Discussion AMD's Future APU Gone ARM?

[SarahKerrigan]

1. They are dropping the licensed cores for accelerating custom cores with customers.
2. They aren't out yet.
3. https://www.geekbench.com/blog/2023/09/geekbench-62/ && https://www.geekbench.com/blog/2021/03/geekbench-54/
They added support awhile ago: https://browser.geekbench.com/v6/cpu/5004506 Can literally just search risc-v and get risc-v scores: https://browser.geekbench.com/search?q=risc-v

SiFive is just the topping, the ones to follow are those coming after: Ventana Micro Systems, Tenstorrent, Rivos and Akeana. Which is the first wave of Datacenter CPUs that will get Desktop CPU/APUs.

I have zero expectation of Ventana announcing a desktop-oriented core or desktop silicon any time soon, and I find their performance projections unimpressive. Tenstorrent, who knows, but it's unlikely. Rivos and Akeana have announced zero public details of what they're building at all.

You're stretching, as usual. How are Tunnelborer and Crane doing? How about the Nintendo "Lady/Lord" console?

 

[Nothingness]

3. https://www.geekbench.com/blog/2023/09/geekbench-62/ && https://www.geekbench.com/blog/2021/03/geekbench-54/
They added support awhile ago: https://browser.geekbench.com/v6/cpu/5004506 Can literally just search risc-v and get risc-v scores: https://browser.geekbench.com/search?q=risc-v

Thanks, I didn't know! So my claim of 0 in 2024 is definitely wrong

 

[igor_kavinski]

That is Apple's CPU Design Team.

Apple is growing its own weed underground under their space campus. That's why their ARM design team is still executing without faltering despite departure of key engineers to Nuvia.

 

[NostaSeronx]

I have zero expectation of Ventana announcing a desktop-oriented core or desktop silicon any time soon, and I find their performance projections unimpressive. Tenstorrent, who knows, but it's unlikely. Rivos and Akeana have announced zero public details of what they're building at all.

data center, edge, and other general computing applications. <=== V2 info
The general computing application is for personal computing. Which I can't find the slide for it but I recreated it for you:


All four of them will be aiming for the server market, then leaning their cores to desktops. It is in the RISC-V international slides on what they are doing.

 

[SarahKerrigan]

data center, edge, and other general computing applications. <=== V2 info
The general computing application is for personal computing. Which I can't find the slide for it but I recreated it for you:
View attachment 94262

All four of them will be aiming for the server market, then leaning their cores to desktops.

Bet you US$200 that there's no Ventana desktop chip by the end of 2026.

 

[igor_kavinski]

3. Why would Primate Labs care about RISC-V at the moment?

Because they must be bored and want a challenge?

 

[igor_kavinski]

https://browser.geekbench.com/search?q=risc-v

These cores working in year 1990 mode?

Those scores sound like you execute a command, lay your head back, dream about your crush, open your eyes and pray that you see output.

 

[Tigerick]

These cores working in year 1990 mode?

Wow, didn't know about Geekbench support RISC-V. And the scores are so low and yet we have people in this forum promoting it...

 

[Tigerick]

In 2025?

It would have to be late 2025 at least given the fastest cadence we have seen from ARM Ltd IP releases to first SoC's using them.

2025 is likely to be full of X5 SoC's - and now that I think about it, I wonder if the reason that Rockchip and Amlogic have been so schtumm on future higher end SKUs is that they are planning to field a really high end SoC for the first time in a while.

When RK3588 was initially announced it was less than a year after A76's IP release so we could potentially see something similar for X5.

Hopefully though it doesn't take quite so long to land in products as RK3588 which was about 1-2 years late to market.

AMD's contract with Microsoft happen around FY2026, so I am not sure which Cortex AMD is using. Anyhow, if NV and AMD are able to replace Qualcomm as SoC choice of upcoming Surface X, then I am expecting better performance or features...

 

[igor_kavinski]

If Nosta is even 10% right about R-V's performance and WinRV, I hope he gets promoted to ARM CPU AT Moderator.

 

[igor_kavinski]

Wow, didn't know about Geekbench support RISC-V. And the scores are so low and yet we have people in this forum promoting it...

They are working in deep stealth mode. When they are finished, they will turn off DEBUG mode and the R-V CPU will run so fast that GB6 will crash. GB7 will be needed and it will use R-V as the base CPU.

 

[Tigerick]

Last time I heard ARM have at least two dedicated CPU design teams at two different places. They are in charge of CPU core design for different year. That's why ARM able to release Cortex X series at yearly basis.

Therefore it is safe for AMD to just license ARM CPU and integrate within upcoming SoC so that AMD could spend resources at upcoming Zen architecture which may take years to accomplish. I heard Qualcomm has delayed the development of Pegasus core by half a year, we don't know what the reason is but compared to ARM with two teams, the risks are sure higher if you rely on one team.

 

[FlameTail]

Anyhow, if NV and AMD are able to replace Qualcomm as SoC choice of upcoming Surface X

Why would that happen?

 

[Tigerick]

Why would that happen?

better performance or features...

