techspot Intel Core Ultra 9 285K Review

[Sir Beregond]

1 hour ago, Mr. Fox said:

 

We agree on most of this. I probably dislike it the most just because I am an old school overclocker that doesn't give a hoot about "efficiency" or reducing power consumption, and while I enjoy gaming now and then it's not a primary concern for me. Had the new Core Ultra line not had issues with gaming performance being less than 13th and 14th Gen i9 CPUs there would likely be no controversy or complaining. It seems to mostly revolve around gaming performance being less than other factors. And, it's not BAD performance per se; it's just not better by much, or slightly less performant, depending on the game title. Lots of media-fueled drama and melodrama going on as it relates to gaming. So, to me that is just as silly or niche as my unhappiness about how this tiled crap screws up overclocking. I'm just part of a smaller niche than gamers. It looks like Core Ultra processors actually bring more to the table for productivity than anything else does and I really like a lot of the feature-rich aspects with Z890.

I totally agree that for a system whose primary use case is benching, overclocking, tweaking, etc, that power consumption / efficiency doesn't matter at all, and in that sense, I know tons of folks that fit in that camp who have had a lot of fun with 13th and 14th gen Intel. For my daily rig that's 4k gaming and I value quiet performance with the watercooling, I went AMD and it works fine for my use case. I still did a lot of tweaking for the RAM timings which need significant gains in latency, as well as tweaking both the PBO boosting behavior and all-core OC's for the chip. My motherboard has a feature to dynamically shift between all-core OC and PBO single-core max boosts based on parameters set for load. So it's not all bad. But for max overclocking and tweaking fun? Got the bench and some Intel systems for that.

 

Quote

What irks me the most about it is the elimination of hyperthreading, reduction in thread count and hybrid cocktail of P- and E-cores. They need to do one or the other. Either give me more P-cores that are hyperthreaded and ditch the E-cores, or ditch the P-cores and give me a ton of E-cores to make up the loss in thread count. I don't care what kind of core it is as long as it performs well and overclocks nicely. It would be better to do one or the other.

Rumor mill suggests Intel is working their way back to a single core and dropping this hybrid P + E core approach. Guess we'll see if that actually happens. Skymont E-cores are supposed to be about the IPC of Golden Cove (Alder Lake P-cores) if I recall? I wonder why Intel didn't drop an 8P + 32E considering the loss of hyper-threading.

 

Quote

Most of the issues with the gaming performance are Windows issues, not actual hardware issues.

 

Yeah I really hope Linux gaming continues to gain ground with the Steam Deck and other efforts. I am really reaching the point I want to just be done with Windows altogether.

[Mr. Fox]

1 hour ago, Sir Beregond said:

Rumor mill suggests Intel is working their way back to a single core and dropping this hybrid P + E core approach. Guess we'll see if that actually happens. Skymont E-cores are supposed to be about the IPC of Golden Cove (Alder Lake P-cores) if I recall? I wonder why Intel didn't drop an 8P + 32E considering the loss of hyper-threading.

Yes, I find it puzzling as well. It seems illogical that they did away with hyperthreading, and even more illogical that they didn't add more cores to make up for the loss. To some degree the Core Ultra launch may have some social experimentation aspects associated with it. They may have deliberately released a CPU that would predictably get rejected by enthusiasts to test the waters and see what the appetite of the general public is for performance downgrades on a platform with otherwise very desirable features. If there is an element of that involved and they are looking for public response as a measurment, then the more hate and discontent they experience the better off we will all be. 

1 hour ago, Sir Beregond said:

Yeah I really hope Linux gaming continues to gain ground with the Steam Deck and other efforts. I am really reaching the point I want to just be done with Windows altogether.

Linux is so much better than Windows, and it is has gained tremendous ground on gaming. Most of the Windows-only titles I have experimented with on Steam function well, if not flawlessly, using Proton Experimental. No Steam Deck is needed. Works just as well on ordinary PC hardware. All you have to do is apply the setting in Steam client and download the games to your Linux machine. It comes as no surprise, but those that are least likely to work are game titles published my Micro$oft Studios. Some of those also don't work if you don't have a current version of the cancer OS with updates applied. The only reasons I still use Windows are because several products I am required to use every day for work only function on Windows; and, because I enjoy benching. Nothing on HWBOT supports Linux benching. (Sad that they do support smartphone benching, which is ludicrous.) All of my systems are multi-boot and my preference is using Linux. Otherwise, I would have no reason to waste drive space on a bloated and gimmick-driven data-mining Windows 10/11 spyware OS installation.

Edited October 31 by Mr. Fox

[Sir Beregond]

1 hour ago, UltraMega said:

 

For reasons I don't fully remember off the top of my head, nor do I really fully understand, there are limits to how large a single die can be on any given node.

You're thinking of the reticle limit which tl;dr dictates maximum physical limit on die size. There are ways around that though, and to be honest, that hasn't been the limiting factor on monolithic chip design.

