvidia an huge thanks, LOL

 

On 11/19/2025 at 10:14 PM, Papusan said:

One thing for sure... You will see less Asus cards in 6090 series thread on OC.net.

 

Edit. What I expect from Asus now... A new firmware for Matrix that lock you to 600W with use of only one power connector. They missed on this. I expect it will be fixed (sarcasm).

 

 

[Papusan]

 

On 11/17/2025 at 2:11 PM, Mr. Fox said:

Jury is still out whether I like 3DvCache. The 9950X3D that I RMA'd was such a pathetic silicon sample that it wasn't fair to draw any conclusions based on what a piece of garbage it was. I'm leaning toward either not liking it or viewing it as an irrelevant selling point/gimmick. (I'm a casual gamer, not a gaming enthusiast, and it seems to provide no tangible benefit for an overclocking enthusiasts.)

 

Not only Asus can make dog-shit. What Asus and AMD have common... The first letter (A).

 

 

[Reciever]

A little preview of what I'm building next 

 

Sage 621e

[Mr. Fox]

16 hours ago, Reciever said:

A little preview of what I'm building next 

 

Sage 621e

Awesome. I am looking forward to seeing how it runs, brother.

---

I got up early to delid the CPU. It always amazes me how much better than makes things. I increased the core clock 100MHz across all 16 cores and my max temperature is still over 20°C cooler. I hate owning CPUs that are not delidded. The experience sucks with the IHS.

 

So, I gave up on trying to use Windoze worthless trash software bloat to determine whether or not 3DvCache actually worked or not. Micro$lop butchers or botches up everything they get their hands on.

 

I used Process Lasso to set parameters on the EXE file to force 3DvCache CCD0 use or all-core use with both CCDs.

 

As you can see below, it DOES make a difference.

 

All Core (no 3DvCache)

CCD0 only (3DvCache only) - I used Process Lasso for this instead of Windoze trash

 

 

All Core (no 3DvCache)

 

CCD0 only (3DvCache only) - I used Process Lasso for this instead of Windoze trash

[Reciever]

2 hours ago, Mr. Fox said:

Awesome. I am looking forward to seeing how it runs, brother.

---

I got up early to delid the CPU. It always amazes me how much better than makes things. I increased the core clock 100MHz across all 16 cores and my max temperature is still over 20°C cooler. I hate owning CPUs that are not delidded. The experience sucks with the IHS.

 

So, I gave up on trying to use Windoze worthless trash software bloat to determine whether or not 3DvCache actually worked or not. Micro$lop butchers or botches up everything they get their hands on.

 

I used Process Lasso to set parameters on the EXE file to force 3DvCache CCD0 use or all-core use with both CCDs.

 

As you can see below, it DOES make a difference.

 

All Core (no 3DvCache)

 

CCD0 only (3DvCache only) - I used Process Lasso for this instead of Windoze trash

 

All Core (no 3DvCache)

 

CCD0 only (3DvCache only) - I used Process Lasso for this instead of Windoze trash

 

Amd chips are cheaper but the motherboard options are slim pickings. 

 

Finally got fed up with oem motherboards on the Dell precisions so I'm selling off a few of them and moving to a DIY solution. The board came with a couple xeon silvers which I s fine for testing but long term I'm probably going with a couple xeon gold 6248's 20c/40t 

 

Ordered a be Quiet 1000w for it as well I'll begin testing it over the weekend. 

[electrosoft]

4 hours ago, Mr. Fox said:

Awesome. I am looking forward to seeing how it runs, brother.

---

I got up early to delid the CPU. It always amazes me how much better than makes things. I increased the core clock 100MHz across all 16 cores and my max temperature is still over 20°C cooler. I hate owning CPUs that are not delidded. The experience sucks with the IHS.

 

So, I gave up on trying to use Windoze worthless trash software bloat to determine whether or not 3DvCache actually worked or not. Micro$lop butchers or botches up everything they get their hands on.

 

I used Process Lasso to set parameters on the EXE file to force 3DvCache CCD0 use or all-core use with both CCDs.

 

As you can see below, it DOES make a difference.

 

All Core (no 3DvCache)

 

CCD0 only (3DvCache only) - I used Process Lasso for this instead of Windoze trash

 

All Core (no 3DvCache)

 

CCD0 only (3DvCache only) - I used Process Lasso for this instead of Windoze trash

 

Did you post the same screen shot for CP2077 twice? I'm not seeing the difference.

 

Second you can definitely see the uplift when in play.

 

When X3D is properly utilized, it hits hard. But when it doesn't especially if you didn't tune your system, whew....brutal.

 

----------------------------------------------------------------

 

So Jufus got scammed from BB. He picked up an open box Gigabyte 5090 and it was missing the GPU and RAM. One plus on Microcenter is they test all 5090 returns real time before issuing a return.

 

I've had a few issues with BB over the years with either the wrong product in the box (7950x when it was supposed to be a 7950x3d) or no product at all (when i bought a sony camera and the box was empty) to back in the day WoW gamecards already being used.

