GPUs & CPUs & Enthusiast hardware: Questions, Discussion and fanboy slap-fights - Nvidia & AMD & Intel - Separe but Equal. Intel rides in the back of the bus.

TheRaptorJesus said:
AIO users lol!

I need to fix my page since I consolidated fans and reinstalled windows and lost my virtual delta sensors. The sensors without the decimals are the shitty sensors built into the mobo and no the octo.
View attachment 6174403
Click to expand...
Nice setup! I love the Octo, but I only have one thermistor plugged into each of mine. One on the external radiator, one internal because for some reason the motherboard can fit four LED headers but no thermistor header even though that’s half the size.
Overly Serious said:
I like this new Wendell, Champion of Truth and Justice. Last week he was chewing out MS for their ARM support. Anyway, bad timing for Intel as AMD are getting very close to their Ryzen 9000 release which is looking very promising.


And review samples have apparently just gone out. Expected launch date, end of this month:

I'm seeing the usual "boorrrrring, 14% improvement is worthless" crowd emerge but I think this looks really promising. Efficiency seems much improved. Example the leaked scores show the 9700X 10% faster in Single Thread (4% faster in MT) over its predecessor 7700X but the 9700X is a 65W processor. Its predecessor is 105W. I could be wrong if the leaked scores aren't on stock voltage and similarly if the benchmark includes AVX-512 then that might throw it off from many real world scenarios. But to me this is looking really hopeful. Dropping CPU power down to 60% of its predecessor would be impressive even if performance was just at parity.

I'm assuming memory controller is the same though if it did let you use faster RAM, that would be awesome. 8000MHz RAM here we come.
Click to expand...
Worth bearing in mind that Wendell has a strong AMD bias. To the point that Intel won’t send him review samples.
But yeah, it does look like this is a real issue.
I’m personally hoping for CAMM on desktop. That should bring us actually quite close to soldered memory performance.
 
Susanna said:
Worth bearing in mind that Wendell has a strong AMD bias. To the point that Intel won’t send him review samples.
Click to expand...
The Alderon Games post is pretty damning. Thermal stresses leading to material failure is my theory.
 
The Ugly One said:
The Alderon Games post is pretty damning. Thermal stresses leading to material failure is my theory.
Click to expand...
Something is definitely wrong, but I’m not sure it’s thermal stress. Intels run hot, but TSMC5 is handling the same kind of thing (7950X overclocked to 105 degrees say) just fine, even for long periods. When the motherboard vendors were “killing processors” by running the memory controller far outside spec the failure rates were still very low. I think the problem is more likely to be inherent to Intel7.
 
  • Thunk-Provoking
Reactions: Vecr
The problem isn't just temperature. Ice Lake Xeons run hot, and Sapphire Rapids is an inferno. But those use nearly square dies. The problem is the shape of the chip. 7950X is three chiplets, one of them square, none of them as long and thin as recent Intels. Not only that, but the E-cores, which run cooler than the P-cores, are all clustered on one end. Much more uneven thermal distribution = much higher physical stresses.
 
  • Thunk-Provoking
Reactions: WelperHelper99, Vecr and Susanna
Susanna said:
Worth bearing in mind that Wendell has a strong AMD bias. To the point that Intel won’t send him review samples.
Click to expand...
I think I recall him being a bit damning in the past about Intel's marketing BS (though I may be thinking of Anandtech) but he's been pretty excited and enthusiastic about upcoming Lunar Lake.

Frankly, if I were Intel, I'd be reluctant to send review samples to a neutral and crazy-thorough reviewer as well.
 
  • Like
  • Agree
Reactions: Vecr and Susanna
Arrow Lake desktop chiplet layout leaked:
Intel Arrow Lake CPU Architecture To Feature Four Tiles, CPU Tile Includes Coherent Fabric Connecting P-Cores With E-Cores
Intel-Arrow-Lake-CPU-Architecture.jpeg

Same number of "tiles" as Meteor Lake. No LP E-cores on the SoC tile (RIP, we hardly knew ye). Like Meteor Lake, display and video decode are located on the SoC tile instead of the GPU tile, which could improve power efficiency when users are mindlessly watching videos.

You get an NPU everywhere, also on the SoC tile. Rumored to be a weak 13 TOPS one, so no official Copilot+ support. It might be the same one in Meteor Lake with higher clocks.

