Question Intel's current woes and the low end of the desktop CPU market

[hemedans]

So you want to compare it went it is heavily overclocked when it comes to performance but at stock when comparing power?

And then compare prices only when you can get it at an all time low price? If you want to argue it can be a good value that's one thing. But get your facts straight.

Video I posted has both Stock and overclock power consumption, even overclocked you can see power consumption is low when Gaming.

The guy I quote he mentioned black Friday deal $55 for Ryzen 5 5500, thats not normal price it's a deal, so was my price, and when you build budget build you need those deals.

 

[DAPUNISHER]

I don't pretend to know anything about the Chinese market. And ignoring the "off the charts" hyperbole; That screenshot is a problem. The rules here require all content be in English. Please replace that screenshot, providing a translated one instead, or at minimum put it in spoiler tags. TIA for your understanding and cooperation.

There is no getting around 4c/8t being an issue in more recent CPU demanding games, here on the precipice of 2026. Including 2077 in its latest iteration. That brings us to the fact the provided TG video is over 3 years old. That alone makes the results irrelevant for shoppers looking to build now. Win 10 is another issue. Few building now will be using it for obvious reasons.

Continuing, I think this is an occasion where the messenger deserves to be shot. TG's testing methodology is erratic and untrustworthy like all side by side channels. Hell, most of them are well known to fake results.

Here's a few things that immediately jumped out at me besides being outdated. The first game is 2077, as noted already, is an old version. It's not being tested with max crowds and traffic. The streets are almost empty and there are no NPCs. Also, not using RT on a 3080tie at 1080 with DLSS in 2077? LOL. Even if the crowds and traffic were maxed, the failure to test with RT on, in a CPU demanding area, results in the test being considerably less CPU intensive i.e. next to worthless.

Next thing that jumped out is the 5600X pulling a 100W+ but the clockspeed is low, something is nerfed. Here's GN's launch review from almost exactly 5 years ago, for power usage in Blender, a very heavy workload.



That's what stock looks like to this day. I've had multiple 5600X and they all did 4.85GHz boost in ECO mode (45W TDP) with +200MHz boost. Takes all of a minute in the UEFI to get that nicely improved performance per watt on just about any B series or higher AM4 board with the latest bios and Zen 3 CPU. You can usually even enable it on really cheap A series boards using Ryzen Master. No being limited to buying boards with external clock gen, which usually cost more here.

I'm not going to debate which is a better value any longer after this post. I do think you did a good job of presenting use case for the i3 f variant. However, as stated, few outside of your market seem to care.

 

[coercitiv]

OP:

I'd like to talk about low budget Intel CPUs with integrated graphics.

The thread:

The King of value CPUs is 5950x - 16 full supa' fast cores with zero iGPU. Use the first CCD for email, keep the other CCD for your next low budget build.

Here's 12100F overclocked on a $600 board running games on a $1200 GPU. Zero iGPU for low budget, use savings to buy MB with Bclk OC support. And a good cooler. The more you buy, the less power it uses!

Also this thread:

CPU with a stock PPT limit of 76W is using 100W at... uhm... "stock settings".

 

[DrMrLordX]

Regardless of how good the 12100f is or isn't, the reality is that Intel can't afford to sell anything newer as a sub-$100 loss-leader (there are 13100f and 14100f CPUs for around $97). Though I kinda wonder how much a NUC based on Arrow Lake-U would cost.

 

[511]

Regardless of how good the 12100f is or isn't, the reality is that Intel can't afford to sell anything newer as a sub-$100 loss-leader (there are 13100f and 14100f CPUs for around $97). Though I kinda wonder how much a NUC based on Arrow Lake-U would cost.

ARL-U is expensive... PTL-U is the replacement with much better cost to Intel

 

[DrMrLordX]

ARL-U is expensive... PTL-U is the replacement with much better cost to Intel

ARL-U is Intel 3, is it really that expensive? Compared to 18a? I'd think not. I'd think the worst part about ARL-U is the iGPU and having no real dGPU option since the board probably wouldn't expose enough PCIe lanes or even a slot for one.

 

[511]

ARL-U is Intel 3, is it really that expensive? Compared to 18a? I'd think not. I'd think the worst part about ARL-U is the iGPU and having no real dGPU option since the board probably wouldn't expose enough PCIe lanes or even a slot for one.