 

I still love them though. When they're firing on all cylinders it's good stuff.

 

Timestamped:

 

 

Pic:

 

[Talon]

2 hours ago, electrosoft said:

 

Did you post the same screen shot for CP2077 twice? I'm not seeing the difference.

 

Second you can definitely see the uplift when in play.

 

When X3D is properly utilized, it hits hard. But when it doesn't especially if you didn't tune your system, whew....brutal.

 

----------------------------------------------------------------

 

So Jufus got scammed from BB. He picked up an open box Gigabyte 5090 and it was missing the GPU and RAM. One plus on Microcenter is they test all 5090 returns real time before issuing a return.

 

I've had a few issues with BB over the years with either the wrong product in the box (7950x when it was supposed to be a 7950x3d) or no product at all (when i bought a sony camera and the box was empty) to back in the day WoW gamecards already being used.

 

I still love them though. When they're firing on all cylinders it's good stuff.

 

Timestamped:

 

 

Pic:

 

 

Damn, scammers now running scams through BB. That is just unreal, I hate people lol. 

[Mr. Fox]

4 hours ago, electrosoft said:

 

Did you post the same screen shot for CP2077 twice? I'm not seeing the difference.

Ooops, I sure did. I updated the post with this image. Pretty huge difference in high and low FPS, and those runs are 4K with no frame generation.

 

CCD0 only (3DvCache only) - I used Process Lasso for this instead of Windoze trash

[Papusan]

On 8/28/2025 at 7:56 AM, Mr. Fox said:

I think we should call BTF trash WTF instead.

 

Spot on. From 55:28

 

[electrosoft]

17 hours ago, Papusan said:

 

Spot on. From 55:28

 

 

Biggest problem to me is the lack of a standard for the BTF connectors. Hyte goes where the market goes as do all case makers. If BTF is properly standardized and gains traction, they will go there but it has been swirling around for years now going nowhere so.....

 

And that makes the Matrix and Astral BTF editions problematic because it suddenly limits your choice of motherboards if you want to switch/upgrade. I don't care for that.

 

-----------------------------

20 hours ago, Mr. Fox said:

Ooops, I sure did. I updated the post with this image. Pretty huge difference in high and low FPS, and those runs are 4K with no frame generation.

 

CCD0 only (3DvCache only) - I used Process Lasso for this instead of Windoze trash

 

Ahhh, much better. Here is a good video from BBT where he pits maxed tuned (memory and CPU)  14900KS vs 285k vs 9800X3D vs 9950X3D in a proper shootout.

 

The overall consensus is still the 9800X3D is  still king. Shines at 1080p. Loses a little steam at 1440p and really loses steam at 4k to pretty much make all tuned chips the same and the 285k actually comes out on top as the GPU experiences being capped. 

 

Like I said, 285k gets a bad rap, but properly tuned and paired with what is gobsmackingly expensive ram now, it can hold its own and then some. Also the newest edition of Intel's APO is actually valid now and works as intended when it can take advantage of its optimizations.

 

For X3D, it comes down to the games being played/tested and the resolution too and how that effects the cache and saturation.

 

Even Jufes, who beats on AMD regularly said if all he played was racing sims or flight sims hell yeah his main rig would be X3D as it dominates in those types of games.

 

I still feel to this day the 14900KS when tuned properly provides a smoother, more steady gameplay experience in WoW for handling the lows but I also play at 4k maxed out so that subjects the X3D cache to more saturation than 1440p and especially 1080p. I actually logged into my WoW account on the wife's system the other day and played for a few hours at 1440p then set internal render to 4k for the next few hours and the gameplay, even on her 9070xt, is just smoother especially in raids.

 

The 4k evidence from BBT's video just makes me want to build out a 285k even more at this point for 4k testing. I said it before, but if a decent 285k sample pops up on the forums, I'll probably pick it up and build out a cheap rig and just use these 8400 Kingspec sticks and see where they can go on it.

 

Overall, when tuned ALL the top end chips provide a good gaming experience.

 

 

[Papusan]

 

People still don't know why you would make two colors tiny trash. For the design ? 

 

I've plugged it in as hard as possible (with care) but there's a little bit of yellow left, is this ok or do I have to reconnect it?

 

 

[electrosoft]

3 hours ago, Papusan said:

 

People still don't know why you would make two colors tiny trash. For the design ? 

 

I've plugged it in as hard as possible (with care) but there's a little bit of yellow left, is this ok or do I have to reconnect it?

 

 

 

As someone who has owned and tested 3x 5000 series MSI cards (5080 Shadow, 5090 Ventus, 5090 Vanguard) yes, there should be no yellow visible lol.......

 

 

[Mr. Fox]

1 hour ago, electrosoft said:

 

As someone who has owned and tested 3x 5000 series MSI cards (5080 Shadow, 5090 Ventus, 5090 Vanguard) yes, there should be no yellow visible lol.......

There should be no yellow visible even when it is still in the PSU box.

