Etern4l

If it's a locked up modem/router combo (as seems to be the case), then it would take a major hack to make progress I guess. Really dodgy ISP.

Interesting. Which specific power limits include power consumed by iGPU? @Papusan

When active, the iGPU consumes some RAM bandwidth (about 1GBps?). Also remember there tends to be a fair amount of variance in CB23 results.

Thanks, appreciate the guidelines. Why does the board try to keep it so low though? Or is it very dynamic when on auto, i.e. it's showing as 1.16V in BIOS but actually goes up significantly under load? Core count is just one part of the overall performance equation. We can keep the discussion more focused by considering options within, say < $2k per CPU range. Not much there apart from the latest consumer blockbusters, right?

Yes, it's not so simple - apart from the measurement issues (including the potential for fake numbers being advertised by less scrupulous manufacturers), you also need the paste not to disintegrate in a matter of days lol. Amazon reviews are definitely not reliable. A ton of fake ones, and the negatives can be suppressed. I have personally had my negative reviews not appear at all. No communication from Amazon, just ignored.

It can be very difficult, we are hardwired to love sugar as you know. What works is getting rid of all the sugary stuff from the house, unfortunately that can be a challenge in itself unless you live alone. Basically, you have to treat this as addiction. It's also possible there are some triggers, like stress for example that need to be addressed as well.

I looked into those boards "with Thunderbolt functionality" for the 5950x and it turned out to be a poor marketing gimmick. It was actually USB4 not TB4, eGPUs wouldn't work for instance. Yet, they were unashamedly marketed as supporting Thunderbolt... Asrock is Asrock I guess. Perhaps something changed this gen, but I doubt it. Even if AMD suddenly managed to come up with a full TB implementation, it would have to be a pure and unadulterated 100% compatible implementation otherwise God, or at least Linus himself, help those trying to get this to work on Linux this decade. This is one of the problems that has always plagued AMD. Yes, looks good on paper, but then there are these seemingly minor incompatibilities which break things in an annoying way. Probably looking better now than in the Opteron era I am referring to, but still def a gamble to some extent, depending on the application. Gaming should be fine. AMD seems to be investing heavily in the area (possibly to the detriment of other, clearly neglected, domains), as evidenced by their aggressive acquisition of exceptional gaming hardware talent such Frank Azor ;) ( @Papusan, you have the floor sir, my popcorn is ready lol ) It used to be that you had to pay a premium for the peace of mind and freedom from compatibility issues, but thanks to "the invisible hand of the market" (concretely: the hand/wallet of gamers buying AMD) now we pay about the same or less, relatively speaking ;) Edit: Quick question about CPU VDDQ on the 13900K: BIOS default is 1.164V, apparently raising this to 1.2V could help with memory OC. Any downsides? Same question regarding SA voltage: also around 1.164V in BIOS. Thanks a lot.

"Look at the TCO graph here. Total annhilation." My preference would be to do my own calculations rather than rely on AMD marketing to do the thinking for me. Have we seen the actual vendor pricing for this versus Sapphire Rapids yet? What about in-depth benchmark results? I'm not saying AMD's assertions are wrong, we just don't really know (presumably). And again, the value proposition may well vary depending on the use case. "I think AMD Ryzen 7000 is a superior product for many users except to those who tweak" I am not sure this has much to do with tweaking, frankly unless you have a ton of time and won big in the silicon lottery, the benefits are not as great as they used do be (except for the unrelated memory overclocking) - look at the minor gains at the end of the day: 43000 vs 41000 in Cinebench 23 which likely wouldn't be stable under extended heavy load anyway. As explained earlier, Intel has significant advantages on scientific computing side. Sure, if you are running something like a web server, AMD looks like a strong option to consider. I just I remembered one more reason why I didn't go with AMD: the lack of Thunderbolt. That's it, if you need it, then AMD is not really an option.

Come on, be fair. We've had good 2 gens on LGA1700. Yay! ;)

I don't think I've seen that many. Mainly one guy posting? I've found a couple of promising comments on reddit as well, but that's pretty much it. It was a gamble. I mean it could have been a bad batch or something, but that's just as unacceptable. Arctic just pulled MX-5 over consistency issues.

Ha, basically a US version of Sherlock, which was pretty good. Noted!

Really nice. Looks like you're ready to attack those IndigoBench and Blender CPU scores :)

Spread. It was perfect, and like I said - temps seemed very good immediately after application.

That's cool. Just checking. Thank you for supporting AMD, without people like you Intel would have probably become another Nvidia. I was very seriously considering going with a 5950x earlier this year, but just couldn't risk the potential issues (and 12th gen/DDR5 was already available). A bit of an edge case on my end really. I just find it a little funny how the tables have turned and now AMD is the more expensive option (although just marginally, it seems).

