News Intel 12th-Gen Alder Lake Release Date, Benchmarks, Specifications, and All We Know

Page 3 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
The flexibility that AMD has with their chiplet based designs and also their experience with chiplets could mean that AMD can respond quickly and effectively to Intel, or they might simply wait and see what the end users think and for the software side of things to catch up.

Infinity fabric relies on the memory subsystem in AMD's design. Intel's die-to-die interconnects and other on-die interconnects do not. Like it or not (AMD fan here), the Intel design is faster (lower latency on-die) and offers a greater amount of throughput on-die.

AMD needs to improve memory controller bandwidth capabilities to keep disparate chiplets and infinity fabric competitive. Additional (stacked) cache will dampen the blow but is not viable long-term.
 
for those confused why we need little cores.
i think little cores can be used for antivirus scan, email notifications, updates, downloader, monitoring
 
for those confused why we need little cores.
i think little cores can be used for antivirus scan, email notifications, updates, downloader, monitoring

Yes, and more. ITD looks at processor thread workloads that consume more or less power and distributes per defined coding to P or E cores or both . A passive anti-virus scan might go to just E cores while a full scan could go to P cores or P + E cores (depending on the software in use).

P and E can work simultaneously on the same threaded workload. Threads can move from P to E or visa versa, on the fly, depending on scheduling, thread count and availability. E cores don't just sit around specifically for low-power tasks.
 
Last edited:
The "little" Gracemont cores perform better than Skylake per clock, and Skylake cores are used in Cometlake chips.

So these aren't to save power, these are there to maximize multi-threaded performance within the same power and die area budget.

Based on leaks they are 7-10% faster per clock. So the Icelake and Tigerlake cores are only about 5% faster than these so-called little cores.

Gracemont improves on the dual 3-wide cluster execution by improving coordination between the two clusters and reducing the chances it acts as a 3-wide engine.

Compared to "big" Intel cores such as Skylake and the recent Golden Cove, it doesn't focus on frequency meaning it has less pipeline stages. Testing indicates the uop caches used in their cores don't have 80% hit rate as Intel suggests. Perhaps that number was true in the Sandy Bridge days. The actual hit rate is close to 50-60%. Meaning Gracemont is an effective 5-15% for the same architecture per clock, or they can reduce architectural features by the same amount but achieve the same per clock performance.
 
Based on leaks they are 7-10% faster per clock. So the Icelake and Tigerlake cores are only about 5% faster than these so-called little cores.
We (people in general) call them small because that's what we are used to and it's just easier.
Intel calls it big.bigger for all the reasons you already went through.
Or now efficiency and performance because they probably realized that people are stuck with big.little in their heads and aren't going to change.
https://www.tomshardware.com/news/intel-alder-lake-specifications-price-benchmarks-release-date
 
The top "12th" gen being cheaper than the 5950X is a bit suspect... Intel has never been shy to price the top accordingly, so I wonder what is going on there. And no, Intel is not going to be the "cheaper alternative", lol. Not with all the fanfare they're surrounding Alder Lake with.

Curious.

Regards.
 
The top "12th" gen being cheaper than the 5950X is a bit suspect... Intel has never been shy to price the top accordingly, so I wonder what is going on there. And no, Intel is not going to be the "cheaper alternative", lol. Not with all the fanfare they're surrounding Alder Lake with.

Curious.

Regards.
Intel also always tried extremely hard to keep each SKU priced the same or as close as possible over generations so that huge customers could just buy the same amount of units each time without much fuzz.

If the top SKU is going up in price it's going to be a new SKU, maybe even tier.
Maybe the 12900k will be without e cores otherwise it can't be much more expensive than the 11900k. If they go with a higher price it's gonna be an i10 maybe.
Going from $550 to $800 plus on the same SKU is impossible.
 
Intel also always tried extremely hard to keep each SKU priced the same or as close as possible over generations so that huge customers could just buy the same amount of units each time without much fuzz.