[electrosoft]

So the itch to try Arrowlake is slowly winning but frugally, so I snagged an open box Asus Z890  Strix-A from da 'zon. When all is said and done, $176 delivered to my door.  It is the cheapest route I could sort out that still gives me access to SP ratings on CPUs as I test. Arrowlake is all new to me so that will be fun as I've twisted AM5 and Intel Skylake++ every which way possible via several motherboards and chips over the last 2-3 years.

 

The plan of attack is to install this into the empty white Hyte Y40 that used to house my secondary/daughter's system and rebuild out of it since she is now ready to come over and game again after the other system gathered dust for well over a year before I broke it down and sold the components so I'll build out again......but she's a good daughter who is working and in school so I don't mind at all. 

 

I have a few spare PSUs floating around including an EVGA P2 850w and MSI Ai1300p so I'm good there along with a few Samsung 980 and 990 Pro 1TB and 2TB SSDs. I have at last count 4-5 AIOs of various flavors on my shelf from 240mm up to 360mm so I'm good there too. As previously noted, I have those 8400 Kingspec 2x24GB sticks floating about so those will be used. I wouldn't splurge for 9000 cudimms even before the memory crazy pricing so I'm sure as heck not dropping $420 for another set of 8400 or a mind boggling $675+ for a set of 9000. I'll just see where these or the TG 8200 sticks go. Just fishing for a CPU to use.

 

Might test a few 265k/kf's or 285k's depending on Black Friday specials. 265k is obviously the cheapest way to go.

 

 

[Papusan]

 

Hmmm. A$us prefer the cheap way. The more you pay the more they save. Nvidia offer their 5090 FE cards at MSRP ($2000) and still come with rubber die guard. I expect Asus prefered to use paste as die guard to save sub 1$ in cost of build. Isn't that cheapo? Or is it the classic greed?

 

[electrosoft]

1 hour ago, Papusan said:

 

Hmmm. A$us prefer the cheap way. The more you pay the more they save. Nvidia offer their 5090 FE cards at MSRP ($2000) and still come with rubber die guard. I expect Asus prefered to use paste as die guard to save sub 1$ in cost of build. Isn't that cheapo? Or is it the classic greed?

 

 

Just watched it a bit ago.....

 

Like I said before, these were available for pre-order from the USA Asus site for well over 6hrs. Sold out eventually, but clearly not in that normal demand window here in the states. As a check, I was at the "checkout" window every hour on the hour and could have checked out and picked one up very easily.

 

Whoever designed it has a thing for phallic designs obviously....

 

I did like the fact you could detach the BTF connector and run it at 600w if you so desire if you can't or won't purchase an Asus exclusive BTF motherboard. Of course the downside is you might not be able to use the 800w but considering non BTF users are pulling 800w I'm curious to see if it will via 12v2x6 only. I am also curious to verify if it can do 600w via the BTF connector like the Astral BTF.

 

Card design is basically the white Astral BTF with a few enhancements but de8auer is going to do some more testing about power delivery.

 

That liquid metal application was atrocious and the barrier with paste only was a slap in the face. I expect a lot more on a $4k GPU.

 

The cooling and especially fan improvements show Asus does know when they bork things up. It is just getting them to fix it....

 

Users with Asus 5090s that can't use the BIOS (IE, all of them except the white astral) need not hold their breath. There is zero chance Asus will release a revised bios that will allow their cards to suddenly become warranty possibilities and also piss off Jensen. I also would not be surprised if Asus patches their BIOS before shipping the next batches of cards to lock it down even further but at this point the cat is out of the bag. Outside of the balanced power delivery via both connectors and admittedly beefy cooler, there is no reason to buy one of these except if you're a fanboy which de8auer readily admitted he's a sucker for this type of stuff.

 

Best thing is the 800w vBIOS. Not only lifting PL to 800w but also better overall clocks and stability. It is the new Gigachad vBIOS by far....unless you have an Asus card (sans white 5090) or Suprim 5090 AIO so far.... As soon as my Wireview Pro II comes in, I'm flashing this to my Vanguard 5090. Since it is a full fat 1.125v model that even when pulling ~630w in some scenarios can't break past 1.10v, I am fully looking forward to opening up as much of that 1.125v as possible in tandem to 800w......Vanguard already has very good cooling and even at 630w I'm sitting at mid to low 60's so bring on the high 60's low 70's! 🤣

 

Having picked up the Vanguard 5090 for $1971 was the icing on the cake. Returning the Astral 5090 on the last day of my 60 day window and getting a full ~$3600 refund back in early June? Priceless especially now. I would have been absolutely sick to my stomach at this point knowing how much I paid and not being able to flash this 800w vBIOS. I would be selling it off at a loss with just as much disgust as I sold off my 5080FE when I found out nothing could flash to it.

 

 

 

 

 

 

 

 

 

 

[electrosoft]

13 hours ago, electrosoft said:

So the itch to try Arrowlake is slowly winning but frugally, so I snagged an open box Asus Z890  Strix-A from da 'zon. When all is said and done, $176 delivered to my door.  It is the cheapest route I could sort out that still gives me access to SP ratings on CPUs as I test. Arrowlake is all new to me so that will be fun as I've twisted AM5 and Intel Skylake++ every which way possible via several motherboards and chips over the last 2-3 years.

