Discussion Speculation: Zen 4 (EPYC 4 "Genoa", Ryzen 7000, etc.)

[Vattila]

Except for the details about the improvements in the microarchitecture, we now know pretty well what to expect with Zen 3.

The leaked presentation by AMD Senior Manager Martin Hilgeman shows that EPYC 3 "Milan" will, as promised and expected, reuse the current platform (SP3), and the system architecture and packaging looks to be the same, with the same 9-die chiplet design and the same maximum core and thread-count (no SMT-4, contrary to rumour). The biggest change revealed so far is the enlargement of the compute complex from 4 cores to 8 cores, all sharing a larger L3 cache ("32+ MB", likely to double to 64 MB, I think).

Hilgeman's slides did also show that EPYC 4 "Genoa" is in the definition phase (or was at the time of the presentation in September, at least), and will come with a new platform (SP5), with new memory support (likely DDR5).

What else do you think we will see with Zen 4? PCI-Express 5 support? Increased core-count? 4-way SMT? New packaging (interposer, 2.5D, 3D)? Integrated memory on package (HBM)?

Vote in the poll and share your thoughts!

 

[StefanR5R]

So for now, the EPYC 4004 lineup is indeed a 1:1 rebadge of Ryzen 7000, except for an additional 4c/8t/16MB configuration.

The PPT limits aren't given in today's news items on EPYC 4004. But from a quick glance at Phoronix' CPU power monitor graphs, it seems to me as if it's the same formula as with Ryzens: default PPT limit = 1.35 default TDP. (All of the big EPYCs have default PPT limit = default TDP.) Edit: Or is it? On second look, do only the "65 W" EPYCs have 88 PPT limit, while the "105"/120/170 W EPYCs actually have 120/120/170 W PPT limits? Edit 2: And the "120 W" 4484PX might actually have a 105 W PPT limit (in the platform tested by Phoronix).

 

[Abwx]

So for now, the EPYC 4004 lineup is indeed a 1:1 rebadge of Ryzen 7000, except for an additional 4c/8t/16MB configuration.

The PPT limits aren't given in today's news items on EPYC 4004. But from a quick glance at Phoronix' CPU power monitor graphs, it seems to me as if it's the same formula as with Ryzens: default PPT limit = 1.35 default TDP. (All of the big EPYCs have default PPT limit = default TDP.) Edit: Or is it? On second look, do only the "65 W" EPYCs have 88 PPT limit, while the "105"/120/170 W EPYCs actually have 120/120/170 W PPT limits?

105/120/170 W TDPs amount surely to 142/162/230W PPTs, these numbers are the ones for the 7700X/7950X3D/7950X respectively.

 

[igor_kavinski]

Ah, I get it now. AMD knows that Zen 4 will become less attractive after Zen 5 launch so they are just going to rebadge most of their existing Zen 4 chips as Epyc 4004 and that way they maintain their value too coz you obviously don't get server chips at budget prices. Kind of a genius move and the launch happening just before the Zen 5 reveal kinda lends credence to my "suspicion".

 

[Joe NYC]

AMD page for EPYC 4004

https://www.amd.com/en/products/processors/server/epyc/4004-series.html

No comment on naming scheme, but one thing I noticed is there is no 8 core epic with V-Cache...

 

[adroc_thurston]

Ah, I get it now. AMD knows that Zen 4 will become less attractive after Zen 5 launch so they are just going to rebadge most of their existing Zen 4 chips as Epyc 4004 and that way they maintain their value too coz you obviously don't get server chips at budget prices.

no, it's just lower-end server parts are a longer rollout.
server Granite Ridge is next year.

 

[adroc_thurston]

man, this hurts. entry DT xeons being 8c coupled to the last low power server Atom being 7 years old are a bad combo.

 

[Joe NYC]

Patrick Kennedy of ServeTheHome reviews EPYC 4004

The AMD EPYC 4004 is Finally Here and Intel Xeon E Needs an Overhaul

The AMD EPYC 4004 series is finally here and it absolutely wallops Intel Xeon E in the entry-level server market

www.servethehome.com

 

[igor_kavinski]

AMD: We wake Intel up from time to time by kicking them in the b***s!

 

[StefanR5R]

one thing I noticed is there is no 8 core epic with V-Cache...

Server admins who do have an application which benefits from 96 MB last-level cache — a database perhaps — could work around this by taking an EPYC 4{4,5}84PX and containing the application into the V-Cache equipped core complex.

 

[Tuna-Fish]

No comment on naming scheme, but one thing I noticed is there is no 8 core epic with V-Cache...