If the top SKU is going up in price it's going to be a new SKU, maybe even tier.
Maybe the 12900k will be without e cores otherwise it can't be much more expensive than the 11900k. If they go with a higher price it's gonna be an i10 maybe.
Going from $550 to $800 plus on the same SKU is impossible.
Hm... Not sure that's totally true, but then again Intel is "under new management", I guess...

Let's see how it is, but I still find it a bit curious.

Regards.
 
So these aren't to save power, these are there to maximize multi-threaded performance within the same power and die area budget.
Skylake on more mature 10nm++/7 should be a fair amount more power-efficient than it was on 14nm+++, especially if Intel carved out or re-engineered some of its more power-hungry lower-yield features to reduce die size and power.
 
The skus, ~pricing and core counts are on page one of the article. 12900k is 8P / 8E.

https://www.tomshardware.com/news/intel-alder-lake-specifications-price-benchmarks-release-date
*Intel has not officially confirmed these configurations or prices. Not all models may come to market.


Retailers have already posted a slew of listings with specifics, but be aware that this could merely be placeholder pricing. However, these price points do generally align with our expectations.
 
I've been rockin' an i7-4790K for over 5 years and the Zen3 processors were a big enticement to finally upgrade. Trouble was I couldn't find one at MSRP, or even close. So, with Comet Lake pricing being reduced, I opted for an i9-10850 on a Z490 MB. I am really happy with the results!

Wonder what will be available in 5 more years? By then I'll be 80 years old and might not care. 🆒

John
I finally upgraded my 3770k to the i9-10850 on a Z590 MB. If i didn't get the 2070 Super I probably would have stuck with the 3770k, but the new card was bottle-necked by the CPU.

The 3770k lasted me about 10 years, I suppose this 10850k will get me to almost 70 years old. ; )
 
Although I am not interested buying an Intel CPU anytime soon, I am actually excited about this CPU release. I'm very curious how many of Intel's performance claims/benchmarks will turn out to be "a stretch" and "benchmark-cherry-picking". 19% of IPC gain on average? When was the last time Intel's IPC improvement claim turned out to be true once tested by independent reviewers? It must have been a while, coz I don't even remember when that was the case.

c' mon, Intel, release this thing already. I'm dying to find out the truth.
 

Fair enough.. As is almost always the case, release details appear and cement over time. 8 days from pre-order and 16 days from general availability I find it hard to fathom, given numerous previous generation releases from blue and red, a significant departure from what is widely reported as the known skus and counts, save for lower end parts.

Additionally difficult to fathom the top-end sku not incorporating both core types given publicly available content and the architecture goals. Developers and board partners have access to the specifics and Intel wants you and I to know what they are (by now).
 
Last edited:
The recent news about older game DRM perhaps not recognizing the CPU is a little disturbing. I'm not planning on upgrading, but just a weird thing. Sign me up for 13th gen or Ryzen 7000? Want DDR5 to be a little more mature before jumping on that bandwagon.
 
The recent news about older game DRM perhaps not recognizing the CPU is a little disturbing. I'm not planning on upgrading, but just a weird thing. Sign me up for 13th gen or Ryzen 7000? Want DDR5 to be a little more mature before jumping on that bandwagon.
Even more of a reason to move away from some of these DRM software going forward.
 
Skylake on more mature 10nm++/7 should be a fair amount more power-efficient than it was on 14nm+++, especially if Intel carved out or re-engineered some of its more power-hungry lower-yield features to reduce die size and power.

These have NOTHING to do with Skylake. These Gracemont cores are direct descendents to the Atom line. The architecture is very very different. It's part of the reason why it's so much more efficient and smaller.

The Sunny/Cypress/Golden Cove cores are in common with Skylake. Yes, a much evolved architecture, but still has the Skylake base, you could say.
 
These have NOTHING to do with Skylake. These Gracemont cores are direct descendents to the Atom line. The architecture is very very different.
The fundamental pipeline hasn't changed much since Nehalem. The only major differences between generations are how wide and deep everything gets along with instruction set extensions.

Atom started with a lot of major stuff like out-of-order execution and branch prediction stripped out, which caused it to be grossly under-powered. All of it got added back in over time and now Gracemont looks very much like a souped-up Skylake.
 