 

The plan of attack is to install this into the empty white Hyte Y40 that used to house my secondary/daughter's system and rebuild out of it since she is now ready to come over and game again after the other system gathered dust for well over a year before I broke it down and sold the components so I'll build out again......but she's a good daughter who is working and in school so I don't mind at all. 

 

I have a few spare PSUs floating around including an EVGA P2 850w and MSI Ai1300p so I'm good there along with a few Samsung 980 and 990 Pro 1TB and 2TB SSDs. I have at last count 4-5 AIOs of various flavors on my shelf from 240mm up to 360mm so I'm good there too. As previously noted, I have those 8400 Kingspec 2x24GB sticks floating about so those will be used. I wouldn't splurge for 9000 cudimms even before the memory crazy pricing so I'm sure as heck not dropping $420 for another set of 8400 or a mind boggling $675+ for a set of 9000. I'll just see where these or the TG 8200 sticks go. Just fishing for a CPU to use.

 

Might test a few 265k/kf's or 285k's depending on Black Friday specials. 265k is obviously the cheapest way to go.

 

 

 

Newegg is kicking off black friday pricing:

 

285k for $499 and includes $60 BF6 code AND a $125 MSI 360 cooler.....

 

265k for $259 and includes $60 BF6 code AND a $110 MSI 240 cooler....

 

--------------

 

Come Friday, I'm specifically on the hunt for the cheapest 9070xt, cheapest 285k/265k and maybe if the Corsair 1000w SF100 drops even lower. I think that's about it this year.....

 

 

[Papusan]

On 11/23/2025 at 10:10 PM, electrosoft said:

I also would not be surprised if Asus patches their BIOS before shipping the next batches of cards to lock it down even further but at this point the cat is out of the bag.

Outside of the balanced power delivery via both connectors and admittedly beefy cooler, there is no reason to buy one of these except if you're a fanboy which de8auer readily admitted he's a sucker for this type of stuff.

 

Balanced power delivery between the two connectors? Can't do that. That would increase the costs of the PCB. 

The ROG MATRIX PLATINUM RTX 5090 800W VBIOS has been revealed!

After resetting the settings, the total power target of 800W was still slightly off ( link ). HPWR2 should be the 16-pin connector on the graphics card, while HPWR1 is the 16-pin connector on the ROG X870E HERO BTF motherboard. The two account for approximately 70% and 33% respectively, which translates to 67% and 33% for a total power of 762W. This shocked many overclocking enthusiasts, as everyone was expecting a 50% 50% performance.

 

 

And what you think will happen once China again can get whatever they want from Nvidia and AMD? Component shortages will increase further and push the prices on pc hardware to new height we have never seen. And China they wiill be delighted... They have more than enough of very cheap electricity to power whatever AI data centers/plants they fire up. And they should really send Trump an huge thank you, sir. And he will be so moved to tears that we will probably see some golden wet tears under the crown, LOL

 

China's huge big fleet of tech engineers will make wonders with the better western tech to boost their military technologies. Greed, flattery and stupidity come at a huge price. Some is so stupid that they can't see it themself. Sad they don't know who'm they themself are. Maybe look into the "golden"mirror above the golden sink. I'm sure you will see stupidity. Fooled by Jensen and Xi Jinping.

The United States wants to give NVIDIA back the ability to sell its GPUs to China.

 

Quote

Hardwate prices kreep slowly upwards....

 

Only $7,999. Hurry up and grab it before the prices kreep up again😀 AMD has already issued an order saying all Radeon cards are way too cheap. But I'm not sure nvidia will follow our good friend F. Azors order right away. 

 

NVIDIA RTX PRO 6000 Blackwell Desktop GPU Drops to $7,999

 

 

Or maybe an Dell Alienware is more tempting? You even get an Ryzen 9 9950X3D in the +$6100 package.

Dell Alienware Area-51 PCs now available with Ryzen ...

 

Alienware Area-51 Gaming Desktop, Ryzen 9 9950X3D, 64GB RAM, 4TB SSD, GeForce RTX 5090. $6149.00.

 

[Mr. Fox]

One thing that I find difficult with overclocking Ryzen versus Intel is how fussy and sensitive they are in comparison. There are lots of settings that do not play nice together and if one thing is just slightly off it is like a house of cards. Performance and stability can go from great to garbage if all of the stars are not aligned. The CPU overclocking on Ryzen is about as fickle and finicky as RAM tuning is on Intel, and RAM tuning is a tedious process on either platform. In a way it is fun, but can be the opposite of fun (a really huge pain in the butt) when finding the right lever to pull is elusive. When you factor in the wildcards of the silicon lottery (trashy CPU and memory quality samples are common) it becomes even harder to be pleased with results. The crazy memory overclocking results posted places like overclock.net are fringe examples of the small winner count in the silicon lottery and they are not representative of the norm. 

