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

[Reciever]

3 hours ago, Mr. Fox said:

If I had to wager I would bet that it's true because NVIDIA is who they are... control freaks. If they can find a way to screw the people that buy their products that is exactly what they will do. That's how they roll and who they are, and that is how they have been for a long time. Also one of the numerous reasons I dislike FE video cards.

---

And, on a happier note, I ordered my second Lian Li O11D XL EVO to house the second AORUS Master. Also ordered a D5 distro block from Radikult Customs and about 12 feet of Alphacool EDPM tubing.

 

Now I hear rumors of a 9080 XT with 32GB of GDDR7. I am so glad I didn't blow $3000+ on a 5090 already. Now it is no longer even tempting. Not caring actually feels really good.

The source on that is very unreliable from what little I understand on the matter. Would be pretty cool if true, but AMD many times now already said they arent targeting high-end, more so marketshare which is the logical course of action given how little they have.

 

It better have 2 12VHPWR cables if it exists, 600w is not a lot of headroom for a card that operates @ 450w. I was pulling probably close to 775w-815w on my 7900 XTX. I need to get one of those Kill-a-Watt's to be more sure.

[Mr. Fox]

12 minutes ago, Reciever said:

The source on that is very unreliable from what little I understand on the matter. Would be pretty cool if true, but AMD many times now already said they arent targeting high-end, more so marketshare which is the logical course of action given how little they have.

 

It better have 2 12VHPWR cables if it exists, 600w is not a lot of headroom for a card that operates @ 450w. I was pulling probably close to 775w-815w on my 7900 XTX. I need to get one of those Kill-a-Watt's to be more sure.

I was watching a Moore's Law is Dead video and apparently it is based on some leaked benchmarks.

 

As long as the temps are not 100°C I would gladly buy a 4090-level AMD card for the price of a 5080 or less. So, I hope it is true, but we shall see. AMD really doesn't cater to performance enthusiasts. Their niche is more geared toward mainstream on everything.

 

 

[Reciever]

2 minutes ago, Mr. Fox said:

I was watching a Moore's Law is Dead video and apparently it is based on some leaked benchmarks.

 

As long as the temps are not 100°C I would gladly buy a 4090-level AMD card for the price of a 5080 or less. So, I hope it is true, but we shall see. AMD really doesn't cater to performance enthusiasts. Their niche is more geared toward mainstream on everything.

Would definitely be of interest to me as well under those circumstances. Not sure if there would be a real need for 32GB VRAM though.

 

Playing Warhammer 40k Space Marine 2 with some 4k texture mods, appears to be sitting around 13GB VRAM @ 3440x1440 resolution. I wouldnt say I always play the latest titles so perhaps my opinion is skewed on that front. Of course everything tends to boil down to price.

[Mr. Fox]

Everyone that purchased parts on overclock.net has been thrilled with their purchases. Especially the guy that bought the Gene parts combo. He made out like a bandit, LOL. Nice to see good parts getting put to use instead of collecting dust on my spare parts shelf. I still have the Strix combo with the excellent bin 13900KS and G.SKILL 7200 kit and the SP104 14900KF for sale. If the Strix combo doesn't sell by the time I do the second AORUS Master build I will probably sell it locally on Facepoot Marketplace as a working system minus GPU or with the old Titan Black GPU. That would probably be better anyhow because I would have to figure out how to get rid of the Antec C8.

 

While I am waiting for Central Computer to send a replacement for the AORUS Master that was damaged from the NVMe heatsink slamming around inside of the unopened motherboard box, the XL EVO and distro block to arrive I decided to go ahead and integrate the RAM water cooling on the existing beast. I used the EK manifold for this as I usually do, but mounted it to a new place. This will also make it easy to connect the chiller and bypass the radiators. The QDC fitting for cold water leaving the chiller is on the dual D5 pump housing (same place as before) and the "hot" water going back to the chiller will use the QDC fitting on the EK manifold. It still purging the air from the loop, but working well otherwise.

