Intel Beast Lake: 10 performance cores and huge clocks

erek

erek

[H]F Junkie
Joined
Dec 19, 2005
Messages
10,894
Intel Beast Lake

"Beast mode?
Even if Intel is mulling over such a Beast Lake range, as MLID contends, we should remember that it's only experimenting with the idea at the moment, so there's always the possibility that this will all come to nothing and end up scrapped.

MLID also mentioned Panther Lake, the next-gen after Arrow Lake, and how this is supposedly now using a new architecture for the performance cores (Cougar Cove, which may mean it's renamed to Cougar Lake, in fact).

Whatever the case here, we can expect a big performance uplift compared to Arrow Lake - and the latter is already promising seriously major gains itself.

MLID reckons Panther (or Cougar) Lake will offer a 30% to 40% increase for single-core performance, and 15% to 20% for multi-core, in comparison to Arrow Lake. That's with the flagship, which is likely to be an 8+32 (performance/efficiency) configuration processor in both cases for these generations (although Arrow Lake may yet be 8+16 at the top-end).

Intel wants to get Panther Lake to market in good time, we're told, and the schedule calls for a late 2025 release, so a year after Arrow Lake which is due late in 2024. Panther Lake will use the same socket as Arrow Lake, so it won't require a motherboard upgrade.

Overall, Intel's strategy for desktop CPUs is to have these massive gains in future generations down the road - Arrow Lake, Panther/Cougar Lake, and maybe Beast Lake - while settling for treading water in the near term with a simple Raptor Lake refresh which won't be very exciting later this year (Q3 in theory)."

1685841476979.png


Read more: https://www.tweaktown.com/news/9170...-huge-clocks-to-beat-amds-x3d-cpus/index.html
 
OutOfPhase said:
I do.

It's a great way to increase parallelism for bursty loads, while not making something which breaks the bank. Also - handy to toss background junk on cores which don't need to be ultra-fast.
Click to expand...
Pretty much anything I do that needs anything beyond a few cores will peg them all for some amount of time.
The E cores are a waste of space and $$$ that could have been used for more powerful cores in many use cases where a more powerful cpu is useful.
 
stormy1 said:
Pretty much anything I do that needs anything beyond a few cores will peg them all for some amount of time.
The E cores are a waste of space and $$$ that could have been used for more po werful cores in many use cases where a more powerful cpu is useful.
Click to expand...
There is an interesting trade-off in die space versus computation, especially for wide workloads.
 
PC-Only said:
Not if you look at their flagship X3D part, which uses slightly faster X3D cores than the 7800X3D, but then simultaneously dumps a bunch of non-X3D trash on the other side, that their firmware/drivers cannot even properly assign to different tasks. Even worse than the e-cores.
Click to expand...
Why would you be looking at X3D to maximize number of powerful cores? You'd get a regular 7950x...
 
stormy1 said:
Pretty much anything I do that needs anything beyond a few cores will peg them all for some amount of time.
The E cores are a waste of space and $$$ that could have been used for more powerful cores in many use cases where a more powerful cpu is useful.
Click to expand...
Apparently for 1 more powerful core for each 4 E-core + 4mb of L2 you remove, if for something 4-ecore beat a p-core.

We could probably find test that look at 2P+4e vs 3P to have some idea, but e-cores should win for the very parallel affair (which would tend to be the workload to reach your cpu after the first 12 threads)

https://www.pugetsystems.com/labs/a...-cores-compare-2289/#Performance-Core_Scaling

2P+8E is the same die size than 4P and it has a 12,844 R23 cinebench vs 10,258 for the 4P core.
 
  • Like
Reactions: erek
like this
LukeTbk said:
Apparently for 1 more powerful core for each 4 E-core + 4mb of L2 you remove, if for something 4-ecore beat a p-core.

We could probably find test that look at 2P+4e vs 3P to have some idea, but e-cores should win for the very parallel affair (which would tend to be the workload to reach your cpu after the first 12 threads)

https://www.pugetsystems.com/labs/a...-cores-compare-2289/#Performance-Core_Scaling

2P+8E is the same die size than 4P and it has a 12,844 R23 cinebench vs 10,258 for the 4P core.
Click to expand...
CPU video encoding and rar or zip tests would be more aligned with my use case.
 
stormy1 said:
Who wants E cores? Give me 16 to 32 P cores and 0 E cores.
Click to expand...

