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

[Talon]

50 minutes ago, Papusan said:

 

Here is where the 5090 chips go. Almost all of them for workstation cards for the China market. I can't understand why this is better than feed China with vanilla 5090's. The US government banned 5090 gaming cads for the China market, but this is so much better. Oh'well, none can really stop Jensen from doing business with his preferred friends. This is disgusting. 

 

NVIDIA's RTX PRO 5500 is China's most wanted GPU, and almost nobody is getting one soon

 

So why is China buying a workstation card instead of a data center accelerator? The RTX PRO 5500 is a Blackwell-based GPU with 84GB of GDDR7 memory that complies with US export rules, which bar NVIDIA's B300 from China and put its H200 under case-by-case licensing, making it one of the most powerful NVIDIA chips Chinese companies can still legally buy.

 

This is just really sad. The 5090 couldn’t be exported but a 84gb card can? Makes no sense. This card is easily better for AI given the vram and same core count. The slightly cutdown bus likely won’t hinder it much at all. 

[Talon]

Have any of you had weird artifact/purple like boxes flicker in your displays before? I can’t figure out what is causing this issue on my new GPU. Tried DDU. Tried swapping cables. It always seems to come back. Sometimes a restart takes care of it for awhile but it reappears randomly. Even happens in games which shocked me as I figured maybe it was an idle memory issue. Sorta stumped on it. 
 

I’ll upload a video later if I can 

[Mr. Fox]

54 minutes ago, Talon said:

Have any of you had weird artifact/purple like boxes flicker in your displays before? I can’t figure out what is causing this issue on my new GPU. Tried DDU. Tried swapping cables. It always seems to come back. Sometimes a restart takes care of it for awhile but it reappears randomly. Even happens in games which shocked me as I figured maybe it was an idle memory issue. Sorta stumped on it. 
 

I’ll upload a video later if I can 

I have seen weird glitches like that with some of the 621.XX drivers and I have seen it when undervolting GPUs. I am assuming it is with the GPU running stock with no tweaks like undervolting or overclocking. If it is bone stock, have you tried an earlier driver branch, like 610.XX or 596.XX to see if the behavior is the same? If it is the same across several driver braches, test with a different monitor and different HDMI/DP cable. (I have an ASUS 4K 120 HZ IPS display that has all sorts of weird issues with various GPUs to the point that I only use it as an office TV now.) If it still persists after all of those things then something has to be wrong with the GPU.

[Talon]

3 hours ago, Mr. Fox said:

I have seen weird glitches like that with some of the 621.XX drivers and I have seen it when undervolting GPUs. I am assuming it is with the GPU running stock with no tweaks like undervolting or overclocking. If it is bone stock, have you tried an earlier driver branch, like 610.XX or 596.XX to see if the behavior is the same? If it is the same across several driver braches, test with a different monitor and different HDMI/DP cable. (I have an ASUS 4K 120 HZ IPS display that has all sorts of weird issues with various GPUs to the point that I only use it as an office TV now.) If it still persists after all of those things then something has to be wrong with the GPU.

 

Went into Microcenter to add the warranty. Brought the card with me and talked with the tech and manager about the artifacts. Had the video on my phone showing it happening on the desktop and in games. They said not normal and wanted to swap the card. They only had the demo unit they had on display behind glass left but offered that card. They took the card, put it into it's retail box and swapped the GPU and added the 2 year warranty. 

 

Just got the card in and it's significantly better and no artifacts yet. 57.6% ASIC, not a lottery winner but not a dud at all. My previous card would crash at 3300Mhz in GPUZ. This card does 3400Mhz and uses 1085Mv max. I'm just happy to have a card that isn't artifacting out of box. 

 

Got my adapter delivered and unfortunately, my Apex board blocks the cable. The heatsink gets in the way. Ugh. So now I'll have to use a small offset or go vertical or something. Not my favorite but it seems no choice. 

