• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

AMD confirms Zen 6 CPU launch plans

erek

erek

Fully [H]
2FA
Joined
Dec 19, 2005
Messages
17,489
Update: https://hardforum.com/threads/amd-confirms-zen-6-cpu-launch-plans.2045513/post-1046350583

"The delayed arrival of AMD compiler support induced significant challenges, especially given GCC's annual release schedule and LLVM/Clang's six-month cycle. Although both compilers release bug-fix updates that could allow for backporting, Linux distributions rarely adopt these versions. This meant that obtaining optimized compiler support for new AMD EPYC and Ryzen processors at launch was especially difficult. However, with the GCC compiler now officially supporting the upcoming "Zen 6" products, these common issues have been addressed. AMD might have changed its approach due to the new instruction support introduced with "Zen 6."

For "Zen 6," AMD is integrating instructions like AVX512_BMM, AVX512_FP16, AVX_NE_CONVERT, AVX_IFMA, and AVX_VNNI_INT8, among others. This is particularly noteworthy because 16-bit AVX-512 calculations will now be possible on consumer-oriented desktop CPUs, allowing developers to efficiently accelerate applications and data paths that leverage AVX-512. The introduction of these instructions means that the regular CPU becomes a universal base platform, fulfilling its intended role with enhanced observability."

Source: https://www.techpowerup.com/344208/amd-zen-6-gets-official-software-enablement-ahead-of-the-launch
 
Last edited:
erek said:
For "Zen 6," AMD is integrating instructions like AVX512_BMM, AVX512_FP16, AVX_NE_CONVERT, AVX_IFMA, and AVX_VNNI_INT8, among others. This is particularly noteworthy because 16-bit AVX-512 calculations will now be possible on consumer-oriented desktop CPUs, allowing developers to efficiently accelerate applications and data paths that leverage AVX-512. The introduction of these instructions means that the regular CPU becomes a universal base platform, fulfilling its intended role with enhanced observability."
Click to expand...
not sure what this mean, zen5 core in a 9600x cpu or an Epyc cpu are the same I though ?, I imagine xeon had those for a while...

But this make it sound as if customer desktop amd cpu have not had been exactly the same as epyc for a while now
 
RDNA 5 appears in LLVM

Screenshot_20260124_055328_Opera.jpg


https://x.com/Kepler_L2/status/2014828790266950028?s=20
 

“AMD’s Mark Papermaster confirms AMD’s Zen 6 CPU launch plans​

AMD’s Chief Technology Officer and Executive Vice President, Mark Papermaster, has confirmed that AMD plans to launch its next-generation Zen 6 CPUs at its “Advancing AI” event on July 22nd-23rd. This will start with AMD’s sixth-generation EPYC processors, which will be part of AMD’s new Helios AI servers.

Papermaster claims that AMD’s Zen 6 EPYC CPUs will deliver “leadership x86” performance and strong performance in both traditional x86 workloads and new AI workloads. Inside Helios racks, these CPUs will be paired with new AMD Instinct MI455 GPUs.

I’ll say for me, every enterprise conversation I’m having now, this comes up. Because enterprises have, you know, decades of running x86. They’re not going to move that install base. And what we’ve done at AMD, you know, since we launched the new Zen processor back in 2017. We’re now on our sixth generation. So, at our Advancing AI event on July 22nd and 23rd, we’re rolling out this new generation. You know, it continues the kind of leadership x86 CPU, but it’s designed in such a way that it matches what I just described a moment moment ago. It’s optimized for standalone x86 traditional workloads.
– AMD’s Mark Papermaster – Chief Technology Officer and Executive Vice President“
Click to expand...

