Zen 2 for Distributed Computing: Any interest?

Page 6 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

biodoc

Diamond Member
Dec 29, 2005
6,257
2,238
136
I use PSensor to monitor CPU temps in Linux

PSensor and other apps like it get their hardware monitoring data from lm_sensors. As far as I know, there's no support for Zen 2 yet but they are working on it. Initial support will come with the latest kernel and then backports to older kernels will eventually show up. Open source developers for the drivers incorporated into lm_sensors ideally need data sheets from the manufacturers of "super I/O chips" to get accurate voltage/temp/fan measurements from the CPU and MB. Support from Super I/O chip manufacturers for the opensource community varies considerably. If you are unlucky enough to get a new MB with an ITE chip (it87 driver), then you can expect low or no support in linux. My assumption is ITE is very reluctant to release datasheets to the opensource community. On top of that, the opensource developer of the it87 driver called it quits last year. When I bought my Gigabyte X570 MB, I neglected to try to find out what super I/O chip is on the board. It's an it8792 chip which is not supported in the it87 driver. My ASRock x470 MB has an nct6779 chip (Nuvoton?) and there's great support for that chip in the opensource community which I assume comes from the manufacturer. I have no doubt I will eventually get temp readings from that MB once support for Zen 2 (k10temp) shows up in a kernel update.
 

Assimilator1

Elite Member
Nov 4, 1999
24,120
507
126
Re the ASUS Prime X570-P, it 'only' has 8 phase power delivery, does that really matter if I'm never going to get the 16C 3950X? And it at least supports the 3900X, am I missing something?

Hmm, I see the Gigabyte X570 GAMING has a 12 phase power delivery, & it's slightly cheaper, so maybe that would be a better choice....
 
Last edited:

StefanR5R

Elite Member
Dec 10, 2016
5,459
7,717
136
When the 1st Ryzen generation came out, there was the highly interesting "Strictly Technical" thread by The Stilt here at the AnandTech forum. There is a Zen 2 counterpart on OCN: "Strictly technical: Matisse (Not really)"

On Ryzen 3000's performance-per-Watt sweet spot:
The Stilt said:
Some of you might notice that Matisse's power efficiency seems to peak at 3.5GHz, despite the fact that semiconductors do not behave like that. The reason behind this was revealed by Vmin testing, which clearly illustrated that Matisse lacks fused V/F (voltage-frequency) curve below 3.4GHz. This means that below 3.4GHz frequencies the voltage is always at the same level,
 
  • Like
Reactions: Assimilator1

Assimilator1

Elite Member
Nov 4, 1999
24,120
507
126
An update to my previously posted power figures, I've just discovered that because my 2 monitors are connected to different type ports this stops my RX 580 from dropping the cards RAM to idle speeds, this adds about 25w to power usage from the wall (I confirmed this from the wall plug meter as well GPU-Z).
I can't remember if I got that MW power figures with the monitors on or off (the 2nd monitor's power button is pretty much dead), so I will have to run the test again!.

But for now, running Asteroids (12T), 2nd monitor off & GPU @ full idle!, it's pulling 222w from the wall. Asteroids pulls more than MW, for those who don't know.
[update] Running MW@H on just the CPU (12T), it's pulling 200w, with the 2nd monitor on my PC was pulling ~227w.
 
Last edited:

StefanR5R

Elite Member
Dec 10, 2016
5,459
7,717
136
Alas there is still no solid info about Threadripper 3000 yet. Only this: All known TR4/X399 mainboards have got a 16 MByte BIOS EEPROM, which is too small to hold AGESAs for both Zen(+) and Zen2.

The same is true for SP3 server mainboards. To add insult to injury, most server mainboard manufacturers cheaped out and soldered the EEPROM, instead of providing a socketed one. In addition, there aren't Zen2-only BIOSes made available which would fit into that small EEPROM, only combined BIOSes which require a 32 MByte EEPROM. The upshot is, practically none of the mainboards which were created for EPYC 7001 alias Naples are compatible with EPYC 7002 alias Rome.

In theory, they can be made compatible either by resoldering the EEPROM, or if the BMC (baseboard management controller) is able to perform the programming, by flashing a special Zen2-only BIOS if you can get one. In practice, new revisions of these mainboards are being manufactured, and owners of revision 1.x boards are left out in the cold.