[tps3443]

On 11/7/2026 at 11:02 PM, Mr. Fox said:

Nice to see you, brother. Hope all is well with the family.

 

With mVolt+ there is no longer any need to cross-flash BIOS, shunt-mod or use EVC2 for voltage control. You can change all of those things with software now... just a portable exe file does it all.

 

 

And, now you can also run the latest DLSS and frame gen on GPUs NVIDIA said "do not support" DLSS and MFG. Yup, on something like a 3060 that would barely manage 30 FPS at 4K now you can do 100+ FPS, thanks to talented modders finding ways to beat and defeat the Green Goblin's lies and shenanigans. DLSS Enabler or DLSS5 Manager even let's you run DLSS on some AMD and Intel GPUs now.

 

 

I’m doing great! I hope you and your family are all good as well! Im still out of work, sleeping until 11am some days and hustling junk cars, hard to believe it’s been about 6 months now. Life is good!
 

I started playing with Mvolt+ tonight, although I already had it, I haven’t really dug in to it. I just got the new one though, amazing app. My 5090 Astral has a pretty low Xbar like 2,490Mhz, so I focused my tuning there to try and gain some performance. I kept the card all stock for now 600w pwr, 0% voltage slider, auto fans, default memory +0 and default core clocks +0. I started adding voltage to my Xbar and saw performance degradation and Xbar even went down even lower lol. So then I started taking away Xbar voltage, I reached -20mv and the score improved slowly in Steel Nomad DX12, then I began adding Xbar clock frequency and reached +164. So I settled on a 15,500 Steel Nomad score on my 5090 Astral which is amazing. All stock power limit of 100%, and 0% voltage slider, no memory or core overclocking, or undervolting at all. So just with Xbar clocks going up. 
 

Then I undervolted the card to 900mv, OCed the memory +3,000, and it hit its 16,001 and even idling at 10 watts on the desktop. 
 

Not even really pushing the card at all. Great daily setup for air card. 
 

 

 

[Talon]

16 hours ago, Talon said:

Got my adapter delivered and unfortunately, my Apex board blocks the cable. The heatsink gets in the way. Ugh. So now I'll have to use a small offset or go vertical or something. Not my favorite but it seems no choice. 

 

Well this was wrong. I foolishly thought the riser card was needed to connect to the BTF port but completely not needed. 🤣 It's for using to connect to a BTF board. The BTF to 12v2x6 cable works flawlessly. Near perfect 12v under 800w load with great per pin distribution. 

[electrosoft]

18 hours ago, Talon said:

 

Went into Microcenter to add the warranty. Brought the card with me and talked with the tech and manager about the artifacts. Had the video on my phone showing it happening on the desktop and in games. They said not normal and wanted to swap the card. They only had the demo unit they had on display behind glass left but offered that card. They took the card, put it into it's retail box and swapped the GPU and added the 2 year warranty. 

 

Just got the card in and it's significantly better and no artifacts yet. 57.6% ASIC, not a lottery winner but not a dud at all. My previous card would crash at 3300Mhz in GPUZ. This card does 3400Mhz and uses 1085Mv max. I'm just happy to have a card that isn't artifacting out of box. 

 

Got my adapter delivered and unfortunately, my Apex board blocks the cable. The heatsink gets in the way. Ugh. So now I'll have to use a small offset or go vertical or something. Not my favorite but it seems no choice. 

 

2 hours ago, Talon said:

 

Well this was wrong. I foolishly thought the riser card was needed to connect to the BTF port but completely not needed. 🤣 It's for using to connect to a BTF board. The BTF to 12v2x6 cable works flawlessly. Near perfect 12v under 800w load with great per pin distribution. 

 

Shout out again to Microcenter.....

 

When I saw you picked up the 20th edition bundle I did giggle a bit and thought to myself, "My Day One source strikes again!" 🤣

 

Our local MC still has a single Asus 20th edition package in stock. I tried to talk a buddy into picking it up to upgrade his aging 2080ti + 5800X3D  rig and his exact words were, "are you daft? YOU pick it up!" 🤣

 

Glad the replacement is a much better sample and in the end it looks slick AF. It is easily the best looking 5090 on the market IMHO.