View: https://youtu.be/Ufz33EGOkfs

source: https://overclock3d.net/news/cpu_mainboard/amd-confirms-zen-6-launch-plans-for-july-22nd-23rd/
 
erek said:

“AMD’s Mark Papermaster confirms AMD’s Zen 6 CPU launch plans​

AMD’s Chief Technology Officer and Executive Vice President, Mark Papermaster, has confirmed that AMD plans to launch its next-generation Zen 6 CPUs at its “Advancing AI” event on July 22nd-23rd. This will start with AMD’s sixth-generation EPYC processors, which will be part of AMD’s new Helios AI servers.

Papermaster claims that AMD’s Zen 6 EPYC CPUs will deliver “leadership x86” performance and strong performance in both traditional x86 workloads and new AI workloads. Inside Helios racks, these CPUs will be paired with new AMD Instinct MI455 GPUs.



View: https://youtu.be/Ufz33EGOkfs

source: https://overclock3d.net/news/cpu_mainboard/amd-confirms-zen-6-launch-plans-for-july-22nd-23rd/
Click to expand...

Isn't that for servers? It's cool to finally see Zen 6 but it's not something anyone can realistically buy. AMD's chips already beat AWS Graviton5 by a good amount, which means Zen6 will just be overkill. But hurray for servers I guess?
 
DukenukemX said:
. AMD's chips already beat AWS Graviton5 by a good amount,
Click to expand...
at 24% the higher price tag for 17% faster geomen it is not a clean win, couild be amazon taking the loss of course, but could be a chip that do more per dollar, being on TSMC 3N.

Amazon put 192 of those significantly smaller cores on a 4 chiplet chip, running those benchmark using $ instead of core count as the base lane would have been an interesting other one to look at and AMD could use a node refresh to match them.

As long as local agentic AI and others stay niche it will be mostly for server yes but that could change.
 
LukeTbk said:
at 24% the higher price tag for 17% faster geomen it is not a clean win, couild be amazon taking the loss of course, but could be a chip that do more per dollar, being on TSMC 3N.

Amazon put 192 of those significantly smaller cores on a 4 chiplet chip, running those benchmark using $ instead of core count as the base lane would have been an interesting other one to look at and AMD could use a node refresh to match them.

As long as local agentic AI and others stay niche it will be mostly for server yes but that could change.
Click to expand...
Um, he did include a performance/$. In terms of cost, Graviton5 doesn't even beat Graviton4, let alone EPYC. It beat Graviton5 by quiet a bit. AMD's Zen6 based EPYC is just going to beat that horse even more so.
Epyc performance per dollar.png
 
DukenukemX said:
Um, he did include a performance/$. In terms of cost, Graviton5 doesn't even beat Graviton4, let alone EPYC. It beat Graviton5 by quiet a bit. AMD's Zen6 based EPYC is just going to beat that horse even more so.
View attachment 815128
Click to expand...
Amazon doesn’t build its servers as a single workload object, they build them to farm.

Any one of those Graviton CPU’s is running 4-12 other customers work and god only knows how many websites and databases.

Amazons mandate for most of its builds is “how many customers can I fit in the smallest space?”.

And for that Graviton does a really good job, 5 handles a crapload of small to medium jobs like an absolute boss.
 
DukenukemX said:
Um, he did include a performance/$. In terms of cost, Graviton5 doesn't even beat Graviton4, let alone EPYC. It beat Graviton5 by quiet a bit. AMD's Zen6 based EPYC is just going to beat that horse even more so.
Click to expand...
for this very specific workload (as they say there is some case), but in general because you pay 24% more per hours for in general 17% more performance than not the usual.
 
Lakados said:
Amazon doesn’t build its servers as a single workload object, they build them to farm.

Any one of those Graviton CPU’s is running 4-12 other customers work and god only knows how many websites and databases.

Amazons mandate for most of its builds is “how many customers can I fit in the smallest space?”.

And for that Graviton does a really good job, 5 handles a crapload of small to medium jobs like an absolute boss.
Click to expand...

