Ryzen "Zen4 " to support AVX-512 operations.

https://www.techpowerup.com/279129/amd-zen-4-microarchitecture-to-support-avx-512

If so, that’s the nail in the coffin for Intel. That’s the last metric they’ve been holding onto lol.

Still at least a year out for these products. So good chance the foundries will have caught up with demand by then.

I honestly think it’ll be a lot longer. They haven’t even released Milan yet.

---

Seti@Home classic workunits: 29,492 CPU time: 134,419 hours

Still at least a year out for these products. So good chance the foundries will have caught up with demand by then.

I honestly think it’ll be a lot longer. They haven’t even released Milan yet.

Can only hope so.
12 months for supply to finally catch up with demand (or demand finally drop enough for supply to meet it) are the more optimistic predictions. 18 months seems to be the general consensus, 2+ years the most pessimistic at this point in time.

---

Grant
Darwin NT

Although still under an NDA, Anandtech have purchased & reviewed a Core i7-11700K Rocket Lake CPU that was sold at retail early, and posted their review based on that retail sample.

Yes, there has been a 19% increase in IPC- however that is for floating point work only (which is great for compute work). But for general usage, the IPC increase isn't so great- Single thread integer: +13.0%, Multi-thread integer: +7.3%
The really big problem is power usage- think of Bulldozer (although at least unlike Bulldozer you do get performance with the huge power usage).
This is a 125W TDP chip.
Depending on the application, it can us 120-160W sustained. Running an AVX2 workload, you're looking at 225W sustained. AVX 512 workload, peaks of 295W with CPU temperature peaks of 104°c (using a ThermalRight Ultra Extreme cooler, with a 170 CFM high static pressure fan). If you have a AVX 512 workload, then it wins by a country mile- just make sure you have excellent cooling.


Apart from AVX512, AMD still leads the way- better IPC performance, more threads & cores, and lower power usage.
Unfortunately the only way to get a AMD CPU is to buy a ready made system, and even then they're not cheap. As for buying one to make your own (or for a shop to make it for you), most of us would have to sell body parts to pay the scalpers the asking price for a Ryzen 5000 series CPU.
The fact that a batch of these CPUs was sold at retail prior to the NDA date (and the retail launch date which is the 30th of this month) would indicate that there are plenty to be had. So there's a good chance Intel will sell a ton of these, as it's just not possible for people to get the better product. Hopefully that income will let them get their 7nm node up and running.

---

Grant
Darwin NT

Yes, Intel will be able to claw back some market share from AMD simply because they have product to sell. And they are reducing the MSRP prices from historical levels to try and undercut the traditionally lower AMD cpu prices.

If you want to update your cpu in the current market, you will have better luck on the Intel platform. Or be very lucky with the rare times that actual AMD product is available. The AMD Ryzen 7 5800 seems to be the most available lately. But no signs of the 5900 or 5950.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

AMD Epyc 7003 "Milan" cpu Launch day event scheduled for March 15. Next Monday.
https://www.phoronix.com/scan.php?page=news_item&px=AMD-Next-Week-Milan
https://www.amd.com/en/events/epyc

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

AMD Epyc 7003 "Milan" cpu Launch day event scheduled for March 15. Next Monday.
https://www.phoronix.com/scan.php?page=news_item&px=AMD-Next-Week-Milan
https://www.amd.com/en/events/epyc

This is something i've been looking forward to.
Given that AMD haven't been talking it up at all, and the very few leaks that have made it out from different sources indicate huge improvements over the current series, it's looking like it will be something special. Given AMD's present lead over Intel when it comes to server CPUs, this new release should pretty much obliterate Intel's server offerings.

---

Grant
Darwin NT

Just watched the presentation and took a look at the product stack. This article shows the pricing and relative performance compared to the Rome and Naples generations.

https://www.nextplatform.com/2021/03/15/the-third-time-charm-of-amds-milan-epyc-processors/

I want the new AMD Epyc 7443P listed for $1337 to upgrade my Epyc 7402P.



That would be a great bump in per clock speed and 19% IPC uplift.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

that does look like the sweet spot for price to performance. and the $1337 price is just fun.

When can you actually buy these? I havent seen them for sale anywhere. And Asrock hasn't pushed a BIOS update to support them yet.

---

Seti@Home classic workunits: 29,492 CPU time: 134,419 hours

That or the new 28 core 7453 at $1570 seem to be the sweet spot for me.

I haven't found any sign of them at retail though either.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

I think she went for a Ryzen...

Beautifully hilarious for a first-time cruncher!


Maybe Building Your Own PC is a BAD Idea


For a first timer, very well done!

And a good reminder for those who are not so new to this for just how much we take completely for granted.

Enjoy :-)


Happy fast crunchin'!
Martin


Edit, ps:

And for the killer quote in the comments:

Negasuki
wrote:

The day Linus watches someone else handle hardware and feels what we've all felt for so long when watching Linus handling hardware.
A day of justice has finally arrived.