 

 

 

 

 

 

[Meaker]

On 11/8/2026 at 2:19 PM, Talon said:

This is just really sad. The 5090 couldn’t be exported but a 84gb card can? Makes no sense. This card is easily better for AI given the vram and same core count. The slightly cutdown bus likely won’t hinder it much at all. 

This way Nvidia gets to keep all the profit so it's fine, just ask Jensen's girlfriend 😄

[Talon]

1 hour ago, Meaker said:

This way Nvidia gets to keep all the profit so it's fine, just ask Jensen's girlfriend 😄

 

She has expensive taste. It all makes sense now. 

[electrosoft]

7 hours ago, Meaker said:

This way Nvidia gets to keep all the profit so it's fine, just ask Jensen's girlfriend 😄

 

5 hours ago, Talon said:

 

She has expensive taste. It all makes sense now. 

 

Reddit FTW....

 

 

 

[Papusan]

Nothing will change as long  third-party integrators prefer own sales channels/grey market where profits is much higher... 

NVIDIA releases Statement on GeForce RTX 5090: GB202 Allocations / 32GB GDDR7 Active Supply

 

Industry analysis indicates that the observed retail tightness stems from high enterprise demand for dense compute hardware. Reports from regional conversion centers indicate that retail GeForce RTX 5090 desktop graphics cards are being acquired in volume by third-party integrators and reconfigured into dual-slot blower assemblies for deployment in AI compute servers. This secondary demand pulls consumer retail stock away from traditional PC hardware channels, compounding perceived supply shortages.

[Talon]

2 hours ago, Papusan said:

Nothing will change as long  third-party integrators prefer own sales channels/grey market where profits is much higher... 

NVIDIA releases Statement on GeForce RTX 5090: GB202 Allocations / 32GB GDDR7 Active Supply

 

Industry analysis indicates that the observed retail tightness stems from high enterprise demand for dense compute hardware. Reports from regional conversion centers indicate that retail GeForce RTX 5090 desktop graphics cards are being acquired in volume by third-party integrators and reconfigured into dual-slot blower assemblies for deployment in AI compute servers. This secondary demand pulls consumer retail stock away from traditional PC hardware channels, compounding perceived supply shortages.

 

I mean, there are multiple RTX 5090 GPUs available shipped and sold by Newegg right now. But the price is over $6000. But shockingly, they've remained in stock for over a day. If stock builds back up, we might actually see prices finally start to drop as well. 

 

https://www.newegg.com/asus-rog-astral-rog-astral-rtx5090-o32g-gaming-geforce-rtx-5090-32gb-graphics-card-triple-fans/p/N82E16814126751

https://www.newegg.com/gigabyte-gv-n5090aorus-m-32gd-geforce-rtx-5090-32gb-graphics-card-triple-fans/p/N82E16814932760

https://www.newegg.com/gigabyte-gv-n5090aorusm-ice-32gd-geforce-rtx-5090-32gb-graphics-card-triple-fans/p/N82E16814932765

https://www.newegg.com/gigabyte-gv-n5090aorusst-ice-32gd-geforce-rtx-5090-32gb-graphics-card-triple-fans/p/N82E16814932810

https://www.newegg.com/gigabyte-gv-n5090gaming-oc-32gd-geforce-rtx-5090-32gb-graphics-card-triple-fans/p/N82E16814932761

 

Don't everyone buy at once! Stand back people! LOL. At least the Gigabyte cards come with a motherboard? 

[Papusan]

6 hours ago, Talon said:

 

I mean, there are multiple RTX 5090 GPUs available shipped and sold by Newegg right now. But the price is over $6000. But shockingly, they've remained in stock for over a day. If stock builds back up, we might actually see prices finally start to drop as well. 