IIRC (been a few years since I needed to worry about it for work), everything except the absolute cheapest AWS tiers gave you dedicated cores not shared ones. IIRC the shared code option one gave you use credits equal to something like 25% load for your nominal count with the option to buy additional credit if needed during extended surges. I don't think I ever knew if Amazon's load balancer could move customers around if one server had a bunch of customers spike loads at the same time or not.

Unless you're buying the biggest size on offer you're still sharing a CPU; but I believe the hypervisor is able to enforce even memory and PCIe bandwidth sharing so that your performance shouldn't be affected by what other customers are doing with their cores.

Edit: At that job we only used x86 not any of Amazon's ARM offerings.
 
If these EPYC turn out to be great performers relative to the price and otherwise trounce Intel's current and near-term server CPU offerings that's nice to see, but I'm more interested to hear about the release date for Zen6 powered Ryzen and Threadripper! While having Epyc release ahead of the rest makes sense, I would be impressed if AMD chose to release Threadripper next and then high end Ryzen (including 3D cache models), returning to the long forgotten paradigm where high end offerings came ahead of those cut down offering enthusiasts and others buying higher tiers an earlier and longer availability as part of the value.

Ideally , regarding 3D Cache this would also mean Threadripper CPUs would finally offer models with it included at allll, and Ryzen would release their higher end cache models concurrently, rather than waiting a considerable amount of time after the rest of the lineup. As someone who used to greatly favor HEDT options over a decade ago, it seems like modern Threadripper is at a loss - greatly increased prices, segmentation for its own sake and fear of stepping on EYPC's toes, as well as greatly reduced use-case from wide ranging enthusiast benefits to primarily focused on high core count, parallelized loads; single or few core performance would be impared vs Ryzen even before the lack of 3D VCache. In the old days, be it Intel or AMD, picking up a HEDT CPU meant you got early or at least concurrent arrival of the newest technology, equal or better gaming/overclocking/enthusiast single/few core performance vs mainstream's highest offerings plus the increased cores for creative/workstation/server software able to make use of it, all at a reasonable price increase given the technical enhancements of the platform including triple/quad-channel RAM and a wider array of PCi-E lanes.

Alas, I don't have much faith that AMD will realign Threadripper this way, but I can at least hope that it and Ryzen will debut in stores before long with the latter's 3D VCache models - clearly some of the company's most desirable and successful sellers - launching alongside the rest of Ryzen's lineup. Come to think of it, its 2026 - dare I hope that AMD will actually implement OpenSIL (FOSS board firmware and updates, replacing AGESA and easily Coreboot compatibls) from the start of Zen6 lineups? This would be a major benefit to privacy and data sovereignty at a time when greatly needed and would even one-up the quasi-unofficial methods used to easily install Coreboot on some Intel platforms, so I hope AMD has not forgotten nor intends to keep it an EPYC only limited feature.
 
DanNeely said:
IIRC (been a few years since I needed to worry about it for work), everything except the absolute cheapest AWS tiers gave you dedicated cores not shared ones. IIRC the shared code option one gave you use credits equal to something like 25% load for your nominal count with the option to buy additional credit if needed during extended surges. I don't think I ever knew if Amazon's load balancer could move customers around if one server had a bunch of customers spike loads at the same time or not.

Unless you're buying the biggest size on offer you're still sharing a CPU; but I believe the hypervisor is able to enforce even memory and PCIe bandwidth sharing so that your performance shouldn't be affected by what other customers are doing with their cores.

Edit: At that job we only used x86 not any of Amazon's ARM offerings.
Click to expand...
Amazon doesn’t share cores, ARM is bad at it and it’s a terrible security practice.
Graviton has an absurd amount of resources for networking and parallel communication to networked storage, pcie lane communication, and isolation for everything.

Other platforms don’t have that, so when Amazon starts farming out 12 cores here and 8 cores there, the hardware has things in place to optimize for those situations and it shows in both the benchmarks and the power bill.
 
You must log in or register to reply here.
Back
Top