 

Over the past week or so I have moved around CPUs and memory between the Strix and Apex. The Strix CONSISTENTLY has higher Cinebench scores compared to the Apex regardless of the CPU sample (9950X SP118, 9950X SP119 and 4585PX SP120). The memory overclock limits are exactly the same on all three CPUs no matter which motherboard they are installed in.

 

Nothing I have tested in terms of settings allows the Apex to beat the Strix X870E-E on Cinebench. EVER. This was equally true comparing Cinebench scores again the X870E AORUS Master. The Apex loses 100% of the time in Cinebench. Why? I wish I knew. Is the the idiotic PCIe lane allocation on the Apex? The 2-DIMM memory topology ASUS used on the Apex? Is it the DIMM.2 somehow dragging down the CPU? Don't have a clue. Whatever it is, Cinebench doesn't like it.

 

As far as memory overclocking in concerned, either the Apex is hardly any better or both of my memory samples have reached their functionality threshold. Same threshold with four different CPUs. The Apex maxes out at 8200 stable with my G.SKILL 6000 C26 1.450V 32GB kit and the Strix maxes out at 8000 stable with that kit. The Apex maxes out at 8400 with sloppy timings that gives performance the same as 8000 with tight timings using the Kingbank 8400 C40 1.450V 48GB kit (which has both XMP and EXPO 8400 profiles). With that kit the Strix maxes out a 8000 with the same tight timings that work on the Apex. The Strix cannot boot 8400 stable but the end result is the same memory performance on both systems. The latency and speeds follow the CPU and RAM kit to achieve identical results in each motherboard. This essentially makes the Strix the winner overall. The fact that it is roughly $300 cheaper, depending on where and when you buy it, makes it an absolute no brainer. The Apex is a great motherboard, but it doesn't deliver based on price or the fact that Cinebench scores are CONSISTENTLY 500-1000 points lower for any given CPU installed in the Apex.

 

I captured this comparison of the small latency penalty of 3D V-Cache. I did not bother capturing screenshots of different motherboard, RAM and CPU combos because it always ends the same. The screenshots would not show anything different overall. The latency results follow the X3D and non-X3D CPU and memory combo regardless of what motherboard they are installed in.

The G.SKILL 8000 C40 kit that I picked up with the Supercool Direct Touch RAM block is an XMP kit. It is virtually impossible to tune it stable at 8000 with all manual tuning in either of the AM5 motherboards. If I use timings that are stable with either of my other two memory kits mentioned above it refuses to boot and I have to clear CMOS. I cannot figure out why that is the case and it does not make any sense. Maybe the kit is rubbish, but I don't have an Intel motheboard to test it in to find out. However, I have experienced the exact same problem trying to use XMP memory that I had that were stable on Z790 motherboards that were also unusable on X670E and X870E motherboards, so I think it is something peculiar with the XMP memory, even though it makes no sense that manual tuning doesn't work.

 

I have been trying to get the memory latency on the 4585PX below 60ns and it is tough to do. This CPU is nothing more than a better binned 9950X3D and I think this one might be slightly better than average. While the memory and L1, L2 and L3 read/write/copy and latency values are a direct result of clock speed, I wondered if the 3D V-Cache was making the latency higher. It seems it is, but not by enough to be statistically relevant. I asked Google Gemini about it, and this is what the AI regurgitated for me. The "chef" analogy (below) is pretty good. I thought @Raiderman and @jaybee83 might find this particularly interesting.

 

---

Per Google Gemini - for whatever that is worth...

Question: Do AMD X3D CPUs have higher memory and L3 cache latency than the non-X3D processors?

---

Yes, technically they do.

 

It is a paradox of the architecture: AMD X3D CPUs generally demonstrate higher latency for both L3 cache (in cycles) and system memory (in nanoseconds) compared to their non-X3D counterparts.

 

However, they are faster in gaming because their massive cache capacity prevents the CPU from needing to access system memory in the first place.

 

Here is the technical breakdown of why this happens and why it doesn’t hurt gaming performance.

1. L3 Cache Latency: The "Stacking Tax"

The primary difference between an X3D chip (like the 7800X3D) and a non-X3D chip (like the 7700X) is the 3D V-Cache technology. AMD physically stacks a slice of SRAM on top of the processor die. This introduces a slight delay.

• The Penalty: Accessing the 3D-stacked cache typically adds a ~4-cycle penalty.
• The Clock Speed Factor: Latency is a product of cycles and clock speed ($Time = Cycles / Frequency$). Since X3D chips run at lower clock frequencies to manage heat, the absolute latency in nanoseconds is slightly higher than non-X3D chips that boost 400-500MHz higher.

Feature	Non-X3D (e.g., 7700X)	X3D (e.g., 7800X3D)
L3 Cache Cycles	Lower (Faster)	Higher (~4 cycle penalty)
Core Frequency	Higher (~5.4+ GHz)	Lower (~5.0 GHz)
Result	Lowest raw L3 latency	Slightly higher L3 latency

 

2. Memory Latency: The Benchmark Illusion

If you run a synthetic benchmark like AIDA64, an X3D CPU will almost always show higher memory latency (worse performance) than a non-X3D chip.