 

https://www.newegg.com/asus-rog-astral-rog-astral-rtx5090-o32g-gaming-geforce-rtx-5090-32gb-graphics-card-triple-fans/p/N82E16814126751

https://www.newegg.com/gigabyte-gv-n5090aorus-m-32gd-geforce-rtx-5090-32gb-graphics-card-triple-fans/p/N82E16814932760

https://www.newegg.com/gigabyte-gv-n5090aorusm-ice-32gd-geforce-rtx-5090-32gb-graphics-card-triple-fans/p/N82E16814932765

https://www.newegg.com/gigabyte-gv-n5090aorusst-ice-32gd-geforce-rtx-5090-32gb-graphics-card-triple-fans/p/N82E16814932810

https://www.newegg.com/gigabyte-gv-n5090gaming-oc-32gd-geforce-rtx-5090-32gb-graphics-card-triple-fans/p/N82E16814932761

 

Don't everyone buy at once! Stand back people! LOL. At least the Gigabyte cards come with a motherboard? 

 

Looking good. The US have catched up and finally offer cheaper cards ta\han here home. Plenty of them in stock. So run and buy, LOOL

 

Here is more disgusting... The first snow dumped down here home today and I already hate it😒 

 

 

[Talon]

35 minutes ago, Papusan said:

 

Looking good. The US have catched up and finally offer cheaper cards ta\han here home. Plenty of them in stock. So run and buy, LOOL

 

Here is more disgusting... The first snow dumped down here home today and I already hate it😒 

 

 

Man I would embrace that so hard. Go outside with the rig and do some OC fun. Covered of course. Then light a fire inside and chill sipping some coffee. 

[Papusan]

22 hours ago, Talon said:

Man I would embrace that so hard. Go outside with the rig and do some OC fun. Covered of course. Then light a fire inside and chill sipping some coffee. 

 

The winter will last 6-7 months forwards. With slushy snow, cold, rain, hail, wind, storm and normal snow.  Man I would love to be an other place with an chiller for benching in the evening and nice warming sun in the days🙂 Etc like my favorite place... Canary Islands. We already ordered a 2 weeks trip for the end of November, months ago. But sadly me and my wife ain't in the best shape right now so we had to cancel it and we pushed the trip to Mars. We ordered 3 weeks to cut down on the long winter. So very needed.

 

 

The big question is.... Do they have the balls?

Nvidia’s Board reportedly preparing to oust CEO Jensen Huang as second extra-marital affair emerges

 

Chinese university with military ties bought two NVIDIA H200 AI chips worth $123,000 despite US export blacklist

 

A Chinese university with close ties to the People's Liberation Army (PLA) bought NVIDIA H200 chips last month despite being on a US export blacklist, according to procurement records Reuters reviewed.

 

[tps3443]

I was looking at PS5 and PS5 Pro specs as we near closer to GTA6 launch! Yeah lol. Console is locked in at 30fps with no performance modes available for 60fps even being an option. Apparently this limitation is due to the games CPU intensive scenarios. And since the PS5 Pro is running something similar to a Ryzen 3700x desktop 8/16 CPU but it’s clocked about 400-500Mhz slower, and only has 1/4 the L3 cache size of a standard Ryzen 3700X. This is why the 30fps cap is required on even the PS5 Pro. 
 

Anyways, I just found that interesting how weak-ish the PS5 Pro’s CPU really is. 

[Mr. Fox]

28 minutes ago, tps3443 said:

I was looking at PS5 and PS5 Pro specs as we near closer to GTA6 launch! Yeah lol. Console is locked in at 30fps with no performance modes available for 60fps even being an option. Apparently this limitation is due to the games CPU intensive scenarios. And since the PS5 Pro is running something similar to a Ryzen 3700x desktop 8/16 CPU but it’s clocked about 400-500Mhz slower, and only has 1/4 the L3 cache size of a standard Ryzen 3700X. This is why the 30fps cap is required on even the PS5 Pro. 
 