 

So, the internal loop routes like this:

1. D5 pump/res > RAM block in
2. RAM block out > EK manifold
3. EK Manifold > CPU in
4. CPU out > EK manifold
5. EK manifold > Top radiator
6. Top radiator > Rear X-Flow radiator
7. Rear X-Flow radiator > Bottom radiator
8. Bottom radiator > IceMan Dual D5 block
9. Dual D5 block > GPU in
10. GPU out > D5 pump/res

Using the chiller the routing will be:

1. Chiller out > IceMan Dual D5 block
2. Dual D5 block > GPU in
3. GPU out > D5 pump/res
4. D5 pump/res > RAM block in
5. RAM block out > EK manifold
6. EK Manifold > CPU in
7. CPU out > EK manifold
8. EK manifold > Chiller in

 

[Papusan]

14 hours ago, Mr. Fox said:

If I had to wager I would bet that it's true because NVIDIA is who they are... control freaks. If they can find a way to screw the people that buy their products that is exactly what they will do. That's how they roll and who they are, and that is how they have been for a long time. Also one of the numerous reasons I dislike FE video cards.

---

And, on a happier note, I ordered my second Lian Li O11D XL EVO to house the second AORUS Master. Also ordered a D5 distro block from Radikult Customs and about 12 feet of Alphacool EDPM tubing.

 

Now I hear rumors of a 9080 XT with 32GB of GDDR7. I am so glad I didn't blow $3000+ on a 5090 already. Now it is no longer even tempting. Not caring actually feels really good.

 

5090 is just an expensive midrange high end gamer cards.  The more you pay the more you save. Here's Nvidia's new  "gamer "flagship" card😀

 

 

Not sure this would please @electrosoft😁 The $10.000 cards almost sounds like a real circular saw. Very special sound from it. Not the worst I have heard but the sound from this is awful and more annoying than this one from my older post

 

[electrosoft]

285k vs 9950X3D both tuned (ish)

 

Overall 9950X3D has higher fps, but the 285k has higher 1% lows sometimes substantially.

 

 

 

4 hours ago, Papusan said:

 

5090 is just an expensive midrange high end gamer cards.  The more you pay the more you save. Here's Nvidia's new  "gamer "flagship" card😀

 

 

Not sure this would please @electrosoft😁 The $10.000 cards almost sounds like a real circular saw. Very special sound from it. Not the worst I have heard but the sound from this is awful and more annoying than this one from my older post

 

 

Now THAT coil whine is identical to my 3090 FE. Just throw in a massive amount of variable pitch changes like the Aorus.....

 

But with that being said, outside of the price, this is what the 5090 should have been from launch with just 32GB for the stupid prices being charged for the 5090.

 

I LOVE the look and design of the Pro 6000 more than the 5090 FE plus you get the full fat GB202. Icing on the cake is the 96GB to run larger LLMs (well for me, I don't know about you). For AI work, it makes more sense for me to pick up a M4 macbook than a 5090. It makes more sense to pick up an M3 Ultra than the RTX Pro 6000.

 

"The 5090 is the waste. The garbage from the production of these (6000)" - brutal.

 

His end rant is spot on. You are getting the die rejects for the 5090 and yields for the full fat are much smaller. I'm sure Nvidia is stock piling the dies that can't even make it to mid range 5090 status and we'll get a 5080ti to dump those off.

 

The fact the 6000 Pro can look and run like a 5090 is all you need to know in regards to Nvidia just pushing and upselling everything to absurd levels.

 

Everything about the RTX 6000 Pro is the 5090 FE on steroids from the looks to the internals. 🤑

 

I still want a 6000 Pro though.....just not at $8k+

[Papusan]

1 hour ago, electrosoft said:

"The 5090 is the waste. The garbage from the production of these (6000)" - brutal.

 

His end rant is spot on. You are getting the die rejects for the 5090 and yields for the full fat are much smaller. I'm sure Nvidia is stock piling the dies that can't even make it to mid range 5090 status and we'll get a 5080ti to dump those off.

 

The fact the 6000 Pro can look and run like a 5090 is all you need to know in regards to Nvidia just pushing and upselling everything to absurd levels.