Here in Germany, there are no Rome compatible mainboards available in retail yet. I read that local system integrators are receiving single-socket mainboards with Rome compatibility since a couple of weeks, but no compatible dual-socket boards yet. Furthermore, Naples processors are being discounted now, helping integrators and vendors to get rid of their stock of old incompatible SP3 mainboards.

In short: DIYers beware.
 

motqalden

Member
Jun 13, 2009
82
111
106
I get reliable temps from Psensor for my 3900x in linux mint 19.1 at least overall temps, not per core. the line: SMBUSMASTER 0 is the core temp and matches what i see in windows (translation hotter than the sun)
 

Markfw

Moderator Emeritus, Elite Member
May 16, 2002
25,478
14,434
136
Alas there is still no solid info about Threadripper 3000 yet. Only this: All known TR4/X399 mainboards have got a 16 MByte BIOS EEPROM, which is too small to hold AGESAs for both Zen(+) and Zen2.

The same is true for SP3 server mainboards. To add insult to injury, most server mainboard manufacturers cheaped out and soldered the EEPROM, instead of providing a socketed one. In addition, there aren't Zen2-only BIOSes made available which would fit into that small EEPROM, only combined BIOSes which require a 32 MByte EEPROM. The upshot is, practically none of the mainboards which were created for EPYC 7001 alias Naples are compatible with EPYC 7002 alias Rome.

In theory, they can be made compatible either by resoldering the EEPROM, or if the BMC (baseboard management controller) is able to perform the programming, by flashing a special Zen2-only BIOS if you can get one. In practice, new revisions of these mainboards are being manufactured, and owners of revision 1.x boards are left out in the cold.

Here in Germany, there are no Rome compatible mainboards available in retail yet. I read that local system integrators are receiving single-socket mainboards with Rome compatibility since a couple of weeks, but no compatible dual-socket boards yet. Furthermore, Naples processors are being discounted now, helping integrators and vendors to get rid of their stock of old incompatible SP3 mainboards.

In short: DIYers beware.
For my x399 Taichi (most of my x399 motherboards) it says this:
BIOS:
- 128MB AMI UEFI Legal BIOS with GUI support

So I should be good, right ?
 

StefanR5R

Elite Member
Dec 10, 2016
5,459
7,717
136
For my x399 Taichi (most of my x399 motherboards) it says this:
BIOS:
- 128MB AMI UEFI Legal BIOS with GUI support

So I should be good, right ?
The Taichi, like the rest of the lot unfortunately, has got a 128 Mbit EEPROM.
 

biodoc

Diamond Member
Dec 29, 2005
6,257
2,238
136
Here in Germany, there are no Rome compatible mainboards available in retail yet. I read that local system integrators are receiving single-socket mainboards with Rome compatibility since a couple of weeks, but no compatible dual-socket boards yet.

At newegg in the States, you can purchase a supermicro board with a Rome processor preinstalled that is tested at the factory. For the Supermicro H11SSL-i board, EPYC Rome processors apparently require version 2.x of the MB. No dual-socket boards are listed at newegg yet.

"Single AMD EPYC™ 7001/7002* Series Processor (*AMD EPYC 7002 series drop-in
support requires board revision 2.x)"

TR 3000 rumors summarized by Tom's Hardware. TRX40 motherboards are on route?

I get reliable temps from Psensor for my 3900x in linux mint 19.1 at least overall temps, not per core. the line: SMBUSMASTER 0 is the core temp and matches what i see in windows (translation hotter than the sun)

That's "cool". My X470 (ASRock) and X570 (Gigabyte) sensors output (Mint 19.2) do not have the SMBUSMASTER 0 listed. I have 3700X processors installed in both. Which MB do you have? I'm thinking of replacing the x470 board with an x570 but I'm reluctant until I find a board that supports accurate sensors output.

After promising Zen 2 linux hwmon support late last year, AMD finally provides the info to linux developers. Hopefully this patch will be backported to the mainstream kernels.
 

TennesseeTony

Elite Member
Aug 2, 2003
4,204
3,631
136
www.google.com
I know a few of you have both, 2nd gen and 3rd gen Ryzen systems. Is this chart from cpubenchmarks.net even remotely accurate, in your experience? I can fully understand 1700 to 2700x performance boost, but I am having MUCH trouble believing a 3600 with two less actual cores is able to so 'handily whoop the tar out of' ("easily defeat", for my English as a 2nd language friends) the 2700X....

If this is even remotely true, I am looking at my fleet of Haswells, as if they are no more than expensive paper weights...

Ryzen progress.PNG



This is the chart that REALLY takes my breath away... That...that CAN'T be correct? Can it?