Atom started with a lot of major stuff like out-of-order execution and branch prediction stripped out, which caused it to be grossly under-powered. All of it got added back in over time and now Gracemont looks very much like a souped-up Skylake.

Yea then I doubt you know enough to say anything about it. According to your (uninformed)logic then Ryzen is based on Bulldozer.

Tremont/Gracemont has the novel clustered decode scheme, while Skylake follows the 4+1+1+1 complex+simple decoder scheme, which hasn't changed since the original Pentium.

Skylake has the uop cache which existed since Sandy Bridge.

Tremont/Gracemont uses distributed scheduling stations while Skylake uses unified ones.

They are very different. The way the hardware blocks are laid out are also very different. You should actually read about them rather than randomly speculating.

This is a good start: https://www.anandtech.com/show/2493

The major advancements from then are:
-Silvermont
-Goldmont
-Goldmont Plus
-Tremont
-Gracemont
 
It's a good thing AMD did not jump on the DDR5-wagon as the first one, there will be many sad faces when they see pricing and availability of faster DDR5-Modules.
Sitting this one out is not so dumb as it sounds, meanwhile Intel buyers will have to deep dive into their pockets. Till AMD is on, supply/availability of ddr5 modules will be much higher and prices will be lower most likely. Being first is not always a benefit, unless money is no object. AMD was always about getting the most return of the market, and ddr5 was certainly not favorable, not yet that is. This is in line with their general strategy of neglecting supply of DGs and cheaper CCGs, while maximizing possible supply of DCG.
 
There will be plenty of mobos with ddr4 support and probably some that will support both, either or.
To support both DDR4 and DDR5 on one board, you need to accommodate two rather quite different wiring stub arrangements for the two slots on each channel which can only ever have one working DIMM each. If such motherboards ever get created, I'd expect them to have severely lacking memory clock support for at least one standard if not both. Having one slot work from 12V and the other from 1.2V may also pose somewhat of a challenge with power distribution.

I'm not particularly optimistic about mixed standard boards. They could be great for A-B direct comparisons (to the extent that mixed design compromises don't get in the way too much) and people desperate to (re)use DDR4 until an eventual switch to DDR5. For most people though, committing to only one standard will make the most sense.
 
Sitting this one out is not so dumb as it sounds,
Sitting what out? Not sure what your point is. DDR5 just came to market. AMD isn't releasing anything new until next year. Whatever is based on Zen 4 will likely support DDR5 (and maybe DDR4 too) with new boards and a new socket - they certainly aren't going to ignore DDR5. AL supports both DDR5 and DDR4 with new boards and a new socket - there's no requirement to use DDR5, but the option is there. More options are good, not bad!
 
  • Like
Reactions: bolweval
Sitting what out? Not sure what your point is. DDR5 just came to market. AMD isn't releasing anything new until next year. Whatever is based on Zen 4 will likely support DDR5 (and maybe DDR4 too) with new boards and a new socket - they certainly aren't going to ignore DDR5. AL supports both DDR5 and DDR4 with new boards and a new socket - there's no requirement to use DDR5, but the option is there. More options are good, not bad!
That will be determined by the performance gap measured via independent reviewers. This being said, they lift the embargo at launch time (I can't honestly remember if this is Intel MO or not), so I can say that's fishy and given all the asterisks on the internal benchmarks they showed, the gap I'm talking about is not going to be big enough to command a full fledged upgrade; or at least, not at first glance. Better to wait for reviews for sure.

So, given the above, keep in mind you'll need:
  1. New motherboard.
  2. New CPU.
  3. New RAM (unless you go for a DDR4 motherboard leaving performance on the table, according to Intel).
  4. New cooler (unless you manage to secure the upgrade or it's actually compatible). Potentially a beefy one.

Each of those items it kind of high in terms of price right now and it wouldn't be a trivial expense for most people. If someone is going to aim for an i5, then even less. Better wait for the B-boards and re-evaluate the situation then, even if the CPUs are actually good.

At least, that's my take on things.

Regards.