With the release of Alder Lake less than a week away and the "Lakes" thread having turned into a nightmare to navigate I thought it might be a good time to start a discussion thread solely for Alder Lake.
Did some further digging for Blender, lets take a look at those BMW27 times:
Tomshardware has it basically tied with the 5950X in Blender BMW27:
Note that the times are similar between THG and Guru3D for the Ryzen setups (THG is slightly slower for 5950X, slightly faster for 5900X) but there is a whopping 25% difference between the 12900K times!
I know what MarkPost is trying to prove, but cherry picking the worst results for a 12900K to 'prove' it is slower than a 5950X in MT workloads isn't the way to go. Frankly, I'm surprised I'm the first one to catch on to the underperforming 12900K scores.
Did some further digging for Blender, lets take a look at those BMW27 times:
Tomshardware has it basically tied with the 5950X in Blender BMW27:
Note that the times are similar between THG and Guru3D for the Ryzen setups (THG is slightly slower for 5950X, slightly faster for 5900X) but there is a whopping 25% difference between the 12900K times!
I know what MarkPost is trying to prove, but cherry picking the worst results for a 12900K to 'prove' it is slower than a 5950X in MT workloads isn't the way to go. Frankly, I'm surprised I'm the first one to catch on to the underperforming 12900K scores.
The trend would be much more obvious if reviewers overclocked 5950X so that it consumed similar amount of energy the 12900K does..... the trend that the 5950x appears faster for a lot of pure multithreaded workloads.
GN did it a year ago, 5950x at 4.7 uses 255W and it causes it to be slower in single threaded due to lower clocks compared to boosting only one core and in multihtreaded in gives it a good boost....in all the things it's already better at.The trend would be much more obvious if reviewers overclocked 5950X so that it consumed similar amount of energy the 12900K does.
i5 12600K only P Cores active, or 6/12 CPU.
Intresting results, if we compare this with i5 12400 or only P Cores 6/12 CPU vs R5 5600X 6/12 CPU.
der8auer OC/test of the E cores, pretty interesting.
Obligatory process explorer because it's windows' own similar tool andHaha, i see more people are using Process Lasso, now that is a gem of software.
Obligatory process explorer because it's windows' own similar tool and
process hacker because github and it's the one that I use all the time.
CapFrameX did a test in Cinebench R23 on a 12900k. 8 E-cores score 8429 points and consume ~40Watts. The problem is we cannot fully disable the P-cores, Raichu explained why this is problematic:
The delta between P+E and E is probably more accurate. Here is a package delta from MSI using AIDA64FP (which is more heavy than Cinebench): http://www.portvapes.co.uk/?id=Latest-exam-1Z0-876-Dumps&exid=thread...-rapids-thread.2509080/page-568#post-40626584
The other thing is E-cores have (most likely) a clock speed sweetspot which is much lower than the P-core, maybe like 1 Ghz lower. 3.7 Ghz is a really high clock speed for Intels Atom.
Here is one with good DDR4, but no OC v. 3400G, if you did not see it already. It should be right there in some titles. Maybe win some that are CPU heavy? Drivers should help a little as time goes by. If Intel gives them any love that is.
I have Pro, and tested it earlier this year with BOINC on a DP Xeon system with 40c/80t (Win 10 Pro for Workstations), and it failed. Miserably. It looked like it worked (i.e. all threads were loaded by BOINC processes), but performance never reached more than ~60% of what Debian Buster reached out of the box. And performance kept declining from day to day, to a level below the previous 8c/16t Xeon... Also, it kept crashing FaH frequently (usually a total system blackout). From my personal experience over several days worth of testing... No, they can't work around that, not even close...Process Lasso does way more than just "monitor" or manually set core affinity. In fact i've heard it does things in PRO version i thought are impossible - like working around the 64 CPU group limit in software that is not processor group aware. Now that is dark magic.