• Reason 1: Clock Speed: Memory latency is sensitive to core frequency. Because X3D chips have lower peak clocks, the time it takes to request data from RAM and process the return is technically longer.
• Reason 2: Testing Methodology: Synthetic benchmarks deliberately force the CPU to fetch data from System RAM (DDR5) to test the worst-case scenario. They bypass the cache to measure the "wire delay" to the memory sticks.

In this specific "worst-case" scenario, the non-X3D chip wins because it has faster clock speeds and no cache-stacking overhead.

3. The "Effective Latency" Paradox

If the latency is higher, why are X3D chips faster for gaming?

The answer lies in Hit Rate.

• Non-X3D: Fast access to L3, but the cache is small (32MB). The CPU frequently "misses" (can't find data) and must go to System RAM.
• Penalty: Trip to RAM takes ~60–70ns (eternity in CPU time).
• X3D: Slightly slower access to L3, but the cache is huge (96MB).7 The CPU almost never misses.
• Benefit: It stays within the L3 cache, which takes ~10–12ns.

The Math of "Effective" Latency:

Even if the X3D cache is 10% slower to access, avoiding one trip to system RAM (which is 600% slower) saves massive amounts of time.

 

Analogy: Imagine you are a chef.

• Non-X3D: You have a small fridge under your counter. You can grab ingredients instantly, but you frequently have to run to the grocery store (RAM) for missing items.
• X3D: You have a giant walk-in pantry. It takes you 2 seconds longer to walk into the pantry than the small fridge, but you almost never have to leave the building to go to the grocery store. You cook faster overall.

Summary

• Raw Latency: Higher on X3D. (Due to lower clocks + stacking penalty).
• Effective Latency: Lower on X3D. (Because you rarely hit the slow System RAM).

 

If your workload fits entirely inside the cache (like most games), X3D wins. If your workload is random and too large for even the X3D cache (like some scientific rendering), the non-X3D chip might win due to higher clock speeds.

 

The video below is relevant because it directly compares the older 7800X3D and 7700X, illustrating how the theoretical latency differences translate into actual gaming performance.

 

Ryzen 7 7700X vs Ryzen 7 7800X3D - Any Difference Or Not?

 

---

TL;DR - See below:

Lesson #1: If you want to go with AM5 and care about Cinebench scores buy an X870E-E Strix or X870E AORUS Master and save yourself $200-$300. If you a lottery winner have an extremely rare Ryzen golden silicon CPU sample and an extraordinarily superior memory kit and only care about synthetic memory benchmarks and nothing else matters, cough up the money for an Apex then have fun watching paint dry overnight with Karhu speed tests that have no measurable bearing on normal, real-world performance results in anything other that Karhu.

Lesson #2: Unless you want to waste a lot of time and drive yourself insane, don't try to get an Intel XMP memory kit to work on AM5. Unless you get lucky somehow, you are doing to hate yourself for not just buying an EXPO memory kit.

[jaybee83]

3 hours ago, Mr. Fox said:

One thing that I find difficult with overclocking Ryzen versus Intel is how fussy and sensitive they are in comparison. There are lots of settings that do not play nice together and if one thing is just slightly off it is like a house of cards. Performance and stability can go from great to garbage if all of the stars are not aligned. The CPU overclocking on Ryzen is about as fickle and finicky as RAM tuning is on Intel, and RAM tuning is a tedious process on either platform. In a way it is fun, but can be the opposite of fun (a really huge pain in the butt) when finding the right lever to pull is elusive. When you factor in the wildcards of the silicon lottery (trashy CPU and memory quality samples are common) it becomes even harder to be pleased with results. The crazy memory overclocking results posted places like overclock.net are fringe examples of the small winner count in the silicon lottery and they are not representative of the norm. 

 

Over the past week or so I have moved around CPUs and memory between the Strix and Apex. The Strix CONSISTENTLY has higher Cinebench scores compared to the Apex regardless of the CPU sample (9950X SP118, 9950X SP119 and 4585PX SP120). The memory overclock limits are exactly the same on all three CPUs no matter which motherboard they are installed in.

 

Nothing I have tested in terms of settings allows the Apex to beat the Strix X870E-E on Cinebench. EVER. This was equally true comparing Cinebench scores again the X870E AORUS Master. The Apex loses 100% of the time in Cinebench. Why? I wish I knew. Is the the idiotic PCIe lane allocation on the Apex? The 2-DIMM memory topology ASUS used on the Apex? Is it the DIMM.2 somehow dragging down the CPU? Don't have a clue. Whatever it is, Cinebench doesn't like it.

