Monday, October 24th 2022

AMD Set to Unveil its Next Generation Server Processors on the 10th of November

AMD appears to like to play coy when it comes to new product announcements, or at least the reveal of upcoming product announcements. Just as with its November 3rd event, the company has put out a miniscule teaser for its November 10th announcement of what the company is simply calling "the unveiling of our next-gen server processors" on Twitter. The event will kick off at 10 am Pacific time and it appears there will be a live stream, as AMD is inviting people to watch the event online. It's highly likely that we're talking about new EPYC parts here, as the event is called "together we advance_data centers".
Sources: AMD, AMD (on Twitter)
Add your own comment

17 Comments on AMD Set to Unveil its Next Generation Server Processors on the 10th of November

#1
defaultluser
definitely time to push all that excess production into server CPUs
Posted on Reply
#2
Aldain
defaultluserdefinitely time to push all that excess production into server CPUs
Genoa came before desktop zen4
Posted on Reply
#3
ThrashZone
Hi,
Yeah everyone wants new 7kx3d but they do this :laugh:
Posted on Reply
#4
cvaldes
ThrashZoneHi,
Yeah everyone wants new 7kx3d but they do this :laugh:
AMD will sell far more server CPUs to datacenter customers than they will sell 7000 Series X3D CPUs to retail.
Posted on Reply
#6
Punkenjoy
This is indeed expected from AMD chiplets strategy. So they start the production in advance, they will bin the best CCD and keep them for Zen 4 X3D desktop, (The one that can keep the highest frequency).

On the server side, they have less constraint because the clocks are much lower. They probably have way more chips that run at those lower frequency that can keep the high frequency of Zen 4 Desktops. Once they have accumulated enough stock, they will sell those to gamers.


The same thing will happen with DDR5 pricing and high frequency. The server demand for DDR5 chip will explode, manufacturer will be able to bin the better chips (as a normal chip would be able to run at JDEC speed no problem). The more they produce, the more they have to produce higher quality memory kit and the price will go down. Indeed, they will also get better at producing them but the binning parts will play a big roles in democratizing the DDR5 memory for the public.

So for AMD, they can push all their standard binned chips to epic since they will all be able to run at low power/low frequency in those, then keep the one that substain high clock for desktop.
Posted on Reply
#7
Patriot
AldainGenoa came before desktop zen4
How so?

Desktop is launched and available ...
Posted on Reply
#8
cvaldes
TheinsanegamerNTheir client market sales are quite a bit higher then their data center sales

(post truncated for brevity)
AMD thinks about Long Term TAM too.
Posted on Reply
#9
Wirko
PunkenjoyThis is indeed expected from AMD chiplets strategy. So they start the production in advance, they will bin the best CCD and keep them for Zen 4 X3D desktop, (The one that can keep the highest frequency).

On the server side, they have less constraint because the clocks are much lower. They probably have way more chips that run at those lower frequency that can keep the high frequency of Zen 4 Desktops. Once they have accumulated enough stock, they will sell those to gamers.


The same thing will happen with DDR5 pricing and high frequency. The server demand for DDR5 chip will explode, manufacturer will be able to bin the better chips (as a normal chip would be able to run at JDEC speed no problem). The more they produce, the more they have to produce higher quality memory kit and the price will go down. Indeed, they will also get better at producing them but the binning parts will play a big roles in democratizing the DDR5 memory for the public.

So for AMD, they can push all their standard binned chips to epic since they will all be able to run at low power/low frequency in those, then keep the one that substain high clock for desktop.
That may be true but it's not a given, and AMD won't tell us much about binning. Perhaps most chips reach 5.3 GHz and qualify for a Ryzen but fewer can run at 3.7 GHz at full load while drawing low power. The 96-core, 12-chiplet Epyc will supposedly have TDP set at 360 W, which is 30 W per chiplet. Throw Threadrippers into the mix and binning becomes even less straightforward.

(3.7 GHz appears to be the maximum for the Epyc 9xx4, except the F series which should go above 4 GHz.)
Posted on Reply
#10
AusWolf
I understand putting out teasers for consumer products to boost the hype, but what's the point with server CPUs that people/companies buy as and when needed?
Posted on Reply
#11
cvaldes
AusWolfI understand putting out teasers for consumer products to boost the hype, but what's the point with server CPUs that people/companies buy as and when needed?
Actually corporate buyers do put together purchasing plans ahead of time. Often their buying cycles are tied to fiscal periods and each organization can have a different fiscal year reporting period.

Oh, it's worth pointing out that usually there's a budget attached to a purchase.

Even if the exact pricing isn't disclosed, just knowing the direction of where product line is headed as well as a general timetable can help these corporate buyers plan accordingly.

A lot of this type of equipment aren't just drop in replacements for something that broke. Often they are tied to a new facility build-out which might have been years in the planning (for the physical building).