 

I have still not heard what nvidia will do with the Chinese market  due possible (5090D /5090) US ban. If they are forced to make castrated 5090DD (dual dumbed down 5090s) I expect 5080Ti will have a hard launch for gamers (complete deficiency). China will grab +20-30% of all sub castrated 5090 cards with +24GB vram. Aka a horrible 5080Ti launch for the global market.

 

 

 

[Clamibot]

1 hour ago, electrosoft said:

285k vs 9950X3D both tuned (ish)

 

Overall 9950X3D has higher fps, but the 285k has higher 1% lows sometimes substantially.

 

 

 

 

Now THAT coil whine is identical to my 3090 FE. Just throw in a massive amount of variable pitch changes like the Aorus.....

 

But with that being said, outside of the price, this is what the 5090 should have been from launch with just 32GB for the stupid prices being charged for the 5090.

 

I LOVE the look and design of the Pro 6000 more than the 5090 FE plus you get the full fat GB202. Icing on the cake is the 96GB to run larger LLMs (well for me, I don't know about you). For AI work, it makes more sense for me to pick up a M4 macbook than a 5090. It makes more sense to pick up an M3 Ultra than the RTX Pro 6000.

 

"The 5090 is the waste. The garbage from the production of these (6000)" - brutal.

 

His end rant is spot on. You are getting the die rejects for the 5090 and yields for the full fat are much smaller. I'm sure Nvidia is stock piling the dies that can't even make it to mid range 5090 status and we'll get a 5080ti to dump those off.

 

The fact the 6000 Pro can look and run like a 5090 is all you need to know in regards to Nvidia just pushing and upselling everything to absurd levels.

 

Everything about the RTX 6000 Pro is the 5090 FE on steroids from the looks to the internals. 🤑

 

I still want a 6000 Pro though.....just not at $8k+

 

Interesting video. I wonder what causes the drastic differences in 1% lows to flip flop depending on the game. I'm guessing it has something to do with inter-CCD latencies and the fact that not both CCDs have the 3D VCache, which in my opinion is a bad idea. Homogeneous designs yield the highest performance.

 

So looks like for Intel to reclaim the performance crown, they need to do 2 things:

1. Add their own version of 3D Vcache underneath their CPU cores (I believe they have something similar to that, which they call Adamantine cache)

 

2. Get rid of the stupid E-cores and make all cores P-cores. I don't get the point of the E-cores. All they do is reduce performance for the most part, so I leave them turned off since leaving the E-cores turned off significantly increases my performance in pretty much everything, with games getting the biggest boost. I only had one case when using my 14900KF where they increased performance, and performance didn't increase by that much. Tripling your core count to get only a 17% increase in performance is an insanely bad ROI in relation to the amount of extra cores (yes, I know the core types don't have the same processing power, but even assuming the E-cores are half the processing power of the P-cores, it's still a very bad performance increase in relation to the number of extra cores). Intel's server class Xeon CPUs have all P-cores and they perform great due to not being shackled by E-cores. No weird scheduling shenanigans.

 

Since Intel doesn't seem to want to ditch their E-core idea, we may have to move to their server class CPUs to escape the E-core insanity.

[tps3443]

19 hours ago, Mr. Fox said:

12VHPWR or whatever popular thing they are calling today, or whatever they decide to rename it tomorrow, is truly a POS connector that never should have seen the light of day. It is a sketchty, ill-conceived, fragile, unreliable and dangerous piece of crap. Nothing they can do to it--ever--will make it a good and reliable power cable. The group at NVIDIA responsible for bringing this abortion to market need to be stripped naked and caned in public until dead, then resuscitated, their employment terminated, citizenship revoked and then deported to a far away horrible place where life is miserable. 

Did you have one melt before?

 

18 hours ago, electrosoft said:

@tps3443

 

Have you heard the rumor that shunt modding the 5090 FE does nothing? I hope it isn't true.

 

https://www.overclock.net/posts/29472599/

 

Yes, I sawed it. I could see your FE sucks flag/banner raised up from my bedroom window. 🤣

 

18 hours ago, Mr. Fox said:

If I had to wager I would bet that it's true because NVIDIA is who they are... control freaks.