---

See new freedom: Mageia Linux
Take a look for yourself: Linux Format
The Future is what We all make IT (GPLv3)

Definitely a "cringe-worthy" view. And a great reminder of our collective experience in building PC's from scratch. I used to say a "trained monkey" could put a PC together after purchasing my one and only pre-built, store bought computer back in the Intel 386SX days.

Now I have a lot more respect for the monkey.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

Definitely a "cringe-worthy" view. And a great reminder of our collective experience in building PC's from scratch. I used to say a "trained monkey" could put a PC together after purchasing my one and only pre-built, store bought computer back in the Intel 386SX days.

Now I have a lot more respect for the monkey.

How about a trained Monkey could take a PC apart? ;)

Tom M

---

A proud member of the OFA (Old Farts Association).

Finally getting around to digesting some of the news coming out of Computex.

The AMD announcement of die stacking an extra 192MB of L3 cache on top of the 5900X/5950X cpus and the processors coming out at the end of the year has me wowed.
Showed an actual physical implementation and gaming benchmarks of the Zen 3+ 5900X.

https://www.anandtech.com/show/16725/amd-demonstrates-stacked-vcache-technology-2-tbsec-for-15-gaming

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

Finally getting around to digesting some of the news coming out of Computex.

The AMD announcement of die stacking an extra 192MB of L3 cache on top of the 5900X/5950X cpus and the processors coming out at the end of the year has me wowed.
Showed an actual physical implementation and gaming benchmarks of the Zen 3+ 5900X.

Although as Anandtech mentioned, the performance improvements from such huge caches are only seen in games & compression/decompression workloads.

However with them fitting more & more cores & threads on a single socket, this technology will be needed to help stop the cache sizes from getting smaller on such massively multi-core chips which would cause a significant hit in performance.



The announcement that has me most excited at this stage is the one from TSMC about their continued ramp up of existing and new process nodes.
TSMC Manufacturing Update: N6 to Match N7 Output by EOY, N5 Ramping Faster, Better Yields Than N7

---

Grant
Darwin NT

Finally getting around to digesting some of the news coming out of Computex.

The AMD announcement of die stacking an extra 192MB of L3 cache on top of the 5900X/5950X cpus and the processors coming out at the end of the year has me wowed.
Showed an actual physical implementation and gaming benchmarks of the Zen 3+ 5900X.

Although as Anandtech mentioned, the performance improvements from such huge caches are only seen in games & compression/decompression workloads.

However with them fitting more & more cores & threads on a single socket, this technology will be needed to help stop the cache sizes from getting smaller on such massively multi-core chips which would cause a significant hit in performance.

I ran into an (I think) Cache limitation on my Amd 3950x. It meant I couldn't run more than 6-8 Einstein at Home cpu Gravity Wave tasks without the processing speed taking more than 100% longer. (4-6 hours vs. 12+ hours). I thought at first it was a ram limitation but when I dropped in a 64 GB kit it still slowed down.

Tom M

---

A proud member of the OFA (Old Farts Association).

First I've heard of a L3 cache issue with Einstein. There was a thread discussion about L3 limitations with a Rosetta application last year.
https://boinc.bakerlab.org/rosetta/forum_thread.php?id=13764#93961

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

All AMD Zen 3 processors available at online retail now.

Newegg has the 5950X for example at $50 off MSRP, so only $749.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

We'll see how long that lasts.

Latest word on GPU pricing is that after falling for a while, it's on it's way back up. Availability is still spotty at best, and you need an arm, leg & two or more kidneys to pay for them.
It's so bad, some companies are re-issuing GT 730 cards (what an abomination they were, and that's the good series. The crappy ones were beyond words).
Car manufacturers are further reducing their production output due to still not having enough chips to meet demand, and all foundries are expected to bump up their prices on all new contracts (I think it was TSMC that made the latest announcement).
Even consumer electronics in the way of TVs is being impacted- prices there are expected to climb (although only slightly compared to most other things) due to the squeeze in supply of display driver chips.

And the latest forecast- it's expected to continue till the end of next year, maybe, if we're lucky...

---

Grant
Darwin NT

A sign of how bad things are- GM to halt production at nearly all North American assembly plants next week.

General Motors told the Detroit Free Press that starting next Monday, September 6, it’ll temporarily halt nearly all of its production plants in the US as the lingering Covid-19 pandemic continues to hamper production of semiconductor chips overseas.

---

Grant
Darwin NT

Milan-X officially announced for 1Q 2022 general availability in today's keynote address.

Microsoft Azure cloud instances available right now.

Triple the L3 cache size compared to existing Zen 3 Milan processors.

Up to 60% performance improvement in some applications.

Processor Cores/Threads Base Clock Boost Clock TDP L3 Cache (L3 + 3D V-Cache)
Epyc 7773X 64/128 2.2 GHz 3.5 GHz 280 W 768 MB
Epyc 7573X 32/64 2.8 GHz 3.6 GHz 280 W 768 MB
Epyc 7473X 24/48 2.8 GHz 3.7 GHz 240 W 768 MB
Epyc 7373X 16/32 3.05 GHz 3.8 GHz 240 W 768 MB

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)