Only the 40mm2 tile is Intel 3 rest is TSMC on PTL except for the IO Tile everything is Intel

 

[DrMrLordX]

Only the 40mm2 tile is Intel 3 rest is TSMC on PTL except for the IO Tile everything is Intel

Okay but the other tiles on Arrow Lake-U are . . . N6 are they not?

 

[511]

Okay but the other tiles on Arrow Lake-U are . . . N6 are they not?

Yes and N5

 

[DrMrLordX]

Yes and N5

As compared to doing everything on 18a, I wouldn't think those tiles would be terribly expensive. Keep in mind the part volume would still be pretty low. Though with the (apparent) decision on Intel's part to cut their main Diamond Rapids platform, they may have a glut of 18a moving forward.

 

[511]

As compared to doing everything on 18a, I wouldn't think those tiles would be terribly expensive. Keep in mind the part volume would still be pretty low. Though with the (apparent) decision on Intel's part to cut their main Diamond Rapids platform, they may have a glut of 18a moving forward.

You are forgetting everything made on TSMC means lower volume for IFS hence Intel's overall profitability 18A is good for IFS books not N5 though

 

[DAPUNISHER]

iGPU isn't something I'd concern myself with for a mom and pop email+internet+streaming daily driver PC. Here in the U.S. a tiny, low power, low profile, 2GB R5 R430 is $10 shipped on Ebay. A modern i3 or Zen 3 can easily handle 4K playback in formats the card does not support.

 

[511]

a N100 has one tbh alongside QSV

 

[DAPUNISHER]

a N100 has one tbh alongside QSV

Yup, I suggested a cheap mini PC early in the thread. QSV is a silly feature that will not get used on mom and pop builds, but it's better to have it and not need it, than need it and not have it. I doubt they are a good revenue generator for a break/fix biz though.

 

[Thunder 57]

Yup, I suggested a cheap mini PC early in the thread. QSV is a silly feature that will not get used on mom and pop builds, but it's better to have it and not need it, than need it and not have it. I doubt they are a good revenue generator for a break/fix biz though.

QSV is nice to have. It has/had a niche but really isn't a big deal compared to 10+ years ago. It certainly not something a basic user cares about or knows how to use. I wouldn't want to be the "friends and family" techie trying to explain that one over the phone.

 

[Thunder 57]

iGPU isn't something I'd concern myself with for a mom and pop email+internet+streaming daily driver PC. Here in the U.S. a tiny, low power, low profile, 2GB R5 R430 is $10 shipped on Ebay. A modern i3 or Zen 3 can easily handle 4K playback in formats the card does not support.

Unless you are going super low budget I would prefer an all in one chip soultion.

 

[DrMrLordX]

You are forgetting everything made on TSMC means lower volume for IFS hence Intel's overall profitability 18A is good for IFS books not N5 though

18a would have been fully-committed had Intel not meddled with their own server lineup. They're gonna be volume-constrained until 14a I thought?

 

[511]

18a would have been fully-committed had Intel not meddled with their own server lineup. They're gonna be volume-constrained until 14a I thought?

18A don't have capacity constrains they have Fab 52/62 both are capable of producing it but it is more Capex and Money problem for Intel ramping fans requires $$.

 

[DAPUNISHER]

Unless you are going super low budget I would prefer an all in one chip soultion.

That's why mini PCs are an attractive solution in the category. It'd be hard at the moment to avoid used parts and build something comparable BNIB to what This mini PC costs - $170 with 10% off at checkout with 16GB of ram.

 

[dullard]

Not Intel 7. Intel raised Raptor Lake prices because people weren't buying enough Arrow Lake.

How would Intel 7 not be substrate / wafer constrained? As in, if there is a significant shortage, how does Intel 7 get around that?

 

[Joe NYC]

How would Intel 7 not be wafer constrained? As in, if there is a significant shortage of wafers, how does Intel 7 get around that?

Do you mean raw, unprocessed silicon wafers? I don't think those are in short supply. if they are, it is news to me.

 

[dullard]

Do you mean raw, unprocessed silicon wafers? I don't think those are in short supply. if they are, it is news to me.

Yes as in raw unprocessed materials including substrates, wafers, and related other supplies. I should clarify that the biggest problem is a shortage of substrates in general--not just wafers. The whole supply chain is under strain. AI demand is sucking up all supplies leaving shortages everywhere.

Seems like memory was hit first. Then HDDs are becoming a shortage. Then geopolitical things have disrupted Nexperia's wafers and auto companies are being impacted. Hopefully the current rare-earths political tensions don't lead to other shortages. Intel 7 is affected too.