 

As far as memory overclocking in concerned, either the Apex is hardly any better or both of my memory samples have reached their functionality threshold. Same threshold with four different CPUs. The Apex maxes out at 8200 stable with my G.SKILL 6000 C26 1.450V 32GB kit and the Strix maxes out at 8000 stable with that kit. The Apex maxes out at 8400 with sloppy timings that gives performance the same as 8000 with tight timings using the Kingbank 8400 C40 1.450V 48GB kit (which has both XMP and EXPO 8400 profiles). With that kit the Strix maxes out a 8000 with the same tight timings that work on the Apex. The Strix cannot boot 8400 stable but the end result is the same memory performance on both systems. The latency and speeds follow the CPU and RAM kit to achieve identical results in each motherboard. This essentially makes the Strix the winner overall. The fact that it is roughly $300 cheaper, depending on where and when you buy it, makes it an absolute no brainer. The Apex is a great motherboard, but it doesn't deliver based on price or the fact that Cinebench scores are CONSISTENTLY 500-1000 points lower for any given CPU installed in the Apex.

 

I captured this comparison of the small latency penalty of 3D V-Cache. I did not bother capturing screenshots of different motherboard, RAM and CPU combos because it always ends the same. The screenshots would not show anything different overall. The latency results follow the X3D and non-X3D CPU and memory combo regardless of what motherboard they are installed in.

The G.SKILL 8000 C40 kit that I picked up with the Supercool Direct Touch RAM block is an XMP kit. It is virtually impossible to tune it stable at 8000 with all manual tuning in either of the AM5 motherboards. If I use timings that are stable with either of my other two memory kits mentioned above it refuses to boot and I have to clear CMOS. I cannot figure out why that is the case and it does not make any sense. Maybe the kit is rubbish, but I don't have an Intel motheboard to test it in to find out. However, I have experienced the exact same problem trying to use XMP memory that I had that were stable on Z790 motherboards that were also unusable on X670E and X870E motherboards, so I think it is something peculiar with the XMP memory, even though it makes no sense that manual tuning doesn't work.

 

I have been trying to get the memory latency on the 4585PX below 60ns and it is tough to do. This CPU is nothing more than a better binned 9950X3D and I think this one might be slightly better than average. While the memory and L1, L2 and L3 read/write/copy and latency values are a direct result of clock speed, I wondered if the 3D V-Cache was making the latency higher. It seems it is, but not by enough to be statistically relevant. I asked Google Gemini about it, and this is what the AI regurgitated for me. The "chef" analogy (below) is pretty good. I thought @Raiderman and @jaybee83 might find this particularly interesting.

 

---

Per Google Gemini - for whatever that is worth...

Question: Do AMD X3D CPUs have higher memory and L3 cache latency than the non-X3D processors?

---

Yes, technically they do.

 

It is a paradox of the architecture: AMD X3D CPUs generally demonstrate higher latency for both L3 cache (in cycles) and system memory (in nanoseconds) compared to their non-X3D counterparts.

 

However, they are faster in gaming because their massive cache capacity prevents the CPU from needing to access system memory in the first place.

 

Here is the technical breakdown of why this happens and why it doesn’t hurt gaming performance.

1. L3 Cache Latency: The "Stacking Tax"

The primary difference between an X3D chip (like the 7800X3D) and a non-X3D chip (like the 7700X) is the 3D V-Cache technology. AMD physically stacks a slice of SRAM on top of the processor die. This introduces a slight delay.

• The Penalty: Accessing the 3D-stacked cache typically adds a ~4-cycle penalty.
• The Clock Speed Factor: Latency is a product of cycles and clock speed ($Time = Cycles / Frequency$). Since X3D chips run at lower clock frequencies to manage heat, the absolute latency in nanoseconds is slightly higher than non-X3D chips that boost 400-500MHz higher.

Feature	Non-X3D (e.g., 7700X)	X3D (e.g., 7800X3D)
L3 Cache Cycles	Lower (Faster)	Higher (~4 cycle penalty)
Core Frequency	Higher (~5.4+ GHz)	Lower (~5.0 GHz)
Result	Lowest raw L3 latency	Slightly higher L3 latency

 

2. Memory Latency: The Benchmark Illusion

If you run a synthetic benchmark like AIDA64, an X3D CPU will almost always show higher memory latency (worse performance) than a non-X3D chip.

• Reason 1: Clock Speed: Memory latency is sensitive to core frequency. Because X3D chips have lower peak clocks, the time it takes to request data from RAM and process the return is technically longer.
• Reason 2: Testing Methodology: Synthetic benchmarks deliberately force the CPU to fetch data from System RAM (DDR5) to test the worst-case scenario. They bypass the cache to measure the "wire delay" to the memory sticks.

In this specific "worst-case" scenario, the non-X3D chip wins because it has faster clock speeds and no cache-stacking overhead.

3. The "Effective Latency" Paradox

If the latency is higher, why are X3D chips faster for gaming?

The answer lies in Hit Rate.

• Non-X3D: Fast access to L3, but the cache is small (32MB). The CPU frequently "misses" (can't find data) and must go to System RAM.
• Penalty: Trip to RAM takes ~60–70ns (eternity in CPU time).
• X3D: Slightly slower access to L3, but the cache is huge (96MB).7 The CPU almost never misses.
• Benefit: It stays within the L3 cache, which takes ~10–12ns.

The Math of "Effective" Latency:

Even if the X3D cache is 10% slower to access, avoiding one trip to system RAM (which is 600% slower) saves massive amounts of time.

