anon2 (anon.delete@this.anon.com) on April 7, 2021 5:17 pm wrote:
> juanrga (nomail.delete@this.juanrga.com) on April 7, 2021 7:36 am wrote:
> > anon2 (anon.delete@this.anon.com) on April 6, 2021 5:43 pm wrote:
> > > Wilco (wilco.dijkstra.delete@this.ntlworld.com) on April 6, 2021 5:37 pm wrote:
> > > > anon2 (anon.delete@this.anon.com) on April 6, 2021 3:23 pm wrote:
> > > > > Wilco (wilco.dijkstra.delete@this.ntlworld.com) on April 6, 2021 2:09 pm wrote:
> > > > > > Adrian (a.delete@this.acm.org) on April 6, 2021 12:01 pm wrote:
> > > > > > > anon (anon.delete@this.anon.com) on April 6, 2021 10:48 am wrote:
> > > > > > > > https://www.anandtech.com/show/16594/intel-3rd-gen-xeon-scalable-review
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > > Good:
> > > > > > > The cheaper SKUs are no longer crippled by disabling 1 of the 2 FMA units.
> > > > > > > No clock frequency penalty for AVX-256 and light AVX-512 instructions.
> > > > > > > > For heavy AVX-512 instructions, the clock frequency 80% of normal frequency.
> > > > > > > There are a few SKUs with good price/performance, e.g. Gold 6312U.
> > > > > > >
> > > > > > >
> > > > > > > Bad:
> > > > > > > The cheaper SKUs (Silver & Gold 5000) are crippled by low memory speed.
> > > > > > > Hopefully ARK contains mistakes, because it shows even some
> > > > > > > Gold 6000 SKUs as being crippled by low memory speed.
> > > > > > >
> > > > > > >
> > > > > > > Very bad:
> > > > > > > At the same number of active cores and the same power consumption, the clock frequency is much lower
> > > > > > > than for Epyc 7003, so the performance is much lower for anything that does not use AVX-512 (the Ice Lake
> > > > > > > IPC is a little lower than that of Zen 3, so at lower clock frequencies it does not have any chance).
> > > > > > > Also the maximum single-core turbo frequency is much lower than for Epyc 7003.
> > > > > > >
> > > > > > >
> > > > > > > Because of the previous facts, in their presentation Intel has wisely chosen to not show
> > > > > > > any comparisons with competitors, except the few that could take advantage of AVX-512.
> > > > > >
> > > > > > The Ugly:
> > > > > > Fastest SKU doesn't even beat Graviton 2 on SPECINT.
> > > > > > Fastest 2S SKU is only 27% faster than 1S Ampere Altra despite double the threads,
> > > > > > double the DRAM channels, 2.5-3 times the power and quadruple the list price...
> > > > > >
> > > > > > Not being able to match standard Arm cores is just embarrassing.
> > > > >
> > > > > It would be indeed, although it matches and beats quite well a standard ARM *core* (see single thread
> > > > > numbers), just not a bunch of ARM cores in a chip. The latter is less embarrassing depending on the
> > > > > test, because of more cores fitting in, even AMD gets a run for their money on some of the tests.
> > > >
> > > > Well the 8380 is only 10% faster single-threaded because of the large L3.
> > >
> > > Ahh, back to the good old Wilco! Of course, I should have known, ARM is actually
> > > _faster_ and the other mean nasty companies are actually just cheating.
> > >
> > > > Altra has higher ST perf than
> > > > the 6330 with its smaller L3 (both 8380 and 6330 have the same 1T frequency of 3.4GHz). One can fit around
> > > > 4 to 5 Neoverse N1 cores in one IceLake core, so you would expect a major ST performance advantage from
> > > > such a huge core. Do you think 6.3% ST perf over the 2 year old 8280 counts as great progress?
> > >
> > > Well Intel's process technology has stalled badly for a lot longer than two years so of course it's not
> > > great progress. What is more concerning for the ARM fanclub is that AMD does pretty damn well when the
> > > process technology is equivalent. Strange, there must be a nasty ARM tax slowing them down to the point
> > > where they're not able to take advantage of the huge alleged x86 tax slowing down AMD so much.
> >
> > I am not sure you understand the concept of x86 tax.
>
> I am.
>
> > It is not about "slowing down", it is about size.
>
> Well you certainly don't understand it. "x86 tax" was absolutely about performance.

Well you just confirmed you don't understand the x86 tax. The tax is about how a better ISA like A64 allows using less transistors than x86 to do the same work. So you can achieve the same performance on a smaller size or achieve more performance using the same size, or a combination of both: e.g. ~10% higher performance and ~10% smaller design.

If you read the quote given. TX3 gets a similar performance with a 20 or 30% smaller size.

> > https://www.hpcwire.com/2020/03/17/marvell-talks-up-thunderx3-and-arm-server-roadmap/
> >
> >

Just to give you an idea, in the previous generation,
> > if you look at ThunderX2, compared to AMD or Skylake,
> > for the same process node technology [we get] roughly 20%
> > to 25% smaller die area. That translates into lower
> > power. When we move to 7nm with ThunderX3, our core compared to AMD Rome’s 7nm is roughly 30% smaller.

> >
> >
>
>