View attachment 99364

One thing I noticed is that they seem to have priced them to push sales on the 12-core. Possibly they have excess 6-core dies, with both servers and most client users favoring 8-core dies?

Server admins who do have an application which benefits from 96 MB last-level cache — a database perhaps — could work around this by taking an EPYC 4{4,5}84PX and containing the application into the V-Cache equipped core complex.

The problem with this is that many important databases are sold on per-core licenses, where you pay for all the cores in the machine running it (and all the cores in your hot spare...), not just the ones that are actually enabled. Lack of a 8-core vcache option seems to be an actual oversight to me.

 

[igor_kavinski]

Lack of a 8-core vcache option seems to be an actual oversight to me.

Seems 7800X3D is selling too well that they can't "afford" to divert some of the dies towards servers. The number of gamers far outweighs the number of SMBs with usecases for this configuration. And since Zen5X3D won't be available until next year, it's going to remain a hot selling part.

 

[biostud]

One thing I noticed is that they seem to have priced them to push sales on the 12-core. Possibly they have excess 6-core dies, with both servers and most client users favoring 8-core dies?



The problem with this is that many important databases are sold on per-core licenses, where you pay for all the cores in the machine running it (and all the cores in your hot spare...), not just the ones that are actually enabled. Lack of a 8-core vcache option seems to be an actual oversight to me.

What happens if you run a VM only on the 3D cache CCX? Will the DB software be licensed for 8 cores?

 

[Tuna-Fish]

What happens if you run a VM only on the 3D cache CCX? Will the DB software be licensed for 8 cores?

No, why would it be?

To be clear, the license is not enforced by some software mechanism, it's enforced by Oracle goons visiting your data center and physically verifying which machine you are running it on and what cores it has. Welcome to the wonderful world of enterprise software licensing.

 

[biostud]

No, why would it be?

To be clear, the license is not enforced by some software mechanism, it's enforced by Oracle goons visiting your data center and physically verifying which machine you are running it on and what cores it has. Welcome to the wonderful world of enterprise software licensing.

Ok, I'm not in the tech business, so I don't know how these things work

 

[StefanR5R]

One thing I noticed is that they seem to have priced them to push sales on the 12-core. Possibly they have excess 6-core dies, with both servers and most client users favoring 8-core dies?

The non-V-cache 12c part, 4464P, is configured to "65 W" TDP (that is, to 88 W power limit). I think its price is proportional to socket throughput.

many important databases are sold on per-core licenses, where you pay for all the cores in the machine running it (and all the cores in your hot spare...), not just the ones that are actually enabled. Lack of a 8-core vcache option seems to be an actual oversight to me.

After they went through the trouble of making an extra 4 core part just for the EPYC 4004 line (but with mere 16 MB L3$), perhaps they should go the extra mile and make a 4c 96MB part in addition. And price it as a "software licensing optimized" part.

 

[DrMrLordX]

Seems 7800X3D is selling too well that they can't "afford" to divert some of the dies towards servers. The number of gamers far outweighs the number of SMBs with usecases for this configuration. And since Zen5X3D won't be available until next year, it's going to remain a hot selling part.

Eh? Margins on the EPYC part would be higher.

 

[StefanR5R]

I think its price is proportional to socket throughput.

Not quite.
part .... SPEC CPU2017 Int Rate Estimates .... 1KU price .... perf/$
4584PX (16c/32t, V$, 120 W?) ......... 180.0 ........ $ 699 ........ 0.258 / $
4484PX (12c/24t, V$, 105 W?) ......... 148.0 ........ $ 599 ........ 0.247 / $
4564P (16c/32t, 170 W?) .................. 175.0 ........ $ 699 ........ 0.250 / $
4464P (12c/24t, 88 W?) .................... 141.0 ........ $ 429 ........ 0.329 / $
4364P (8c/16t, 120 W?) .................... 105.0 ........ $ 399 ........ 0.263 / $
4344P (8c/16t, 88 W?) ...................... 103.0 ........ $ 329 ........ 0.313 / $
4244P (6c/12t, 88 W?) ........................ 83.4 ........ $ 229 ........ 0.364 / $
4124P (4c/8t, 88 W?) .......................... 58.2 ........ $ 149 ........ 0.391 / $

(SPEC rates are from AMD's press briefing slide deck. Wattages are my guess of the SKU's default PPT limit based on Phoronix' tests.)

 

[eek2121]

Hopefully we get some pro oriented boards to go with these chips along with better stability.

