I'm really not sure how all this applies to Sierra Forest. If the argument is that so many cores would be so strong relative to the competition, Intel would instead prefer to sandbag, then I heartily disagree.
I am not saying they are going to sandbag, just that whatever they have might be enough. AMD isn't using the more area efficient Zen 4c to build many, many more cores than Genoa either. It's "just" 33% more, which when you are talking about being in a ballpark top of the field, it's enough to make a difference.
We don't know enough about Bergamo to know why you'd get it over Genoa. The leaks show similar 300-400W TDP level and it's being called "Density Optimized". Is it cheaper then? Somehow I have a feeling a 350W chip isn't going to be <$1000 cheap. So the benefits of Bergamo compared to Genoa is narrower than we initially expected. It's not a low TDP chip, and it's core count isn't immensely greater.
Maybe Sierra Forest is different and more "density optimized" that way but usually the end result is surprisingly similar. Maybe Sierra Forest will go to 144 cores but Bergamo or whatever it'll compete with will be faster per core, making SRF more of a purpose oriented cloud chip while Bergamo is more in between.
Early leaks for Sierra Forest had a 7000-pin -AP version using what's called "SSHPmont". I assume that means "Super Speed High Performance-mont". Because it's AP it's very plausible it was a version with AVX-512 support, maybe with 1x-512 units like client units or 2 cycle like with AMD which is double Gracemont.
Maybe they have changed this. If you read twitter Retiredengineer replied to witeken saying "you'll be disappointed" when witeken was expecting 448 Crestmont-based cores. And Retiredengineer referred back to that quote, so he knew that way back.
I know with Smartphones some were predicting we'll have 32+ cores very soon, and clearly it stopped as well. Arrowlake was rumored to be 8+32 and with future chips some were thinking 8+40 or even greater, but real Arrowlake is 8+16.
I am saying taming expectations is reasonable because the balance is a bell curve. It's with everything. Power is very simply put P=CV2F, but you aren't seeing 0.6V GPUs in operation. Because below 0.9V the frequency drops non-linearly, which is a nice way of saying it plummets. So rather than 0.6V being 2/3rds the frequency of 0.9V, you get something like 1/3rds the frequency and sure it uses very little power but it's not a useful workload frequency.
And if you lower chip power too much, then the IO and memory is more or less fixed and they start dominating so it's again a bad balance. So in reality the engineers drop it by 100mV, do workload optimizations, cut leakage, and dozen other things.
Even if you could lower whole chip power by simply lowering voltage, you'd have to significantly increase die space. If engineering such things were easy then they could just hire college graduates and be done. But it's not at all. In fact it's one of the most complex things humans can do!
So while the E cores are actually power efficient when you put it in the proper habitat, it also follows a bell curve and in case of desktop chips it's entirely about area density. So if Gracemont goes back into being a N-series chip, it'll again be very, very power efficient, because it's a perfect match for the IO and the frequency balance is good, etc. And I bet the P core chips seem efficient in power for the desktop chips but they might have high leakage and/or worse scaling at lower frequencies so it's a terrible chip for N series platform.
@ashFTW I appreciate that. Just saying I am on the conservative side when it comes to expectations. But that in itself is a contradiction, because computing is absolutely cutting edge. You still have a balance though.
Look at technologies today that are seen as clear advancements over predecessors. LEDs and Solid State Drives for example. Were they that good initially? LEDs were mere indicator lights for dashboards! Solid State drives were either very expensive or the performance(with 2-bit "MLC) completely sucked. Again, it took over a decade! But of course without them the advancement would slow to a near halt. So revolutions break the wall. They aren't super duper good out of the gate.
Facebook
Twitter