 

The real limit is that the bigger the die, the higher the probability of defects. The bigger the die, the fewer chips per wafer. With potential for more defects, the fewer yields per wafer too. This becomes a big problem as we start hitting 3nm, 2nm and beyond where TSMC functionally holds a monopoly  or at least technological dominance in the fab market and can price wafers however they want. TSMC 2N wafer is set to be 2x the cost of 5N and still significantly more than 3N (Source). So the real problem Intel, AMD, and Nvidia are going to hit is designs that minimizes potential for defects, maximizes yields per wafer, so they can get the maximum profit out of each wafer. At the same time they need to still deliver on performance expectations.

 

We have not seen any consumer monolithic dies from Nvidia, AMD, or Intel anywhere close to the reticle limit. The closest was probably Turing with TU102 hitting a die size of 754mm2, but the dies since then have dropped back to just over 600mm2 for the biggest ones, with the step down dies being significantly smaller than that. Shifting back to Intel, the die size of Raptor Lake 14900K is 257mm2 - again a far cry from the physical reticle limit.

 

So in the end, I don't think reticle limit is really a factor here currently and its more probably down to the cost per wafer of the more advanced nodes, and the increasing complexity of architectures necessitating a rethinking of how to manufacture and package, hence chiplets and tiles.

Edited October 31 by Sir Beregond

[UltraMega]

9 minutes ago, Sir Beregond said:

You're thinking of the reticle limit which tl;dr dictates maximum physical limit on die size. There are ways around that though, and to be honest, that hasn't been the limiting factor on monolithic chip design.

 

The real limit is that the bigger the die, the higher the probability of defects. The bigger the die, the fewer chips per wafer. With potential for more defects, the fewer yields per wafer too. This becomes a big problem as we start hitting 3nm, 2nm and beyond where TSMC functionally holds a monopoly  or at least technological dominance in the fab market and can price wafers however they want. TSMC 2N wafer is set to be 2x the cost of 5N and still significantly more than 3N (Source). So the real problem Intel, AMD, and Nvidia are going to hit is designs that minimizes potential for defects, maximizes yields per wafer, so they can get the maximum profit out of each wafer. At the same time they need to still deliver on performance expectations.

 

We have not seen any consumer monolithic dies from Nvidia, AMD, or Intel anywhere close to the reticle limit. The closest was probably Turing with TU102 hitting a die size of 754mm2, but the dies since then have dropped back to just over 600mm2 for the biggest ones, with the step down dies being significantly smaller than that. Shifting back to Intel, the die size of Raptor Lake 14900K is 257mm2 - again a far cry from the physical reticle limit.

 

So in the end, I don't think reticle limit is really a factor here currently and its more probably down to the cost per wafer of the more advanced nodes, and the increasing complexity of architectures necessitating a rethinking of how to manufacture and package, hence chiplets and tiles.

 

The reticle limit gets smaller in corelation with the node size, so it's smaller for current chips than it was with older ones. 

 

It's definitely true that the reticle limit is not the only factor, but one of many compounding factor that are forcing chip makers to go in the chiplet direction.

 

Iirc, one of the bigger factors is power leakage which gets worse the larger the die is. 

[Mr. Fox]

This nice thing is I have enough to keep me going for a long, long time if they don't do something I care about. I will just keep doing what I am doing and using what I already have and say no to chintzy new tech that I do not approve of. 

[Mr. Fox]

12 hours ago, Mr. Fox said:

Probably more in line with not having any to sell, especially the 285K flagship CPU.

Intel Core Ultra 9 285K Sold Out In a Flash In Japan & Out of Stock In US Too, Retailers Confirm Supply Is Low

WCCFTECH.COM

---

Intel Core Ultra 9 285K has seen good initial sales in the US & Japan's DIY market but the total shipped volume to retailers isn't high.

Intel Core Ultra 285K already sold out, retailers expecting re-stock this weekend - NotebookCheck.net News

WWW.NOTEBOOKCHECK.NET

---

Intel's Core Ultra 285K processor, which is the flagship model from the newly released Arrow Lake-S series, has been sold out in Japan as well as the US. As per reports, retailers in Japan said that the processor is...

 

Edited November 1 by Mr. Fox

[Sir Beregond]

10 hours ago, Mr. Fox said:

This nice thing is I have enough to keep me going for a long, long time if they don't do something I care about. I will just keep doing what I am doing and using what I already have and say no to chintzy new tech that I do not approve of. 

Yeah...been watching that thread on OCN and there sure is the crowd that will just buy anything because it's new. 

[Mr. Fox]

On 01/11/2024 at 07:51, Sir Beregond said:

Yeah...been watching that thread on OCN and there sure is the crowd that will just buy anything because it's new. 

 

 

https://www.overclock.net/threads/msi-x870e-x870-meg-mpg-mag-overclocking-discussion-thread.1812293/page-3?post_id=29391228#post-29391228

[Sir Beregond]

1 hour ago, Mr. Fox said:

 

 

https://www.overclock.net/threads/msi-x870e-x870-meg-mpg-mag-overclocking-discussion-thread.1812293/page-3?post_id=29391228#post-29391228

How do you like the MSI board / BIOS so far? Been mostly on ASUS myself with some EVGA sprinkled in for the bench (Z490, Z790).