Screenshot_2019-10-05 PassMark - CPU Comparison AMD Ryzen 7 2700X vs AMD Ryzen 9 3900X vs AMD ...png
 
Last edited:

Markfw

Moderator Emeritus, Elite Member
May 16, 2002
25,478
14,434
136
I know a few of you have both, 2nd gen and 3rd gen Ryzen systems. Is this chart from cpubenchmarks.net even remotely accurate, in your experience? I can fully understand 1700 to 2700x performance boost, but I am having MUCH trouble believing a 3600 with two less actual cores is able to so 'handily whoop the tar out of' ("easily defeat", for my English as a 2nd language friends) the 2700X....

If this is even remotely true, I am looking at my fleet of Haswells, as if they are no more than expensive paper weights...

View attachment 11647



This is the chart that REALLY takes my breath away... That...that CAN'T be correct? Can it?

View attachment 11649
Yup, its WAY distorted. Just look at the 2990wx scores !!! And yes, I can't sell my 2683v3's or motherboards to save my life, so a couple my go back into service.....
 

crashtech

Lifer
Jan 4, 2013
10,521
2,111
146
Are those Passmark scores? Passmark really seems to love Zen2 in a big way.

Yeah, they are. Not sure what the deal is, because before Zen2, Passmark really seemed to favor Intel.

 

VirtualLarry

No Lifer
Aug 25, 2001
56,226
9,990
126
Strictly speaking, as far as the 2700(X) goes, and the 3600, I sold a 2700 that I was using under 240mm AIO WC, running @ 4.0Ghz, and got a 3600, which has two fewer cores, but the IPC increase, combined with the 256-bit AVX2 support, seems like it performs better, and uses less power, at least on things like PrimeGrid, that can actually take real advantage of the 256-bit AVX2 support. It auto-clocks at ~4Ghz or just below, for an all-core boost speed, depending on temps.
 
  • Like
Reactions: TennesseeTony

StefanR5R

Elite Member
Dec 10, 2016
5,459
7,717
136
Yup, its WAY distorted. Just look at the 2990wx scores !!!
I agree. While I don't know what the benchmark is actually measuring, it must be something which does not scale as well as most DC applications do on the WX models of Threadripper.

Regarding Ryzen 2000 -> Ryzen 3000 performance gain in DC applications, @biodoc's (edit:) short-term Rosetta results show a ratio of 1 : ~1.3 comparing an 8c/16t/105W CPU with an 8c/16t/65W CPU (IOW ~30% per-core throughput gain from Zen+ to Zen2 combined with significantly lower power use). I think this may be the order of magnitude of the gain to expect in many DC projects. Though a few like PrimeGrid's LLR-based applications may show a higher gain.
 
Last edited:
  • Like
Reactions: TennesseeTony

biodoc

Diamond Member
Dec 29, 2005
6,257
2,238
136
Those benchmarks appear to be from PassMark's CPU benchmarking software (CPU Mark). It looks to be a combined score of multiple single and multithread benchmarks as tested on Windows OS (no linux version). For DC computing, the single thread score is the most relevant (second to last line in the tables above). If you all recall, the high core count TR WX models did not do well in multithreaded benchmarks when using Windows as the OS as reported by Anandtech and others. The 2990WX did much better on linux in these benchmarks as tested by phoronix. I think this is a good explanation for the poor CPU Mark score for the 2990WX.

I agree with @VirtualLarry . For Zen 2, the ~13% increase in IPC, lower power consumption, and fully functional AVX2 make them attractive for DC use. Multithreaded benchmarks aren't really that relevant for most DC work.
 
  • Like
Reactions: TennesseeTony

Assimilator1

Elite Member
Nov 4, 1999
24,120
507
126
Well, I still haven't upgraded to Zen 2, I'd decided that I wouldn't get an X570 mbrd (mainly due to the high cost) & that I would get a Zen2 compatible 400 series mbrd instead. I thought I'd found the ideal mbrd in the MSI B450 TOMAHAWK MAX (although I'm not generally keen with MSI), it has an excellent power delivery system but the audio is meant to be pretty bad, and 1 of the worst! I couldn't find anything comparable, so I gave up (for the time being), anyone got any suggestions?