This isn't limited to Intel. It is an industry-wide shortage. The short term impacts will be low-profit items will become more expensive or almost non-existent for a while. This includes the low-end of the desktop CPU.

https://www.waferworld.com/post/top-challenges-silicon-wafer-manufacturing-will-face-in-2025

Wafer Shortage: The global supply chain disruption has affected the availability of semiconductor wafers. This shortage has significantly impacted various industries, causing delays, increased prices, and shortages of products that rely on semiconductors.

The shortage came about due to low 2023 and 2024 wafer production:

https://www.semi.org/en/news-media-press-releases/semi-press-releases/global-silicon-wafer-shipments-to-remain-soft-in-2024-before-strong-expected-rebound-in-2025-SEMI-reports

That low production hit just as AI started its boom. Wafer production was supposed to increase in 2025 but now there are those geopolitical fights.

 

[DavidC1]

As compared to doing everything on 18a, I wouldn't think those tiles would be terribly expensive. Keep in mind the part volume would still be pretty low. Though with the (apparent) decision on Intel's part to cut their main Diamond Rapids platform, they may have a glut of 18a moving forward.

The alternative is 18A + Intel 3. That should be cheaper and more profitable for Intel as it will fill fab space and that directly affects entire revenue not just margins. 18A and Intel 3 would have to be dramatically more expensive than TSMC in order for going to TSMC be cheaper.

They moved off their own process way too quickly, without having a viable 3rd party to replace the lost volumes. Now they have to deal with increased fixed costs of building a cutting-edge fab and loss of volumes of significant Intel products going off their own process. But this isn't the first time I saw companies in a decline make choices that accelerate the decline, when the management in charge intended to be otherwise. RIM/Blackberry did the same thing, and HP announced that they would cancel all future PC development, when they were #1 volume leader of PCs at that time.

I think moving off their own process too quickly might have started under Swan, but Pat didn't seem to have done anything to slow down that trend, and think a bit before making decisions. Another bad decision Pat made was giving IFS "unlimited budgets" when actually setting a clear boundary and limits are what forces the team to be more efficient. Of course if you truly have unlimited budgets, you may brute force the win, but no one has that, certainly not Intel, not even in 2016.

 

[regen1]

I think moving off their own process too quickly might have started under Swan, but Pat didn't seem to have done anything to slow down that trend, and think a bit before making decisions.

Under Bob Swan designs had to move to TSMC(most notably N3 for Arrow Lake Compute) because original Intel 7nm(later Intel 4/3 family with updated targets) development was late. Also the path they were going with, Intel's existence in leading edge node itself would have been in question. Under Pat internal nodes were to be the vast majority with target of minimizing external nodes to only when required. PTL is majorly internal nodes. NVL-H compute tile(huge volume) was likely moved to external N2/P from 18A-P under Interim Product CEO after Frank Yeary and Board's December mess. The target was to only use N2/P for 8+16 compute Tiles and a few more things, all the major volume was largely internal. This was to be maximized further while transitioning to 14A in future but this changed somewhat under the next CEOs.

Intel Outsources 30% of Wafer Production to TSMC - SmBom

Intel shifts to a multi-foundry model, outsourcing 30% of wafer production to TSMC while balancing manufacturing self-sufficiency with product competitiveness.

www.smbom.com

Another bad decision Pat made was giving IFS "unlimited budgets" when actually setting a clear boundary and limits are what forces the team to be more efficient. Of course if you truly have unlimited budgets, you may brute force the win, but no one has that, certainly not Intel, not even in 2016.

That "unlimited budget" thing is somewhat overrated. R&D was increased, the test wafers were increased many folds over which was required. They are not as much behind in node which once they were. The foundry balance-sheet is/was also dragged down by need for new equipments/tools, fabs to catch up and also Intel 10nm/7 not being anywhere near margin-friendly as previous long nodes like 14nm was.
The foundry approach did change. To be accepting of TowerSemi deal, to collaborate with UMC for "12nm" instead of going solo so that they can learn and the targets of being no. 2 foundry by 2030 behind TSMC were huge change in attitude from telling the equipment vendors "drop it on the shipping dock and we'll take care of it from there."

 

[Joe NYC]

NVL-H compute tile(huge volume) was likely moved to external N2/P from 18A-P under Interim Product CEO after Frank Yeary and Board's December mess. The target was to only use N2/P for 8+16 compute Tiles and a few more things, all the major volume was largely internal.

I think it may have been Michelle Johnston Holthaus' doing (and undoing).