I think the issue is Intel can't keep it's power consumption/temps under control though with more P cores. A 10P + 24E configuration or 12P + 16E configuration would probably be impossible to cool without a redesigned IHS + an insane liquid cooling system, and a 16P configuration would probably need a nuclear reactor to run. It's not that Intel can't give you more big cores, but rather it would be completely uncompetitive with AMD from both multithreaded workloads and be impossible to keep/maintain P core performance because it would be thermally and/or power throttling so fast.
 
Last edited:
pippenainteasy said:
I think the issue is Intel can't keep it's power consumption/temps under control though with more P cores. A 10P + 24E configuration or 12P + 16E configuration would probably be impossible to cool without a redesigned IHS + an insane liquid cooling system, and a 16P configuration would proably need a nuclear reaction to run. It's not that Intel can't give you more big cores, but rather it would be completely uncompetitive with AMD from both multithreaded workloads and be impossible to keep maintain P core performance because it would be thermally and/or power throttling so fast.
Click to expand...
Pretty much. If Intel could make a 16 P core cpu to directly compete with the 7950X, they would. Instead they combine the P and E core numbers to come up with....dubious core counts on the label.
 
pippenainteasy said:
I think the issue is Intel can't keep it's power consumption/temps under control though with more P cores. A 10P + 24E configuration or 12P + 16E configuration would probably be impossible to cool without a redesigned IHS + an insane liquid cooling system, and a 16P configuration would probably need a nuclear reactor to run. It's not that Intel can't give you more big cores, but rather it would be completely uncompetitive with AMD from both multithreaded workloads and be impossible to keep/maintain P core performance because it would be thermally and/or power throttling so fast.
Click to expand...
I am really unsure about that, it is not easy for me to make it out, but I feel e-core are hotter by mm they take than P-core (i.e. 4core generated more heat than a single pcore).

Wouldn't the CPU die need to be physically bigger to put more core (like they do with the Xeon line) that put up to 56 p core on there.

A single e-core doing a single thread task save you heat versus a p-core doing it, but when you go full load it is more heat from 4 of them than a single P-core on the same space. E-core for when heat become an issue are more space economics than power economy, and I imagine Intel had to go that way to rank up cores because of lesser nodes and not being chiplet-architecture.

For example:
In SPEC, in terms of package power, the P-cores averaged 25.3W in the integer suite and 29.2W in the FP suite, in contrast to respectively 10.7W and 11.5W for the E-cores

Has you see E-core are not really significantly under 25% of a P-core (to be less power hungry), they even use significantly more watt (40-45 instead of 25-30),otherwise I imagine it would be tempting for the Xeon to have a 224 e-core version (or 4 p-core and 2xx e-core)
 
LukeTbk said:
Assuming they get the scheduler under good control, is there a particular reason why ?
Click to expand...
AMD offloads scheduling duties to the Windows team. If anyone is expecting that to work well until second or even third generation for AMD, I think they'll be in for a surprise.
 
AMD is doing big little on midrange desktops, laptops. High end is staying 16c/32t for now (at least thats the last I heard). With the expected big jump in performance with Zen5, this shouldn't be an issue.

Both Intel and AMD are cooking with good stuff lately, great time for CPUs at least. I don't like having to turn on the AC 24/7 so my choice was obvious. I 'd rather not cook like Chernobyl. I'll give up 5% or whatever ultra top end, locking my framerate to the display refresh (144Hz) means my CPUs aren't nearly close to getting taxed.. Doesn't matter to me.

As far as Big Little - idk. Seems like a bit of a crutch but if it works, it works I guess. I prefer full cores, turned off SMT on day 3 for good (or until I actually need it 😃 ).
 