Anyways, I just found that interesting how weak-ish the PS5 Pro’s CPU really is. 

They are about like one of the crappy little Minisforum NUC-sized (a.k.a turdbook in a box) "gaming" PCs, maybe a little less powerful but close. The PS5 is just more expensive, more limited in function, less repairable and more proprietary.

---

So, these G.SKILL 8400 UDIMM kits aren't that great. I have two kits of them (one is a spare part for emergency backup) and their rated speed is close to their max capability. 8600 is still pretty  impressive for 24/7 stability, but I got instantly spoiled by CUDIMMs now that I have platforms supporting CUDIMM. It doesn't really make any difference in the grand scheme of things though. Measurable in memory benchmarks, but not much else once you go beyond 8000-8200 it quickly starts to be irrelevant for anything other than bragging rights.

[tps3443]

2 minutes ago, Mr. Fox said:

They are about like one of the crappy little Minisforum NUC-sized "gaming" PCs, just more expensive and proprietary.

---

So, these G.SKILL 8400 UDIMM kits aren't that great. I have two kits of them (one is a spare part for emergency backup) and their rated speed is close to their max capability. 8600 is still pretty  impressive for 24/7 stability, but I got instantly spoiled by CUDIMMs now that I have platforms supporting CUDIMM. It doesn't really make any difference in the grand scheme of things though. Measurable in memory benchmarks, but not much else once you go beyond 8000-8200 it quickly starts to be irrelevant for anything other than bragging rights.

Is memory overclocking on Z890 similar to Z790? Also, did you sell your 5090? Are you running a 5080 and a 5090? 

[Mr. Fox]

13 minutes ago, tps3443 said:

Is memory overclocking on Z890 similar to Z790? Also, did you sell your 5090? Are you running a 5080 and a 5090? 

It is not remarkably different. Much easier than Ryzen and not nearly as persnickety. Z890 is mostly different on the CPU-side of overclocking. There is more crap like NGU and D2D that comes as baggage with the new CPU architecture.

 

No, I am not planning on selling the 5090. It is in one of my Z890 Apex builds. I generally use that one for benching with the chiller and it stays turned off most of the time. I just dropped a new 270K Plus into the Z890 Apex with the water-cooled 5080. The other 5080 is the air-cooled Prime GPU on the open bench with the Z890 Unify-X.

 

I need to delid this new CPU, but it will have to wait until my thumbs are healed up. I had surgery on both thumbs... left one last month and right one on Monday, and I do not have the necessary dexterity at  the moment. My right thumb is still bandaged up with stitches and the left one is still limited range of motion.

[Mr. Fox]

On 10/26/2026 at 11:04 PM, Mr. Fox said:

So, I got another 8800 CUDIMM kit (Corsair Vengeance) to use in the Unify-X with the CPU I was given by Brother @Talon. I started studying Intel DDR5 memory tuning and learned something new using AI (with links to the points of reference authored by humans). tRTP and tWR should be left on "Auto" because the timings get jacked up. Instead, set tRDPRE instead of tRTP and tWRPRE instead of tWR. (This may apply to AMD as well, but I have not tested it.) 

 

What to set for tWR using tWPRE instead:

To get a target tWR of 48, you need to work backward using your tCWL value:

• Formula: tWRPRE = 48 + tCWL + 8

Example 1: If your tCWL is 38, you would set tWRPRE to 94 (48 + 38 + 8).

Example 2: If your tCWL is 40, you would set tWRPRE to 96 (48 + 40 + 8).
(Leave tWR set to Auto for this to work.)

@tps3443 - try  the above to see if it works for Ryzen the same. Leave tWR on Auto and use the formula above to control tWR using tWRPRE. With tWR set to Auto you should see the actual value displayed (48 in the above example) in ZenTimings.

---

Tweaked a couple more settings to drop latency a bit more.