Actually, I don't know why I didn't consider the
MSI B450 GAMING PRO CARBON, has it only recently gained Zen 2 support?
THGs review of it is ok, although the memory overclocking wasn't great (although that was on a Zen CPU), is it good enough for Zen 2 to hit highest default memory speeds? I can't remember what that is now!.... [update] 3200, so that's a no then, back to the drawing board :(
 
Last edited:

biodoc

Diamond Member
Dec 29, 2005
6,257
2,238
136
Well, I still haven't upgraded to Zen 2, I'd decided that I wouldn't get an X570 mbrd (mainly due to the high cost) & that I would get a Zen2 compatible 400 series mbrd instead. I thought I'd found the ideal mbrd in the MSI B450 TOMAHAWK MAX (although I'm not generally keen with MSI), it has an excellent power delivery system but the audio is meant to be pretty bad, and 1 of the worst! I couldn't find anything comparable, so I gave up (for the time being), anyone got any suggestions?

Actually, I don't know why I didn't consider the
MSI B450 GAMING PRO CARBON, has it only recently gained Zen 2 support?


I believe you'll need a ZEN or ZEN+ processor in a pre-570X board to upgrade to a bios that supports a ZEN 2 processor. The problem is bios capacity in most if not all cases. Read this from MSI that explains the issue.

Maybe the prices on x570 boards will come down in the coming months.
 

Assimilator1

Elite Member
Nov 4, 1999
24,120
507
126
Don't want to wait that long really.

Btw, according to ebuyer at least, both those MSI mbrds support Zen 2 out the box. And so does MSI's site for the B450-GAMING-PRO-CARBON-AC at least.
(edited my previous post btw re RAM speeds). Further to that, it seems some users are achieving much better RAM speeds with that mbrd than THG were :), maybe it was just their 'old' CPU they were using? (2700X). Looks like it might be a good option afterall :).
 
Last edited:

crashtech

Lifer
Jan 4, 2013
10,521
2,111
146
@Assimilator1 , I'm going for an X470 board because of an aversion to proprietary chipset fans, but it's not clear that all of them would work out of the box with Zen2. Some of them advertise this explicitly, like the X470 Taichi.
 
Last edited:

extide

Senior member
Nov 18, 2009
261
64
101
www.teraknor.net
Alas there is still no solid info about Threadripper 3000 yet. Only this: All known TR4/X399 mainboards have got a 16 MByte BIOS EEPROM, which is too small to hold AGESAs for both Zen(+) and Zen2.

Why do you think that? All of the current consumer boards with 16MB BIOSes are able to support Zen, Zen+ and Zen2. It's only the PRE-Zen chips that they have dropped.
 

Assimilator1

Elite Member
Nov 4, 1999
24,120
507
126
@Assimilator1 , I'm going for an X470 board because of an aversion to proprietary chipset fans, but it's not clear that all of them would work out of the box with Zen2. Some of them advertise this explicitly, like the X470 Taichi.
Yea I'm not keen on the chipset fans either, but that alone wouldn't have stopped me.
I've got to do some more reading on achievable RAM speeds with the MSI mbrd I mentioned, if it looks good I'll be going for that :).
Then my next major buy is a BMW M3 e92 as my fun/fair weather car!! :D
 

StefanR5R

Elite Member
Dec 10, 2016
5,459
7,717
136
Why do you think that? All of the current consumer boards with 16MB BIOSes are able to support Zen, Zen+ and Zen2. It's only the PRE-Zen chips that they have dropped.
I was assuming that there is going to be the same split of firmwares as between Naples and Rome. Maybe this is a premature conclusion.
 

biodoc

Diamond Member
Dec 29, 2005
6,257
2,238
136
All of the current consumer boards with 16MB BIOSes are able to support Zen, Zen+ and Zen2. It's only the PRE-Zen chips that they have dropped.

Support for Zen, Zen+ and Zen 2 is a bit confusing to me. Take the ASRock Taichi x470 and x570 boards for example.

Taichi X470 support

The CPU support list does list Zen, Zen+ and Zen 2.
The latest bios offers AMD AGESA Combo-AM4 1.0.0.3 ABB support but with this disclaimer:
"ASRock do NOT recommend updating this BIOS if Pinnacle, Raven, Summit or Bristol Ridge CPU is being used on your system."

No bios support for AMD AGESA Combo-AM4 1.0.0.3 ABBA yet

Taichi X570 support

The CPU support list only includes Zen+ and Zen2

The latest bios supports AMD AGESA Combo-AM4 1.0.0.3 ABBA

My question is will B450 and X470 boards offer AGESA 1.0.0.3 ABBA and further updates in the future?