How much do you wager?  😎 

[electrosoft]

10 minutes ago, tps3443 said:

Yes, I sawed it. I could see your FE sucks flag/banner raised up from my bedroom window. 🤣

 

 

If only to counter your "I ♥️ 5090FE" tramp stamp...  🤣

 

But in all seriousness, I like the 5090FE. It still is in my top three list based on price. size and aesthetics.

 

If one pops on BB or VPA, I'd pick it up and give it the coil whine test. If it had bearable or no coil whine like my 5080FE (showing it is quite possible), I'd keep it. If not, off to eBay it would go.

 

It is an amazing SFF card. I would then seek out a SFF case small as possible that supports either a top horizontal inverted mount so the card would exhaust directly out the top of the case or a classic vertical mount that could invert to exhaust out the side. You gotta dump that heat somewhere if you don't have a big enough case.

 

5090FE + 9800X3D air cooled tuned SFF build would be the end game for desktop and portable power.

 

 

 

 

[Mr. Fox]

2 hours ago, Clamibot said:

 

Interesting video. I wonder what causes the drastic differences in 1% lows to flip flop depending on the game. I'm guessing it has something to do with inter-CCD latencies and the fact that not both CCDs have the 3D VCache, which in my opinion is a bad idea. Homogeneous designs yield the highest performance.

 

So looks like for Intel to reclaim the performance crown, they need to do 2 things:

1. Add their own version of 3D Vcache underneath their CPU cores (I believe they have something similar to that, which they call Adamantine cache)

 

2. Get rid of the stupid E-cores and make all cores P-cores. I don't get the point of the E-cores. All they do is reduce performance for the most part, so I leave them turned off since leaving the E-cores turned off significantly increases my performance in pretty much everything, with games getting the biggest boost. I only had one case when using my 14900KF where they increased performance, and performance didn't increase by that much. Tripling your core count to get only a 17% increase in performance is an insanely bad ROI in relation to the amount of extra cores (yes, I know the core types don't have the same processing power, but even assuming the E-cores are half the processing power of the P-cores, it's still a very bad performance increase in relation to the number of extra cores). Intel's server class Xeon CPUs have all P-cores and they perform great due to not being shackled by E-cores. No weird scheduling shenanigans.

 

Since Intel doesn't seem to want to ditch their E-core idea, we may have to move to their server class CPUs to escape the E-core insanity.

It seems like all of the hardware manufacturers are full of bad ideas. They are the proverbial "insane clown posse" by most measurement. Stupid is the new normal in their line of business. Maybe we should form our own posse, saddle up and go take them out. Right now we are surrounded and outnumbered by stupid, but we can use WMDs to level the playing field. Intel, AMD, NVIDIA, Apple, Microsoft, Google, Meta... the list of dishonest, crooked, stupid idiots that are only capable of stupidity is a long one. Even when they do something right it gets tainted by their abject stupidity and immorality.

[electrosoft]

So I returned my Astral 5090 today (tomorrow was the last day), and I have to admit it feels like a monster sized weight has been lifted off my chest. No matter how I tried to spin it, I could not accept paying ~$3600USD for a consumer grade GPU to game and bench no matter how perfect it was.

 

Pros: low to no coil whine, 1.110 ripe for XOC vBIOS, +3000, handled a +375 V/F curve offset no problem, clocked decently, aesthetically gorgeous, ran quiet, zero fans outside of meaningful load at all times.

 

Cons: expensive, heated up my entire office within an hour turning it into a sauna so I had to keep the ceiling fan going and hoping the AC would kick on. 

 

I collected a lot of WoW data and swapped between the 9070xt  routinely for various scenarios I would run including lots of LFR Raids, 5 mans and Delves along with ORBs and outdoor assaults and the 9070xt along with AMD made insane strides in driver stability and performance after launch on a level that has left me impressed but that's for another post.

 

 

---

 

Fun convo with the guy who rang up the return. He immediately said "such a nice card." and once we got to talking, he said he noticed on my account I had picked up a Gigabyte 9070xt OC Gaming. I said I picked up an open box the second time around. He noted he used the same card and picked it up open box too a few weeks back. Yeah, that was the one I returned! It didn't even make it to an open box listing. He scooped it up. He games at 1440p, so you know it is tearing it up.