I think I might be looking for another 8800 or 9000 CUDIMM kit in a few months. (I am seeing some indications that DDR5 prices might start to go down soon. Hope so.) For the interim, I think I am going to strip the chrome heatsinks off my new Corsair Vengeance 8800 CUDIMMs and put them in the second Apex on water and put these 8400 G.SKILL UDIMMs in the Unify-X (also on water) since it is not quite as good at memory overclocking as the Apex boards are.

[Mr. Fox]

I discovered something else that improves CPU performance. There is an option found on the CPU Power Management Control page that needs to be disabled for maximum CPU performance on Windoze and especially so for Linux. On older BIOS (like I use) it is named "Resource Priority Feature" (sounds like something good, but the "priority" is wrong) and on newer BIOS it will be listed as "HWP Autonomous EPP Grouping" (or something similar to that). It is in the same screen where you disable C-states, SpeedShit and Intel Dynamic Platform Framework crap. This setting allows Windoze to control CPU behavior to conserve energy at the expense of maximizing overclocked performance. Linux is even more aggressive at doing that and the power management crap requires more effort to manipulate in Linux. (Maybe because European countries are more Nazi-like in their quest to reduce power consumption and Linux is far more common in Europe than the US. Just a guess on my part.)

 

Anyhoo... disabling this setting increases my Cinebench scores by about 600 points and also see higher CPU-Z single and mult-core benchmark numbers. Using the same CPU overclock settings, the Cinebench score is consistently in the 48.2-48.3K range enable versus 48.8-48.9K disabled.

With the setting enabled, CPU-Z multi-thread is in the high 20K range (always under 21K) and single thread in the 950 range. Disabled, it jumps up to these values.

[Papusan]

 

Not so sure this will help gamers/consumers. Jensen made graphics cards for games and gamers, what will the new leaders do? They can probably nothing about Geforce and gaming. They will probably put all the eggs into AI and severs.

Jensen Huang leaves Nvidia with immediate effect

The founder is leaving after 33 years as CEO. The CFO is taking over and the product plan will be finalized.

 

There is a lot more into this. All the smuggling of chips and helping build up China's AI capacity (in the grey market) is probably the main reason. 

 

And China... They does what they always do. First fill up (stock up) with chips and vram, then make it work the way they want it....

Modded Nvidia RTX 5090 with 96 GB VRAM and custom BIOS surfaces in China

 

[Mr. Fox]

24 minutes ago, Papusan said:

 

Not so sure this will help gamers/consumers. Jensen made graphics cards for games, what will the new leaders do? They can probably nothing about Geforce and gaming. They will probably put all the eggs into AI and severs.

Jensen Huang leaves Nvidia with immediate effect

The founder is leaving after 33 years as CEO. The CFO is taking over and the product plan will be finalized.

 

There is a lot more into this. All the smuggling of chips and helping build up China's AI capacity (in the grey market) is probably the main reason. 

We are living in dark times worldwide and things stand to get much worse in the near future. Wickedness is a way of life now. Setting aside all of the extremely undesirable political troubles and inevitable fallout looming on the horizon and focusing on the technology space, a huge part of it is a concerted effort to monetize everything and convert personal computing into a subscription model in every way possible. The big players don't want PC enthusiasts or general computer users to own long-lasting, powerful and capable hardware or locally installed software that allows a buy-once/use forever scenario to occur. We can say we "don't blame them from a business perspective" but the intent is self-serving and nefarious at its root, designed to extract every ounce of revenue possible and make being dependent versus self-sufficient the end goal.

 

This touches all aspects of life in the world we live in. Essentially the same business model as big pharma. Nobody wants cures or permanent solutions. They want lifelong prescriptions that feed the pig for the same reasons that big tech wants lifelong subscriptions. The end user is no longer a valued customer, we are a source of revenue that needs to be captured in perpetuity. No thanks. Hard pass. I do blame them and that business model. It's messed up and unacceptable.