Hacker News Reader

Five years? Make that ten; the progress since Sandy Bridge has been incredibly slow. If you take a Sandy Bridge i5 or i7 CPU and turn off the mitigations, it's still within spitting distance of the current i5/i7 models. The only reason I changed out my old i5-2500K two years ago was that the motherboard had finally failed. It was holding up fine with any software I was running, including modern games from that time such as Doom 2016.

Intel really has got to be in an internal panic. The last time AMD pushed Intel this hard, Intel went to extraordinary tactics to push AMD out of the market. I wouldn't be surprised if pushing their workers well beyond safety limits was in the current playbook.

That's a good point; past a certain point I think it's less about the single threaded performance and more about interconnect constraints— slower RAM speeds, slower PCIe lanes, slower SATA, USB, etc.

There's a funny series of LTT videos from this past fall where they bring up a 2008 Skulltrail mobo and try running some modern games on it, with and without overclocking:

https://www.youtube.com/watch?v=wNo7qoLRtkQ

https://www.youtube.com/watch?v=Hl7Dx895ND4

Intel really made their own bed on this one. Anyone with a few grad classes and some interest could tell you what changes to make at each generation to capture more performance. Intel consistently under-delivered on those fronts. It was not because they did not have the technical knowhow, because that part is easy. The people making decisions at Intel were not interested in giving customers the same or better value for their money without competition. Competition came and their pants were still down. Intel's pants continue to stay down since it takes so long to pull them up. They got greedy and they're paying for it.

Memory bandwidth is a big limitation with the older i5/i7s, at least when trying to pair them with anything beyond say a gtx 970.

I found an ITX board for that chip from Asus or Asrock...it's still living as extra workstation for the office!

Userbenchmark is useless, it can barely compare between Intel microarchs. Best it can do is """help""" compare two processors from the same family.

The last big microarch jump for intel was Skylake (6th gen) and then 7, 8, 9 and some of 10th gen (but not all, but they needed "10th gen" for marketing reasons) are just incremental improvemnts over it. So Skylake was about a 15-20% increase in performance (real) but took a few shortcuts (Meltdown and Spectre). From then on, you had a 5-10% performance improvement per generation (let's round it at 7%). That's not bad per se, but it was done out of comfort. The "cursed" 10nm node which never came is also part of the reason why Intel stalled, they insisted on "tick-tock" and were unwilling (or unable?) to deliver a new microarch over 14nm+++, the super refined node (and the extra refinement allowed them to pull the clock higher on each new gen, now alowing those sweet close-to and with this new weird turbo [that seems more marketing than substance] surpass the 5Ghz barrier).

And that's how you come to Intel's situation on 2020: no 10nm node, no real new microarch in the last 5 years.

Edit:

As for why we never saw something like this... I think we never had a situation before where one of the competitors had such an advantage (the microarchs before Zen on AMDs part were bad, and that hit them hard for a long time) that they could pull this duckish moves for so long. For a long time, Intel chose not to improve their offering, because they were so far ahead of competition and they could keep milking the Whatever-Lake microarch family. They'd blame on the difficulty of extracting more IPC, Moores Law coming to an end (or not coming to an end, depending on what marketing was trying to sell).

In the end it came to AMD+TSMC offering superb processors for Intel to slash the prices in half, and then still being crushed performance wise, in desktop, server and now mobile too. And all Intel has to show is AVX512, which after having it gatekeeped for years in the server market, they've released on consumer processors - but hey, nobody's using it because there was no support!

> And that's how you come to Intel's situation on 2020: no 10nm node, no real new microarch in the last 5 years.

Chips fabbed on 10 nm have been shipping in volume since last summer. The Sunny Cove CPU cores found in Ice Lake chips, fabbed on the 10+ node, average roughly around 16 % higher performance vs. Skylake, clock-for-clock, before accounting for improvements in memory bandwidth. Tiger Lake (fabbed on 10++) will be shipping this year, alongside with Ice Lake Xeons.

That 10nm node (10nm+ as they call it) is not as mature as 14nm+++ and they can't reach the same clock speeds (or yields) and that's why they have to resort to comparing clock-for-clock against their previous generations.

Extract from Anandtech[0] comparing Comet Lake vs Ice Lake:

    Some of these tests rely heavily on turbo, such as the PCMark tests, and so the Comet Lake i7-10710U can hit 4.7 GHz on the latest variant of Skylake, while the Ice Lake i7-1065G7, despite its higher IPC difference, can only do 3.9 GHz. This means in a lot of bursty workloads (which a lot of business workloads are), the Comet Lake wins and we see that play out.

Again, that means that some chips aren't produced in 10nm, because if they were, they'd end up being straight up slower, despite having an 18% IPC improvement.

As I said before: I really hope Intel sorts this mess up and comes along with something that can compete against AMDs offering. Otherwise we might see a turn of the tables, with AMD eventually stalling progress, just because they aren't pushed hard enough by the competition.

[0] https://www.anandtech.com/show/15385/intels-confusing-messag...

Are they just going to skip to 7nm at this point?

Intel's 7nm should be comparable to TSMC's 5nm (actually Inte'ls original 10nm, the one that failed spectacularly, would've been better than TSMC's 7nm and 7nmP in terms of transistor density see [0]). And it was started a long ago (like new nodes usually are) so it should be coming along somewhere around the same time maybe? Intel hasn't said much about its future nodes, but there's some information and a few marketing slide around[1].

They seem to have learned from their 10nm mistake, so they're apparently planning to have + and ++ versions of the following nodes, that could handle production of a new microarch if the next node isn't ready, but only time will tell how that goes along.

[0] https://en.wikipedia.org/wiki/7_nm_process#7_nm_process_node...

[1] https://www.anandtech.com/show/15217/intels-manufacturing-ro...

Userbenchmark is totally useless because too many users run benchmark in very varying environments. Better to just watch proper review sites like AnandTech, GamersNexus, Phoronix.

Yup. Same boat. I have an i7 3770k that I purchased in 2012 with the intention of overclocking it. It turned out to be a great chip that was stable at 4.8ghz. I use this computer mainly for gaming.

Last month, I started getting some kernel_trap BSODs so I reduced voltage a bit and also the clock speed down to 4.7ghz.

I don't see a reason for me to upgrade my CPU for another few years unless it burns out or something. The only component I've upgraded is my GPU, from a GTX 670 to a 980 TI.

The newer chips may be great for specific needs, but for gaming and light development work I don't see any reason why I need to upgrade. Sure, a quad core from 2012 isn't the hottest thing anymore, but along with my 980TI it runs all my games at 2k without any issues.

It is also much greener, to reuse your old hardware, especially if it is Sandy Bridge and later (because 1st gen and earlier hardware consumed lots of energy, even in idle; Sandy dropped idle consumption from 50 to 5 watts).

AMD users and overclocking has always gone hand and hand 4 years before the 4770 came out i had a FX8120@5.1Ghz that's faster benchmarks than said 4770 4 years prior and I daily that chip @ 5.1 4.9 for 5 years. had a 9590 that did 5.2 FX was amazing in all the work I did and fraction of cost. In it's idle time these cpus pushed me to the top of the F@H charts with BIGadv work units.

That's because chip makers have been focusing on energy efficiency far more than chip speed.

Yet the TDPs of desktop processors are around 100W, just like ten years ago. And apparently current Intel CPUs Turbo Boost themselves to 200+W!