 

4 hours ago, Clamibot said:

 

Interesting video. I wonder what causes the drastic differences in 1% lows to flip flop depending on the game. I'm guessing it has something to do with inter-CCD latencies and the fact that not both CCDs have the 3D VCache, which in my opinion is a bad idea. Homogeneous designs yield the highest performance.

 

So looks like for Intel to reclaim the performance crown, they need to do 2 things:

1. Add their own version of 3D Vcache underneath their CPU cores (I believe they have something similar to that, which they call Adamantine cache)

 

2. Get rid of the stupid E-cores and make all cores P-cores. I don't get the point of the E-cores. All they do is reduce performance for the most part, so I leave them turned off since leaving the E-cores turned off significantly increases my performance in pretty much everything, with games getting the biggest boost. I only had one case when using my 14900KF where they increased performance, and performance didn't increase by that much. Tripling your core count to get only a 17% increase in performance is an insanely bad ROI in relation to the amount of extra cores (yes, I know the core types don't have the same processing power, but even assuming the E-cores are half the processing power of the P-cores, it's still a very bad performance increase in relation to the number of extra cores). Intel's server class Xeon CPUs have all P-cores and they perform great due to not being shackled by E-cores. No weird scheduling shenanigans.

 

Since Intel doesn't seem to want to ditch their E-core idea, we may have to move to their server class CPUs to escape the E-core insanity.

 

 

 

In previous reviews of the 9800X3D vs the 9950X3D, gaming comparisons were basically a wash overall but I did not pay close attention to the 1% lows. I'll have to go back and take a look to test your theory. Before 9950X3D launch, the biggest wish was dual 3D-Vcache but alas.....

 

Intel and AMD are both guilty for abandoning a proper monolithic and/or homogeneous design. At least AMD went back to it for their 9000 series GPUs.

 

Raptor lake was too hot and a mess for more P-Cores (even though rumors of 10 and 12 core P-Core versions had hope springing eternal everywhere), but I thought they could have made an all  P-Core variant of Arrowlake. At least 12 P-cores would have been enough for me. For RPL it is a per game hit or miss fest. For Alder, it was almost universal to turn them off.

 

In Intel's defense, the E-cores on arrowlake are much better than RPLs but I still want a proper P-core only CPU. You might be right in the pursuit of Xeons.

 

 

[Mr. Fox]

1 hour ago, electrosoft said:

In Intel's defense, the E-cores on arrowlake are much better than RPLs but I still want a proper P-core only CPU. You might be right in the pursuit of Xeons.

^^^ This ^^^ AND, Hyperthreading. Super-retarded that they did away with that. In fact, this is the primary reason I went with an AM5 build. Had Core Ultra 285K not been missing hyperthreading I am 99% certain I would have bought A Z890 Apex and 285K instead of 9950X. They screwed themselves extra hard on that one, and customers have spoken loudly with their wallets. Overall, I still think Intel has a better platform than AMD and I am sad they've screwed it up by making stupid choices. (Why does it seem like I am saying this frequently about all brands now? Because they are all idiots. None of them deserve our support at this point, only our wrath.)

---

On 5/18/2025 at 1:07 PM, Mr. Fox said:

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

Item #3: Intel Core i9 14900KF - $375 + Shipping

 

14900KF (SP104_P116_E82_MC88) w/ lapped Copper IHS SP Rating Photos

(the MC SP is a screenshot saved from when it was in an Apex since Strix has no MC SP)

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

On 5/18/2025 at 1:07 PM, Mr. Fox said:

SP104 CPU - Lapped Copper IHS

Only item remaining is the 14900KF (SP104). Everything else has sold. @YoungChris at oc.net is interested but having to scrape up the cash. It's not being held for him. First-come, first-serve.

 

I've got a WireView and WireView Pro that I am not able to use any longer. I sold the Gigabyte 4090 that one was used on and the other one will not fit my 4090 Suprim with the new Alphacool Core waterblock. (It did fit with the Byksi block.) I will probably list both of them and the MSI 4090 Byski block. If anyone is interested before I list them drop me a PM to let me know.