AMD is supposedly planning a cheaper Hawk Point SKU with the NPU disabled: the 8745HS. I wouldn't count on seeing much of that from either company if there's pressure to keep them in and they are physically smaller than the iGPUs which are the first target to disable on desktop. The move to higher yield chiplets might make it more feasible to give every desktop CPU an iGPU/NPU, with no more 'F' models. If Intel/AMD boost their NPU performance to 100+ TOPS in the future, they could disable some bad yields down to half the performance and still meet the Copilot+ requirement.

Leaks have said the Arrow Lake iGPU will have up to 4 Xe-LPG cores (64 EUs). Compare to UHD 770 on the top Alder/Raptor Lake desktop CPUs, which has only 32 EUs of Xe-LP (Gen 12.2). UHD 770 is about as fast as the iGPU in Ryzen 7000 and presumably Ryzen 9000 desktop CPUs, so Arrow Lake will surpass AMD in what would be considered non-APU graphics. Meteor Lake-U (not -H) also has 4 Xe-LPG cores (64 EUs).

The Ugly One said:
An i9-12900 is basically an i7-13700 with slightly lower clock speed (and less cache? Don't remember and don't feel like looking it up)
Click to expand...
I heard the 12900K is fine. Raptor Lake variants 13900K(S)/14900K(S) may be the only ones to watch out for.

I don't think Alderon Games named any specific chips in their statements. If Core i5/i7 owners start having problems, that's a much greater disaster for Intel.
 
Last edited:
  • Like
Reactions: Vecr and Grim_Autismo
If Intel's planning to release a flagship CPU that doesn't support Copilot+, what even is the point? Suggests nobody really gives a shit about these AI features.

My i9-12900 is fine. Given that 12th gen isn't named as having issues, and i3-i7 13th & 15th gen CPUs are nearly identical to 12th gen CPUs, that suggests the problem really is that the i9s just pushed the die size & clock speed too far.
 
  • Agree
  • Like
Reactions: WelperHelper99, Lone Wandering Courier, Vecr and 1 other person
The Ugly One said:
If Intel's planning to release a flagship CPU that doesn't support Copilot+, what even is the point? Suggests nobody really gives a shit about these AI features.
Click to expand...
Making NPUs ubiquitous makes it more likely for software developers to support them, even if the performance is low. Then the performance can get scaled up in subsequent generations.

It looks like Intel is copying and pasting the same IP block from Meteor Lake. So why was an even weaker ~10 TOPS NPU in Meteor Lake? Many say that was also some bullshit but Intel's reams of marketing materials say otherwise.

What we can say is that these weaker NPUs are physically smaller and therefore cheaper to include. We don't have die areas for Lunar Lake and AMD XDNA2 NPUs yet, but during the Strix Point announcement, AMD CEO Lisa Su joked on stage with a Microsoft guy that they were using a lot of die area for theirs. It's probably 2-3x larger than the previous one.

New ARM-based devices include NPUs as weak as 0.5 TOPS (500 GOPS), which is apparently enough for simple tasks like real-time facial/object recognition.

As seen from Meteor Lake demos, basic stuff like webcam blurring, image generation, etc. can use this level of performance. Cooler stuff like Microsoft's AutoSR (apparently) require the Copilot+ level (40 TOPS or greater).
 
  • Thunk-Provoking
Reactions: Vecr
Get the rope Macaulay! said:
any news on the 5000 series? will they learn from their fuck ups with the 4000 series?
Click to expand...
5090 and 5080 come out first. 5090 should be a 28 GB card using GDDR7 (448-bit memory bus), easily 50% faster than the 4090. The 5080 needs to be gimped below 4090 performance to be able to avoid US sanctions and sell to China, with only 16 GB.

Other cards below that will come out later (to give more time to sell off low/mid 4000 cards) and will be boring and similar to 4000 series. Nvidia is definitely thinking of making a 128-bit 5060 8 GB, for example.

Use this as a starting guide:


Warframe devs report 80% of game crashes happen on Intel's overclockable Core i9 chips — Core i7 K-series CPUs also have high crash rates



recall.png

Recall Lake!
 
Last edited:
  • Thunk-Provoking
Reactions: Wooo and The Ghost of Kviv
Cards such as the 7700 XT and 6800 non-XT were as low as $360 (US) recently, probably better buys than remaining 6750 XT stock which was what, $300-320? But I hope you have a fun time with it.

