AMD Ryzen 2021-2022 roadmap with codenames leak - Van Gogh and Warhol

I´m very curious about the next CPUs from AMD. I hope they continue to deliver.

 

Saying it's going to to perform just about as well as said potato...

 

Van Gogh is a potato cultivar from Netherlands. I don't know if the potato variety was given that name because of the van Gogh painting depicting potato eaters, though. Maybe Hilbert would know, being from Netherlands himself. He has probably seen the painting if he's interested in art.

 

But AM1 were DDR3... (the black sheep lol)

 

It is not that easy they count wrong ! The right way is 2 3.0 ...3.1 3.2 gen 1 3.2 gen 2 .....

 

Maybe a bit late and out of the blue, but witht he number of cores in AM4 processors (up to 16, of course), I think quad channel is REALLY needed when the socket will be upgraded. I hope higher frequencies and dual channel per DIMM of DDR5, that is touted for next generation, will deliver more than what just quad channel on DDR4-3600 would. The CPUs are starved for memory.

If you don't believe me, a dual socket Xeon E5-2687W v2 (Ivy Bridge EP, 8 core, 3,4 GHz CPU - 16 cores in total) smashes (being up to 50 % faster) a 3900XT in real life compute (finite element analysis). Both CPUs were equipped with 128 GB ECC RAM (3200 MHz for the Ryzen, 1333 MHz DDR3 for the Xeon) with maxed out RAM slots, so 2 DIMMs per channel for both, 2 channels for the Ryzen and 8 channels for the Xeon system.

I think it's obvious the Xeon system has much slower cores and the 4 additional cores don't do much for it capability wise. The RAM bandwidth though was almost twice as high on the Xeon system (rated 10600 MB/s theoretical for the DDR3-1333 MHz, effectively quad that due to four times the channels, so 42400 MB/s, vs. 25600 MB/s for DDR4-3200). The SSDs were the same, Samsung's 970s, so PCIe 3.0, but the SSD doesn't play a huge role as putting a 970 into my 8700K system didn't give any improvements compared to the SATA Crucial M550. And the 3900XT hardly any faster than my 8700K even though it has twice the cores.

Memory bandwidth starvation is the only reason I can still see with the Ryzen system. Cinebench benchmarks on the Ryzen system were along the published results though, so the system is performing as it should.

 

.... If you are going to compare a server chip to anything else, you compare it to another server chip...

What you want already exists: threadripper/EPYC

 

Intel would dissagree.

 

why would you throw them ?
4 sticks in dual channel running dual rank will be fast on mainstream platform too

 

Kind of what I'm hoping for. Been dragging this DDR4 with me since 2014-2015 when I bought a 5820K as well. Overclocked it a bit and tightened the timings. 9900KS hopefully will suffice for gaming (for higher framerates) until we get to DDR5. Then I've basically done the entire DDR4 lifecycle with the same 4 sticks.

Anyway, not sure if a lot of people need quad channel as much as they think they do. Some applications benefit, some don't, but gaming wise I haven't seen a huge gain tbh.

 

Isn´t that overkill???

 

There isn't such a thing as overkill. Go back few years and quad core was "overkill". Technology and games will go forward in few years we might see 10-12 core a standard for gaming.

 

Moved to a 5900X since this post, and was the lucky winner of a 64Gb G.SKill kit from Guru3d december contest !

 

Why exactly? Buying a 16-core 5950X costs a bit more compared to ebaying old workstation hardware. Sure, there's a power usage penalty, but the speed is an advantage. And does it matter if a Xeon (not actually server, but workstation) was used for a comparison, that clearly shows memory bandwidth being a problem?

Though on the other hand I did do a comparison with the Threadripper, where it's more or less clear the issue is either the same or worse, if anything.

 

Yes but your looking at 2 different markets or use cases, you want a Machine to do that kind of work thats what you buy you want a decent all rounder that doesnt need to have that kind of memory bandwidth then buy a desktop system your comparison is a non starter, I think you will also find the threadrippers now dominate all Intel workstation class CPUs because they now have new specs for the pro workstation market so a single 3995wx is quicker in Adobe than dual Xeon 8280s and thats before the new versions come out

https://www.amd.com/en/processors/ryzen-threadripper-pro

Right tool for the job buy second hand by all means but its a false economy as your upgrade path is nil

 

I take it back, we just found out Intel's compiler is the issue on Ryzen CPUs, MKL_DEBUG_CPU_TYPE=5 flag fixes the issue... Ryzen is fast again, the time to compute halved with this flag.

Carry on.

Though, regarding upgradeability, on a Ryzen 12-core system with 128 GB of RAM, where do you upgrade to? You can go to 16 cores and a 5000 series CPU, which is a bit of an upgrade, but not a large one. With the used Xeons you can easily start out with for example 8 cores per socket and go to 12 cores per socket, up to 18 if you go LGA-2011-3, and with RAM, because you're on a Xeon with 16 slots in a dual socket system, you can also go up to half or even a full terabyte of RAM. For cheap too.

But in either case, you buy something and use it for a few years, then buy again. Upgradebaility is rarely a useful bonus (if it's present at all in modern platforms).