[Papusan]

47 minutes ago, Mr. Fox said:

Overall, I still think Intel has a better platform than AMD and I am sad they've screwed it up by making stupid choices. (Why does it seem like I am saying this frequently about all brands now? Because they are all idiots. None of them deserve our support at this point, only our wrath.)

 

Hmmn. I wonder why I'm still on older Raptor lake refresh. Maybe the newer products from AMD and Intel doesn't offer much more than what I already have or want? Paying premium for the modern tile mess isn't exactly what I want. Nope, I'm fine with what I already have. New is just newer, isn't it?🙃

[Mr. Fox]

In spite of my high ambient temps, (30°C/86°F water) this BarrowCH RAM cooler does a nice job, especially considering it is not a direct touch block. After 30 minutes of TM5 the temps were the same at about 3 minutes of TM5 with the stock garbage heating blankets, and I couldn't run more than about 5 minutes until TM5 would start spitting out errors and the modules were close to 62°C. I just couldn't see spending over $100 for another IceMan direct-touch RAM block. The 4-DIMM version has to weigh a ton as the 2-DIMM version is insanely heavy.

 

The water temperature gauge on the RAM block also matches exactly my Thermaltake meter.

 

[electrosoft]

Signal RGB is *almost* perfect for controlling all my devices. For some odd reason, it appears it won't work with AMD GPUs? It controls my Steel Series, Razor and Asus devices perfectly. Luckily the 9070xt has minimal RGB on it.

 

 

[Mr. Fox]

2 hours ago, electrosoft said:

Signal RGB is *almost* perfect for controlling all my devices. For some odd reason, it appears it won't work with AMD GPUs? It controls my Steel Series, Razor and Asus devices perfectly. Luckily the 9070xt has minimal RGB on it.

 

 

Give Argus Monitor a try. It works for RGB and fan controls on all of the ASUS and Gigabyte motherboards I have used it on. I disable all of the drive and temperature monitoring, system tray temperature icons and other things I do not care about (I have HWiNFO64 for that). The UI looks a little bit dated, but the important thing is that it works well for me for controlling the fans and RGB headers. It also does not need to stay running once the settings are applied.

[jaybee83]

On 6/1/2025 at 4:01 AM, Mr. Fox said:

If I had to wager I would bet that it's true because NVIDIA is who they are... control freaks. If they can find a way to screw the people that buy their products that is exactly what they will do. That's how they roll and who they are, and that is how they have been for a long time. Also one of the numerous reasons I dislike FE video cards.

---

And, on a happier note, I ordered my second Lian Li O11D XL EVO to house the second AORUS Master. Also ordered a D5 distro block from Radikult Customs and about 12 feet of Alphacool EDPM tubing.

 

Now I hear rumors of a 9080 XT with 32GB of GDDR7. I am so glad I didn't blow $3000+ on a 5090 already. Now it is no longer even tempting. Not caring actually feels really good.

 

On 6/1/2025 at 7:24 AM, Mr. Fox said:

I was watching a Moore's Law is Dead video and apparently it is based on some leaked benchmarks.

 

As long as the temps are not 100°C I would gladly buy a 4090-level AMD card for the price of a 5080 or less. So, I hope it is true, but we shall see. AMD really doesn't cater to performance enthusiasts. Their niche is more geared toward mainstream on everything.

 

 

 

oh that would be very sweet if they came out with such a card. there is a possibility that AMD didnt expect Nvidia to nerf the 5080 THIS HARD, so theyre now seeing an opening to slot in smth between the 5080/4090 and 5090.

 

21 hours ago, Papusan said:

 

5090 is just an expensive midrange high end gamer cards.  The more you pay the more you save. Here's Nvidia's new  "gamer "flagship" card😀

 

 

Not sure this would please @electrosoft😁 The $10.000 cards almost sounds like a real circular saw. Very special sound from it. Not the worst I have heard but the sound from this is awful and more annoying than this one from my older post

 

 

boiz@5090 owners, welcome to midrange 🥲 even more so cuz the Pro 6000 can be oveclocked! now lets see if Roman will design and make his very own custom water block for this card. seeing as he actually has all the milling equipment and know-how, this would be perfect as a part 2 for his "project irrationality" 😄 

[Papusan]

On 6/1/2025 at 2:55 AM, Mr. Fox said:

12VHPWR or whatever popular thing they are calling today, or whatever they decide to rename it tomorrow, is truly a POS connector that never should have seen the light of day. It is a sketchty, ill-conceived, fragile, unreliable and dangerous piece of crap. Nothing they can do to it--ever--will make it a good and reliable power cable. The group at NVIDIA responsible for bringing this abortion to market need to be stripped naked and caned in public until dead, then resuscitated, their employment terminated, citizenship revoked and then deported to a far away horrible place where life is miserable. 

