4 ms·
Copying my answer [1] from the Alder Lake Preview. 1. Base Power and Turbo Power. Finally a simple term so we dont have to constantly argue about TDP and PL1 /
by ksec 5y ago
Copying my answer [1] from the Alder Lake Preview.
1. Base Power and Turbo Power. Finally a simple term so we dont have to constantly argue about TDP and PL1 / PL2. Which still happens fairly often on HN. And some of us have been ranting about this for almost a decade.
2. + 19% IPC improvement vs 11900K which is basically an Icelake. Or ~10% improvement over TigerLake ( Which was never made available on Desktop )
3. PCI-E 5 and DDR5. Intel went from falling behind in PCI-E offering to leaping ahead of AMD. We should expect PCI-E 5.0 SSD shipping soon after ( 14 - 16GBps )
4. Chipset is on 14nm. No USB 4 or Thunderbolt 4. But integrated WiFi 6E MAC.
5. For those interested, this put Alder Lake, on a Desktop Platform, with Intel 7nm High Performance Node, at roughly 20-25W per Performance Core. With a ( non-verified ) Geekbench Score fairly similar ( ~6% faster ) to M1 HP Core. The Apple M1, on a TSMC 5nm Low Power node, at roughly 5W per core.
6. It will be interesting to see how the new efficiency cores perform. Which is basically a new generation Atom Core. These were previously scheduled for a Graviton like 64 - 128 Core chip on server. Not sure if that is still the case with Intel's chiplet strategy or what they called Tiles.
7. Worth wondering, Alder Lake was originally scheduled on Intel 7nm or what is now called Intel 4/3nm in 2019. What would happen had Intel not been so badly managed? But if that didn't happen, Pat Gelsinger may never be back at Intel.
8. AMD Zen 4, also with DDR5 and PCI-E 5.0 will be coming in late 2022. Depending on Intel's pace of execution, which seems to be getting better and better every time Pat Gelsinger provides an update, AMD may have some tough competition.
[1] https://news.ycombinator.com/item?id=29017221 https://news.ycombinator.com/item?id=29017221
- tagrun 5y agoThere is USB4 but it's limited to higher end motherboards for now.
- ju-st 5y ago> 3. PCI-E 5 and DDR5. Intel went from falling behind in PCI-E offering to leaping ahead of AMD. We should expect PCI-E 5.0 SSD shipping soon after ( 14 - 16GBps ) The x4 lanes usually used for m.2 SSDs are PCIe4.
- liuliu 5y agoAlder Lake is on Intel 7 otherwise known as 10nm process.
- emn13 5y agoThe claim I've heard is that intel 7 (previously 10nm) is approximately as good as tsmc's 7nm (but which 7nm?). It sounds like most reviewers seem willing to accept that it's at least a good enough match that you're unlikely to do better. I mean, those literal nm measurements had no reasonably obvious basis in measurement for either intel or TSMC even before the intel rebrand, so... It'd be nice if they advertised with a few real benchmarks (e.g. logic density + sram density, some kind of clockspeed metric, some kind of powerleakage metric)... but as is, it's just a branch anyhow. Whether it "is" a 10nm process depends on what your definition of "is" is ;-). It's certainly not the case that a transistor is 10nm large, not even close.
- liuliu 5y agoDefinitely, no argument here. I meant Intel 7 is the previous Intel's 10nm. It is confusing in my parent thread where the commenter uses Intel 7nm, which is not a thing in any case. Of course, this is probably all intentional from Intel's point of view (letting people say "Intel 7nm" rather than "Intel 7").
- reacharavindh 5y agoBetter single thread performance is always a nice win because it is immediately perceivable on unmodified software. I believe Intel is still counting on using x4 PCIE4 for SSDs. I wonder if the PCI-E 5 x16 makes any difference for the GPUs? DDR5 may bring in nice capacity boosts. However, I read somewhere that at lower capacities it is almost indistinguishable perf wise from high-end DDR4. The key to M1's attraction is the cohesiveness of its implementation. Apple can make MacOS tightly knit to the heterogenous core types and caches. and essentially gate keep applications to behave optimally. From Intel's announcements, it looks like Intel came up with Thread Director to _advice_ Windows on how to schedule stuff and leave it to MS to implement things. Not sure how that synergy is going to take shape. Perhaps Linux will come out and eat this. if AMD can improve on Zen 3 and come up with Zen 4, they already provide much more PCI-E lanes and ECC on DDR4/DDR5 memory support, with better than intel multi-threading perf. Competition sure is nice!
- uweihfuio 5y agoI think your data is not correct. H-series CPUs from Intel always had 35W-45W TDP. Now, you write here that a single core will consume 20-25W... That will make TDP of 200W when all cores are running. That doesn't make any sense at all.
- spacedcowboy 5y agoI don’t know about the 20-25W figure, but I do know that Intel’s TDP figures are not maximum TDP like Apple’s are. Intel chips are well known for blowing through that TDP under load, whereas the M1* figures are absolute, as far as I have read. In short, 35-45W is a pipe-dream if you’re pushing the chip. I personally doubt it gets near 200W though…
- ksec 5y ago>That doesn't make any sense at all. Intel has always used TDP as normal TDP under baseline usage. They have now introduced a term called Turbo Power or previously known as PL2, when you max out the CPU to better reflect on true TDP. And I should have wrote it better. It was compared to i9-12900K on Desktop which has a Turbo Power of 240W. But on a single core basis, the difference should be minimal. The Mobile CPU will still uses ~20W max assuming information and Geekbench number being equal. We will have to wait for final review to confirm.