Not quite.
part .... SPEC CPU2017 Int Rate Estimates .... 1KU price .... perf/$
4584PX (16c/32t, V$, 120 W?) ......... 180.0 ........ $ 699 ........ 0.258 / $
4484PX (12c/24t, V$, 105 W?) ......... 148.0 ........ $ 599 ........ 0.247 / $
4564P (16c/32t, 170 W?) .................. 175.0 ........ $ 699 ........ 0.250 / $
4464P (12c/24t, 88 W?) .................... 141.0 ........ $ 429 ........ 0.329 / $
4364P (8c/16t, 120 W?) .................... 105.0 ........ $ 399 ........ 0.263 / $
4344P (8c/16t, 88 W?) ...................... 103.0 ........ $ 329 ........ 0.313 / $
4244P (6c/12t, 88 W?) ........................ 83.4 ........ $ 229 ........ 0.364 / $
4124P (4c/8t, 88 W?) .......................... 58.2 ........ $ 149 ........ 0.391 / $

(SPEC rates are from AMD's press briefing slide deck. Wattages are my guess of the SKU's default PPT limit based on Phoronix' tests.)

PPT and related values should be the exact same as Ryzen.

 

[LightningZ71]

If there ever was a time to put a PLX chip on a motherboard to improve I/O throughput, this would be it. PLX the PCIe 5.0 x16 GPU lanes from the processor out to 64 lanes to drive 4 full length slots on the mobo at PCIe 4.0 (2x over-subscribe) and you'd meet the needs of a huge chunk of the market. Precious few cards out there see any sort of notable performance benefit running at PCIe 5.0 speeds and not many really load up a PCIe 4.0 slot either. If you had a PLX chip that could bifurcate each of those x16 slots down to x4x4x4x4, you'd have a killer I/O machine for M.2 SSD arrays.

 

[StefanR5R]

Hopefully we get some pro oriented boards to go with these chips along with better stability.
[…]
PPT and related values should be the exact same as Ryzen.

Arguably, server chips should have different (read: saner) power limits. Anyway; as I said, Phoronix' power measurements indicated that only the 65 W TDP parts implement the same PPT limit as the Ryzens. The other ones might have a different PPT limit, or the test system had such a marginal cooling that TDP actually hit (but why did the 4484PX pull ≈105 W, not 120 W, and the 4364P ≈120 W, not 105 W?), or there were conservative Voltages in combination with a conservative V/f curve applied, or the measurements or diagrams have mistakes...

I looked, but as of yet, I haven't found a document at AMD's public web site which lists PPT limits. It's straightforward with all of the other EPYC lines; they should make it as straightforward with EPYC 4004 too.

 

[SteinFG]

It's literally cheaper to get a threadripper 24-core + a 700 dollar motherboard, than a gen5 PLX

 

[igor_kavinski]

or there were conservative Voltages in combination with a conservative V/f curve applied, or the measurements or diagrams have mistakes...

Could be AGESA related issues too.

 

[Joe NYC]

Seems 7800X3D is selling too well that they can't "afford" to divert some of the dies towards servers. The number of gamers far outweighs the number of SMBs with usecases for this configuration. And since Zen5X3D won't be available until next year, it's going to remain a hot selling part.

7800x3d was released some time after the other (more expensive) V-Cache models.

Maybe that's what will happen here as well, that 8 core V-Cache model will be released later.

 

[SteinFG]

7800x3d was release some time after the other (more expensive) V-Cache models.

Maybe that's what will happen here as well, that 8 core V-Cache model will be released later.

Nope, it's just that 8-core with 96MB L3 may be prefered over a 12-core, so they upsell it

 

[StefanR5R]

If there ever was a time to put a PLX chip on a motherboard to improve I/O throughput, this would be it. PLX the PCIe 5.0 x16 GPU lanes from the processor out to 64 lanes to drive 4 full length slots on the mobo at PCIe 4.0 (2x over-subscribe) and you'd meet the needs of a huge chunk of the market. Precious few cards out there see any sort of notable performance benefit running at PCIe 5.0 speeds and not many really load up a PCIe 4.0 slot either. If you had a PLX chip that could bifurcate each of those x16 slots down to x4x4x4x4, you'd have a killer I/O machine for M.2 SSD arrays.

Doesn't the cIOD already support x4x4x4x4 bifurcation of the "GPU" PCIe lanes?

And for NVME fanout: ASMedia Promontory 21, also known as AMD B650, can already be cascaded 1+1 (also known as AMD X670), which gives up to 20 fanout lanes (minus up to 8 SATA lanes) plus a bunch of USB ports. If four B650s could be combined into a 1+3 tree, you'd get 36 fanout lanes (minus up to 12 SATA lanes) plus more USB ports than any server needs. The performance loss due to the PCIe cascade might be a tolerable compromise for entry level servers.