 

We have now come so long that buying etc a new PSU ain't enough. You need buy spare power cables to be able to use your PC. Good luck try do that from some PSU vendors/manufacturers (Msi we look at you). Some will refuse offer you spare parts. Either due nothing in stock, don't sell spare parts or being difficult. Newer is newer. Not always better. The trend is crystal clear... You don't get what to expect. Not even if you buy the more premium products.

 

 

[Mr. Fox]

1 hour ago, Papusan said:

 

Looks like NVIDIA moderators removed the post. What did it say?

[win32asmguy]

15 hours ago, Clamibot said:

Intel's server class Xeon CPUs have all P-cores and they perform great due to not being shackled by E-cores. No weird scheduling shenanigans.

 

Oh, the Xeons can also have scheduling issues. I have had trouble with CPPC, so processes on the 2495X do not automatically run on the highest clocked core or the favored cores. Also with hyperthreading enabled it can cause issues with the scheduler similar to what you would see on the 7950X and 9950X. In latency sensitive applications some cores are also more mesh hops away from a given DDR5 or PCIe5 block so depending on which slot you have a GPU in some cores may be better to use than further away favored cores.

 

It will get more interesting if we end up getting a Granite Rapids WS Xeon at some point, with Redwood Cove P-cores. Maybe a LCC or MCC monolithic variant could be tuned to give a 285K or 9950X a run for its money. It would be great if it could also not require registered memory (as was the case on the 3175X) and the latency penalties associated with it.

[Papusan]

10 minutes ago, Mr. Fox said:

Looks like NVIDIA moderators removed the post. What did it say?

 

He did some changes to his pc. + maintenance. Aka he did a few disconnects of the trash cable. Once he should start use it the Astral monitoring showed imbalanced amps. Total out of specs. He bough new cable and everything was well again. Just show how unreliable this tiny trash is. Aka saved by the Astral monitoring. nvidia should have done what Asus did with the ROG tax and implemented monitoring of the power pins/rails. Maybe they don't like critics of Nvidia?

[Mr. Fox]

2 hours ago, Papusan said:

 

He did some changes to his pc. + maintenance. Aka he did a few disconnects of the trash cable. Once he should start use it the Astral monitoring showed imbalanced amps. Total out of specs. He bough new cable and everything was well again. Just show how unreliable this tiny trash is. Aka saved by the Astral monitoring. nvidia should have done what Asus did with the ROG tax and implemented monitoring of the power pins/rails. Maybe they don't like critics of Nvidia?

Everything would be so much better if NVIDIA would just admit 8-pin legacy PCIe power connections are better and recommend their AIB partners go back to it. But, they'll never do that. It is hard to respect people or companies that cannot admit that sometimes change is not good and demonstrate unwillingness to correct their own mistakes. Pretending to be right when it is obvious to everyone they are wrong makes them unworthy of respect.

[Papusan]

 

LOL The 5090 is all too cheap. 

 

[Clamibot]

2 hours ago, Mr. Fox said:

Everything would be so much better if NVIDIA would just admit 8-pin legacy PCIe power connections are better and recommend their AIB partners go back to it. But, they'll never do that. It is hard to respect people or companies that cannot admit that sometimes change is not good and demonstrate unwillingness to correct their own mistakes. Pretending to be right when it is obvious to everyone they are wrong makes them unworthy of respect.

 

This is the phenomenon of the "Intelligent Idiot". It's a very strange phenomenon where some otherwise intelligent people double down on their position even when they've been proven wrong because they cannot accept that they're wrong. Some highly intelligent individuals are also higly out of touch with reality.