Seems like budget is a consideration. Why AMD then? Intel is supposed to be better value this time around (in addition to be being better overall, arguably).

Re Intel: It's an interesting one. On one hand, the software is available for free in many cases, they invested a lot of money in it, and it is much better than the open source alternatives, so it's hard to blame them for using it to promote their hardware. Same for NVIDIA. One could say that the blame here rests squarely on AMD for failing to compete on the software side. On the other hand, this could be viewed as anti-competitive practice. NVIDIA is safe here, because their software only works with their hardware, but Intel is treading on dangerous grounds. Clearly, if AMD could sue them over this, they would, so more stringent laws would need to be in place to address this particular case, and this is always tricky. Presumably Intel can just say the optimize the code specifically for their CPUs and don't care about AMD compatibility or performance. BTW The SYY 157 turned out to be so bad, I posted a mini-review to warn people. All good after Phobya repaste.

I followed an interesting tip and recommendation of a new paste called SYY 157 that was provided in the Thermal Pastes thread. The product is marketed as rivaling the thermal conductivity of Phobya NanoGrease Extreme, which is what lured me in. It is available for an affordable couple of bucks from a Chinese brand on Amazon. Executive summary: everyone likes a good deal, but in this case you get much less than you paid for, taking the waste of time into account the value is hugely negative. The packaging was nice (spatula and spreading card included). The application looked perfect, I even waited the 15 min they strangely recommended before applying the heatsink. Temps seemed fine right after application, but initially stealthy deterioration started basically right after, manifesting itself in progressively more instability. This resulted in enormous waste of time as I blamed the problems on everything but the paste, I couldn't imagine it could be so bad to cause the issues. After about 10 days it became clear something is wrong. Noticeably higher idle temps, instability on stock settings etc. When I removed the heatsink, it felt like there was some gray talc sprinkled on the IHS. The paste almost turned into powder. Luckily I still had enough Phobya Nanogrease Extreme left, everything has been working great after the repaste so far. Now, the paste was used on a large desktop CPU, with around 270W of power going through it for very long periods of time, and temps in 80-90C range. It's possible, that the paste would deteriorate at a slower rate when used more casually, but I wouldn't risk it. SYY 157 Product Rating: 0% (absolute garbage, dangerous, bin on sight)

The SYY 157 turned out to be so bad, I posted a mini-review to warn people.

Correct me if I'm wrong but AMD has nothing in response to Intel MKL for instance. Heck, Intel even have their own Linux distro which absolutely destroys Windows on performance. On the NVIDIA front, their response to CUDA is called ROCm, and apparently doesn't even run on recent (or any, I'm not sure) consumer cards.

Yes, but I'm sure they did the math and worked out they would get creamed by AMD in multicore had they not pulled the E-Core trick, and probably concluded that 8 high-clock cores are enough for most consumer use cases. Will be interesting to see what they are planning to do next. My iniital impression of what bro Papu posted earlier was that 14th gen looks like 11th all over again. Oh dear, might not be good enough against the chiplet swarm. I will be surprised if they can't get those e-Core clocks much higher up, and - for their sake - I hope they do.

I don't think it's about the cost. Probably cost them a lot upfront in R&D to come up with the architecture, drivers etc. The E-Cores are significantly slimmed down: no HT, no AVX. Kind of moving a bit towards the RISC architecture. What's not clear to me though is why they actually have to clock lower - I would have expected the opposite, given the simpler/streamlined design. Haven't done any reading on this though. Put another way: had they gone all-PCore, you would have something like 14/28 (if that) vs 24/32, where half of the 28 would be due to HT, which slower than an E-Core.

I hear you, this almost put me off the Intel route and certainly cost me a lot of time in getting stuff to work properly, but they probably had no choice given the ancient process. How many more P-Cores would they fit in the space taken up by E-Cores? Good news is that the end result is surprisingly effective. And don't get me started on silicon lottery. All those cr@p chips you guys return in disgust? They just sigh and send them down to these sunlit uplands lol I feel Dell does the same with laptops.

My impression is that AMD is lacking mostly on the software side, and that gap is closing somewhat. Similar on the CPU side, maybe less of a gap in practice, but it's still there. They need to be doing more hardcore work! ;)

In other news, AMD is opening a network of friendly kidney extraction clinics to help customers deal with the adoption cost of their latest EPYC 'Genoa' platform. Meanwhile "Black Friday" means "Blackout" for Nvidia. Only 3090Ti is released in small, AI-optimised batches. Even the 4080 cr@p sold out here.

I would not really fault Intel, they did us a solid with the 13900K, however, Jensen could use a vacation or ideally exit the company for sure. The problem with them is that they have AI as a presumably massive backup - I wonder what share of their business consumer video card market represents (I'm sure @electrosoftknows!). Similarly, AMD and Intel have the server market to support them, they would just deprioritise consumer product lines.