@The Ugly One From the video I linked:

moorenpu.jpg

Moore's Law is Dead alleges that the ~13 TOPS NPU in Arrow Lake desktop CPUs may be used for "Machine Learning Thread Scheduling". Quote: "I think this is probably going to be some sort of APO 2.0 that is only going to be on Arrow Lake-S, because Arrow Lake-S is a desktop part and therefore they don't care if they're wasting power powering on the NPU constantly for scheduling when it needs to."

Furthermore, Arrow Lake Refresh will get a Copilot+ ready NPU (40 TOPS or greater). A planned die with 8 P-cores and 32 E-cores was cancelled.

Another interesting thing to me is that Arrow Lake-U (for mobile) supposedly has 6+8 cores instead of the typical 2+8, along with 2 LP E-cores, and that Intel will use their own node for that one to diversify and ensure a higher volume. If that's the die used for low power non-premium laptop chips, then the dual-core i3-1115G4 which became a 6-core i3-1215U is now likely getting even more cores.
 
Last edited:
  • Informative
Reactions: WULULULULU
The Mass Shooter Ron Soye said:
Cards such as the 7700 XT and 6800 non-XT were as low as $360 (US) recently, probably better buys than remaining 6750 XT stock which was what, $300-320? But I hope you have a fun time with it.
Click to expand...
Thanks. I wanted those two but sadly nada from my place. To compare: (converted from my currency)
6750XT: 385.19 USD
6800 (non XT): 436.54 USD (Newegg doesn't count due to high shipping fees, also only stock comes from a shady store)
7700XT: 453.58 USD
Feelsbadman. I wish I was in the US for easy GPU stocks without high shipping costs.
 
  • Informative
Reactions: The Mass Shooter Ron Soye
The Ugly One said:
If Intel's planning to release a flagship CPU that doesn't support Copilot+, what even is the point? Suggests nobody really gives a shit about these AI features.
Click to expand...
I'm pretty sure Copilot+ is already DOA. Sales for the new AI PCs are already kinda anemic (although that's partially a product of just low demand for PCs right now) and among the people who actually buy them, you know what the biggest reason they give is? Better battery life.

People in this thread who don't think battery life is a top 3 factor when it comes to normal consumers buying laptops are delusional.

Whether or not NPUs will be relevant or not remains to be seen. Anecdotally though I can say that Microsoft's marketing of AI integration in Windows has been a colossal failure and might poison the well for this shit. Which is a shame because most of the functionality we saw was pretty decent, it's just that Microsoft could not control themselves and had to make the whole thing absurdly intrusive to pump some metrics so a middle-manager could finally get his wings or some shit.
 
The Ugly One said:
Interesting that it wasn't a problem with 12th gen, since 13th and 14th are barely changed from 12th.
Click to expand...
Every day I am so glad I chose 12th gen. I heard rumors about stability issues and went NOPE, not on my rig.
The Ugly One said:
Since it gets worse over time, and it only seems to be happening to the i9s with the super-sized die, it probably is the heat from being clocked to the edge degrading the chips. They pushed this thing as hard as they could to win internet pissing matches, and now they have a failing product. These rectangular chips have a flexing issue (an aftermarket bracket mitigates this), and I can't believe it's any less of an issue on the largest, hottest chips.
Click to expand...
The speeds on 13th and 14th gen are nuts and suck power like no tomorrow. If it isn't this, they have something more fundamentally wrong wearing out 13 and 14th gen i9s, which is not fun to think about for anyone that bought those chips
The Ugly One said:
An i9-12900 is basically an i7-13700 with slightly lower clock speed (and less cache? Don't remember and don't feel like looking it up)
Click to expand...
Basically yes. A 13700 is just a product improved i9 12900 with newer manufacturering processes. I looked into it when building my PC, didn't find it worth it to go to 13th gen when a 12 gen i9 was basically equal to a 13 gen i7 and with potentially better stability (which I've been vindicated).
 
Somewhat unconfirmed, but there's reports of the 13700s and 14700s having above average failure rates, albeit nowhere near as high as their bigger brothers.

And that the Handbreak devs have also been seeing reports of those CPUs failing.
 
You must log in or register to reply here.
Back