Maybe you're an IT staffer who is used to working with a couple of 19" racks and just replacing things when they stop working. Not everyone does that. There are other usage cases beyond what you experience.
Posted on Reply
#12
Punkenjoy
WirkoThat may be true but it's not a given, and AMD won't tell us much about binning. Perhaps most chips reach 5.3 GHz and qualify for a Ryzen but fewer can run at 3.7 GHz at full load while drawing low power. The 96-core, 12-chiplet Epyc will supposedly have TDP set at 360 W, which is 30 W per chiplet. Throw Threadrippers into the mix and binning becomes even less straightforward.

(3.7 GHz appears to be the maximum for the Epyc 9xx4, except the F series which should go above 4 GHz.
it's actually much lower than 30w because the I/O die on Genoa probably use quite a lot power, not sure how much but it was up to 100w on previous gen.

But if you look at laptop parts, a 6800U would Zen3+ on 6nm have a base of 2.7 GHz at 15watt and this include the power budget of the memory controller and igpu.

For Epyc Genoa, on the 96 cores part, the base is not 3.7, that is the boost frequency, it's 2.15 GHz and that should be easily achievable with low power.

For Threadripper, they can bin them, but it is worth it for AMD? probably not and it's probably why it took so long to get Zen3 Threadripper.
cvaldesActually corporate buyers do put together purchasing plans ahead of time. Often their buying cycles are tied to fiscal periods and each organization can have a different fiscal year reporting period.

Oh, it's worth pointing out that usually there's a budget attached to a purchase.

Even if the exact pricing isn't disclosed, just knowing the direction of where product line is headed as well as a general timetable can help these corporate buyers plan accordingly.

A lot of this type of equipment aren't just drop in replacements for something that broke. Often they are tied to a new facility build-out which might have been years in the planning (for the physical building).

Maybe you're an IT staffer who is used to working with a couple of 19" racks and just replacing things when they stop working. Not everyone does that. There are other usage cases beyond what you experience.
Also, release announcement aren't just for buyers, they are also for shareholders.

Also, now with cloud vendors, it's quite easy to swap what you use, in few minutes you can go from cascade lake to Genoa in Azure. If they convince people to rent the Azure VM, Microsoft will have to order more cpu from AMD. If they are confident on their performance, they will incite people to use it.

It's just good marketing practice to let your people know that your CPU is available and run well. AMD still need more mindshare on datacenter. There are more and more people open to AMD in Datacenter but the de facto option remain Intel.
Posted on Reply
#13
cvaldes
PunkenjoyAlso, release announcement aren't just for buyers, they are also for shareholders.

Also, now with cloud vendors, it's quite easy to swap what you use, in few minutes you can go from cascade lake to Genoa in Azure. If they convince people to rent the Azure VM, Microsoft will have to order more cpu from AMD. If they are confident on their performance, they will incite people to use it.

It's just good marketing practice to let your people know that your CPU is available and run well. AMD still need more mindshare on datacenter. There are more and more people open to AMD in Datacenter but the de facto option remain Intel.
Yes, shareholders too absolutely.

But also hardware and software partners as well. It's not like AMD manufactures datacenter boxes or writes the software for data mining, etc.

There are far more people involved than AMD and some IT flunky who mounts it into place with a cordless screwdriver.
Posted on Reply
#14
Wirko
AusWolfI understand putting out teasers for consumer products to boost the hype, but what's the point with server CPUs that people/companies buy as and when needed?
The retail market for Zen 3 Epycs and associated motherboards seems to be quite alive. I'm looking at the data on geizhals.eu (EU, mostly Germany) and there are tens of retailers you can buy an Epyc from (any model, or almost). There's also a good choice of 1P and 2P motherboards at prices that aren't exorbitant. Now who's buying all that, I don't know, I can just guess. Maybe small and medium sized businesses with IT departments large enough to build and maintain their own servers, and unwilling to rent. Also, people and companies building clusters for rendering/simulations/etc.
Posted on Reply
#16
Patriot
cvaldesYes, shareholders too absolutely.

But also hardware and software partners as well. It's not like AMD manufactures datacenter boxes or writes the software for data mining, etc.

There are far more people involved than AMD and some IT flunky who mounts it into place with a cordless screwdriver.
Lol, the partners have had the chips for ~9mo-1yr before launch date. The 10th is also probably the NDA lift on when they get to talk about their next gen AMD based products.
Posted on Reply
#17
cvaldes
PatriotLol, the partners have had the chips for ~9mo-1yr before launch date. The 10th is also probably the NDA lift on when they get to talk about their next gen AMD based products.
Yes for OEM partners but not customers. And certainly not software developers. We have already gone through the fact that there are multiple parties that are affected by these types of upcoming announcements.

I'm not sure how AMD does this on the server side but on the consumer side, AMD and NVIDIA bask in the limelight on launch day. OEM partners must typically wait a day or so before they can make their announcements as to not overshadow the chip designers.

So this heads-up from AMD gives the media and their readership an early notice to expect a number of announcements on the 10th and for several days after.
Posted on Reply
Add your own comment
May 19th, 2024 21:34 EDT change timezone

New Forum Posts

Popular Reviews

Controversial News Posts