 

Analogy: Imagine you are a chef.

• Non-X3D: You have a small fridge under your counter. You can grab ingredients instantly, but you frequently have to run to the grocery store (RAM) for missing items.
• X3D: You have a giant walk-in pantry. It takes you 2 seconds longer to walk into the pantry than the small fridge, but you almost never have to leave the building to go to the grocery store. You cook faster overall.

Summary

• Raw Latency: Higher on X3D. (Due to lower clocks + stacking penalty).
• Effective Latency: Lower on X3D. (Because you rarely hit the slow System RAM).

 

If your workload fits entirely inside the cache (like most games), X3D wins. If your workload is random and too large for even the X3D cache (like some scientific rendering), the non-X3D chip might win due to higher clock speeds.

 

The video below is relevant because it directly compares the older 7800X3D and 7700X, illustrating how the theoretical latency differences translate into actual gaming performance.

 

Ryzen 7 7700X vs Ryzen 7 7800X3D - Any Difference Or Not?

 

---

TL;DR - See below:

Lesson #1: If you want to go with AM5 and care about Cinebench scores buy an X870E-E Strix or X870E AORUS Master and save yourself $200-$300. If you a lottery winner have an extremely rare Ryzen golden silicon CPU sample and an extraordinarily superior memory kit and only care about synthetic memory benchmarks and nothing else matters, cough up the money for an Apex then have fun watching paint dry overnight with Karhu speed tests that have no measurable bearing on normal, real-world performance results in anything other that Karhu.

Lesson #2: Unless you want to waste a lot of time and drive yourself insane, don't try to get an Intel XMP memory kit to work on AM5. Unless you get lucky somehow, you are doing to hate yourself for not just buying an EXPO memory kit.

 

the timing of your post is hilarious, ive been going nuts these past weeks since i cant get my kit stable at 8000 anymore, had to back down to 7600 ugh. the same exact setup hardware wise, only two differences being the new case / cpu cooling / fans and ofc i updated to the newest bios. imagine that i currently cannot even get 6000 stable anymore in synchronous mode, like wtf.... and even at 7600 with tightened timings the bandwidth and latency results are absolute garbage. theres definitely something funky going on here, and as of yet i have no idea what it is....

[Mr. Fox]

9 hours ago, jaybee83 said:

 

the timing of your post is hilarious, ive been going nuts these past weeks since i cant get my kit stable at 8000 anymore, had to back down to 7600 ugh. the same exact setup hardware wise, only two differences being the new case / cpu cooling / fans and ofc i updated to the newest bios. imagine that i currently cannot even get 6000 stable anymore in synchronous mode, like wtf.... and even at 7600 with tightened timings the bandwidth and latency results are absolute garbage. theres definitely something funky going on here, and as of yet i have no idea what it is....

I suspect that it is something to do with the firmware/AGESA version. Try going back to an older bios version that was allowing you to run stable with your 8000 memory settings. If you already flashed one of those firmware that tells you you cannot go back once flashed you probably still can if you do the BIOS flashback from USB. That almost always works for me. Just remember to rename the file correctly before putting it on a USB thumb drive all by itself.

 

You could even try it with the same BIOS version first because it could be something that got corrupted in the firmware and the best way of fixing that is doing the flashback feature manually instead of trying to reflash the normal way.

[Mr. Fox]

https://www.overclock.net/posts/29531606/

safedisk said: 

hello Can you provide me with the cinebench comparison results between apex and x870e-e?

Sure. Thank you. Of course when you try to plan something it never turns out exactly as you thought. The gap is normally wider than this, but the Strix still wins in Cinebench. I have all BIOS settings the same. The Windows installation is optimized exactly the same. I have set a static overclock with 55x on favored CCD and 54 on the other CCD. Both are running DDR5 at 8000 MT/s, although that does not influence Cinebench too much. Where I see most frequently the very wide variance (500-1000 Cinebench points) is when I am trying to beat my previous highest score and using chilled water and max stable overclock or when I randomly run Cinebench as a random stability check. Even with chilled water at 8-10°C the Apex cannot surpass the Strix. It is very frustrating (as you can tell by the fact I mentioned it often). I settled on the Apex for benching, forgiving its poorly appointed PCIe slot placements and it doesn't bench as well. The Strix has a more usable motherboard layout in terms of the PCIe bifurcation and slot placement. The only thing I like better about the Apex is 2-DIMM slots and dual BIOS. I like everything else more with the Strix.

The first screenshot today is per your request with the static overclock mentioned above. The best of three runs on each system this morning to respond to your request.
The second screenshot are my highest scores with each system posted on HWBOT.

 

Chilled water highest achieved scores submitted to HWBOT. This is after literally hundreds of Cinebench runs with both systems, so the variance I normally see is minimized due to the cherry-picked highest score examples of the best CBR23 runs ever produced by both systems.

 

(As a side note, the 5090 was moved from the Apex to the Strix since it is the better platform for benching. Not what I hoped for with a $700 motherboard, but it is what it is.)