Last edited:
ZeroBarrier said:
AMD offloads scheduling duties to the Windows team. If anyone is expecting that to work well until second or even third generation for AMD, I think they'll be in for a surprise.
Click to expand...
But not being a fan in advance seem to assume even if well executed and work well someone will not like it, it is more profound than about the quirks.
 
LukeTbk said:
A single e-core doing a single thread task save you heat versus a p-core doing it, but when you go full load it is more heat from 4 of them than a single P-core on the same space.
Click to expand...
But the flip side for low-power tasks is you can be doing 4 things at once instead of 1 or 2.
 
LukeTbk said:
But not being a fan in advance seem to assume even if well executed and work well someone will not like it, it is more profound than about the quirks.
Click to expand...
The word fan is derived from the word fanatic. Being a fanatic is not a good thing, but it does aptly describe those who subscribe to the notion that one hardware manufacturer deserves their praise over another.
 
ZeroBarrier said:
The word fan is derived from the word fanatic. Being a fanatic is not a good thing, but it does aptly describe those who subscribe to the notion that one hardware manufacturer deserves their praise over another.
Click to expand...

He's not talking about Intel or AMD, he's not a fan of asymmetrical architectures.
 
Axman said:
He's not talking about Intel or AMD, he's not a fan of asymmetrical architectures.
Click to expand...
I didn't reply stating intel nor AMD either. He's saying that not being a fan assumes that you won't like what they put out even if it's good. I'm saying being a blind fan is actually a bad thing, symmetrical or asymmetrical architectures be damned.
 
  • Like
Reactions: Axman
like this
Gotcha. I see it as simply an inevitability.
 
Story isn't believable. Intel reusing a socket? Pshaw! (75% kidding)
 
pippenainteasy said:
I think the issue is Intel can't keep it's power consumption/temps under control though with more P cores. A 10P + 24E configuration or 12P + 16E configuration would probably be impossible to cool without a redesigned IHS + an insane liquid cooling system, and a 16P configuration would probably need a nuclear reactor to run. It's not that Intel can't give you more big cores, but rather it would be completely uncompetitive with AMD from both multithreaded workloads and be impossible to keep/maintain P core performance because it would be thermally and/or power throttling so fast.
Click to expand...
It’s not just that, it has a lot to do with the California and EU regulations for OEM systems and idle power usage, the E cores when running in sleep mode with wake on Lan and other common business features enabled sip power. You can achieve similar numbers from AMD by under-volting and using high end PSU’s but the Intel ones will do it on a cheap ass 12VO Sparkle Power with anything from the gen 12 and 13 lineup stock.
The Intel Thread Director also does a very good job at keeping all the background tasks that the network does relegated to those E cores. AV, VPN, One Drive, Creative Cloud, Teams, and a dozen other minor things all take those E cores like a champ and it keeps the P cores free for the actual work. You could get the same full CPU load performance from say 8p cores instead of a 6+8 setup but those E cores handling that but during the day to day when it’s probably only 20% at best there is a much nicer feel with all those E cores. Nothing lags, there are no delays, and the lack of SMT or HT in the E cores leads to way fewer crashes in Office background tasks.
AMD is going to need to make the same E core concessions at some point soon or they are giving up the business OEM desktop market they have been trying so hard for over the past 3 years.
Though AMD can likely have an easier go of it, one CCX of all P cores 8/8 and a second of all E cores 16. For a total of 32 threads, though it would hopefully means AMD can have their own version of Thread Director by then, relying on Xbox Game Bar for their thread priority is kind of a cop out.
 
Intel can fit 8 skylake-IPC level cores in the same die space as 2 powerful HT cores. Those 8 E cores get more parallel work done than 2 P cores in nearly every single workload, (though I'm sure there are some armchair experts here who will gladly 'uhm actually' me with a specific task they do that anecdotally runs 4x faster on P cores) going from 8+16 to 12+0 makes little to no sense at all, as the only workloads that need more than 8 cores are going to benefit WAY MORE from an additional 16 good enough cores than 4 "balls to the wall" cores. And any task that favours single threaded performance already has 8 of those single threads.
I've said it before about P/E cores and I'll say it again

If it's stupid, but it works: it's not stupid.
 
You must log in or register to reply here.
Back
Top