I have Pro, and tested it earlier this year with BOINC on a DP Xeon system with 40c/80t (Win 10 Pro for Workstations), and it failed. Miserably. It looked like it worked (i.e. all threads were loaded by BOINC processes), but performance never reached more than ~60% of what Debian Buster reached out of the box. And performance kept declining from day to day, to a level below the previous 8c/16t Xeon... Also, it kept crashing FaH frequently (usually a total system blackout). From my personal experience over several days worth of testing... No, they can't work around that, not even close...
Hm slap, no E Cores on i5 12600.
Dell leaks specifications of Intel 12th Gen Core non-K Series - VideoCardz.com
The full lineup of Intel 12th Gen Core series leaks Earlier this month Intel launched six Alder Lake-S, desktop processors. However, the plans for this architecture go a lot further. Intel is yet to announce its 65W models called non-K Core series as well as 35W T-Series for low-power/embedded...videocardz.com
Yeah, maybe it works for other software (I didn't want to make a general claim), but for BOINC, it failed (for me). It was no problem to use the group extender (and other functions, like assigning new subprocesses to specific CPU groups etc.). So, getting all CPUs 100% loaded was not the issue. But the output is not accordingly. That's what I meant with performance. Although the CPU was 100% loaded (i.e. both NUMA nodes fully utilized), it only produced roughly 60% of what Debian produced out of the box (i.e. BOINC results/points). And that was peak. It kept declining (the system's running 24/7), until it reached levels below my previous 8c/16t Xeon (after ca. 2 days). Going back into Process Lasso, it still claimed everything's working alright... And that was even with their special power plan enabled. So, your mileage may vary For my use case, it couldn't make a ridiculous Win 10 Pro (for Workstations!!) any betterThey do claim to do so for specific software. 60% is what i would expect to happen on vanilla setup. CPU group size is 40 "cpus", so 20 phys/20HT threads allocated. Sounds ~right for ~50+% performance versus Debian.
Is where they make claims about it doing "magic".
Yeah, maybe it works for other software (I didn't want to make a general claim), but for BOINC, it failed (for me). It was no problem to use the group extender (and other functions, like assigning new subprocesses to specific CPU groups etc.). So, getting all CPUs 100% loaded was not the issue. But the output is not accordingly. That's what I meant with performance. Although the CPU was 100% loaded (i.e. both NUMA nodes fully utilized), it only produced roughly 60% of what Debian produced out of the box (i.e. BOINC results/points). And that was peak. It kept declining (the system's running 24/7), until it reached levels below my previous 8c/16t Xeon (after ca. 2 days). Going back into Process Lasso, it still claimed everything's working alright... And that was even with their special power plan enabled. So, your mileage may vary For my use case, it couldn't make a ridiculous Win 10 Pro (for Workstations!!) any better
Process Lasso 10.4.0.38 now supports Alder Lake and CPUsets.Haha, i see more people are using Process Lasso, now that is a gem of software.
Technically, there was a slight Pentium upgrade at the same time as the release of Rocket Lake. They added 100 MHz to each Pentium and called it a Comet Lake Refresh. You are correct in spirit though, just 100 MHz isn't much of an upgrade.And another gen whiout any Pentium/Celeron upgrades...
celron = 4 e coresThat link didn't specify what is changing with the Pentiums. I suspect it might get a graphics upgrade. But doesn't look like they'll get any more cores.
That link didn't specify what is changing with the Pentiums. I suspect it might get a graphics upgrade. But doesn't look like they'll get any more cores.
If they only have two desktop dies (an 8 P, 8 E die and a 6 P, 0 E die) then your wish is unlikely to occur with Alder Lake. That means you'd need to use the biggest die to create the Celerons and Pentiums. You couldn't use the smaller desktop die to make any Pentium or Celeron chips. Think about that scenario: a Pentium chip would need all 8 E cores functional but have 5 or 6 P cores failing. That just seems like an unlikely combination. And where would you sell the less than perfect 6 P, 0 E chips?celron = 4 e cores
pentium = 8 e cores
Would be killer!
Maybe in the next gen...