 

[FlameTail]

Last time I heard ARM have at least two dedicated CPU design teams at two different places. They are in charge of CPU core design for different year. That's why ARM able to release Cortex X series at yearly basis.

Cambridge Team and Sophia-Antipolis Team?

I heard Qualcomm has delayed the development of Pegasus core by half a year, we don't know what the reason is but compared to ARM with two teams, the risks are sure higher if you rely on one team.

Yeah I was wondering this too. Apparently, Qualcomm is working on an E-core of sorts.

According to leaks, 8G4 has a 2+6 configuration.

2 × Phoenix-L
6 × Phoenix-M

Phoenix-L acts as performance core, and Phoenix-M acts as efficiency core. Beyond the names of the cores itself, we don't know any details.

If Phoenix-M is either a ground up new E-core design or a derivative of the Phoenix-L. I suspect it's the latter. Either way, designing that E-core would have taken time and resources, which may be what delayed Pegasus.

 

[NostaSeronx]

These cores working in year 1990 mode?

Those scores sound like you execute a command, lay your head back, dream about your crush, open your eyes and pray that you see output.

The highest performing core is 2 ALUs + 2FPALUs(pipes shared with Vector units) + 1 BR + 1 AGU aka XT910/C910.
The lowest performing core is a FPGA. The one above that is U74 which is in-order 2-issue.

I can't find anything about the extensions used in GB5/6 so I am going to assume the app is only:
RVA20 Profile => RV64GC

~~~ no end-products(SoC+board) till 2H'24 and beyond ~~~
RVA22 Profile (Mandatory) => RV64GCB
RVA22 Profile (Optional) => RV64GCBV
RVA23 Profile (Mandatory) and RVA23 "Compatible" RVA22 => RV64GCBV <== P670 is this (Compute resources: 208 regs, 4 ALUs, 208 regs, 2 FPALU-FMAs, 128 regs, 2 VALUs)
RVA24 "Compatible" RVA23 => Compare and swap, vector carryless multiply and more

Most of the known High Performance ones are faster than the P670. Ascalon/Aegis-16 to Aegis-128 w/o no AI component is "Zen5" target. Ventana is the strangest one, but is targeting Zenxc products. Rivos is targeting Apple/Nuvia/Qualcomm type SoCs that are HEDT/Server(Mac Pro). Akeana is targeting succeeding SPARC(Sun/Oracle/Fujitsu) cores, alternative to A64FX/Monaka but toned down to laptop power efficency.

 

[eek2121]

If Nosta is even 10% right about R-V's performance and WinRV, I hope he gets promoted to ARM CPU AT Moderator.

🤦

 

[igor_kavinski]

🤦

I see it! Didn't get much sleep today...

 

[Shivansps]

RISC-V is no where near ARM 10 years ago which had no desktop/server demand. RISC-V is more inline with Itanium being server[/desktop] first.

May 29, 2001 - Itanium releases
Windows XP 64-Bit Edition for Itanium System releases -> October 25, 2001
Windows Server Limited Editions came out as early as June 2001.

Instant those 64-core >256-bit LPDDR5/120W TDP and 16-core 192-bit LPDDR5/30W TDP puppies get over here. Microsoft's RISC-V taskforce is going to jump on them like ravenous hungry beasts.

You are right, it is more like 13 years ago, right before the first Raspberry PI that fire up the demand and interest of ARM chips outside phones.

RISC-V first needs to prove itselft in the SBC area, right now the only RISC-V SBCs for general computing that are avalible you only buy it if you are interested in RISC-V, they are more expensive, have less features and are slower that ARM counterparts. Something like the VisionFive 2 is not that much faster than a $15 Zero 2... not to mention 3rd party options with H618 or RK3588 at the $20 price range, vs +75 for the VisionFIve2.

Microsoft is not going to launch a RISC-V build of windows anytime soon, on what SOC they plan people to run it on? the VIsionFive2 SBC? you cant run Windows on that, it would be just as slow as Windows 11 on A55 cores, it takes 10 minutes to open task manager, i tried,

Ive following ARM progress closely, and i consider myself a SBC enthusiant as i have tons of ARM SBC, the progress of making ARM viable for desktop use as been slow, it took time until Windows 10 ran on it and even more time for the x64 emulator to come in, and right now, the RK3588 and the Raspberry PI 5 SOC are the first 2 ARM soc that are generally avalible to consumer that we can say they can drive Windows as a everyday computer, and there still a ton of problems, drivers that no one wants to make, PCIE is a general mess, UEFI implementations are petty much community made, and you want RISC-V to do it next year? There is just no way. Before anything we need an actual RISC-V SBC that we can buy because the price/perf is good, otherwise you dont even have something to try that RISC-V versions of WIndows on, or even compile your programs for, we dont even have an official .NET build for RISC yet.

 

[NostaSeronx]

You are right, it is more like 13 years ago, right before the first Raspberry PI that fire up the demand and interest of ARM chips outside phones.

RISC-V is more inline with AMD64/Itanium. They have no demand and interest heavily leveraging the designs toward phones. Rather, they are purely aimed at Datacenter/Desktops/Laptops/etc.

April 2003 -> Release of AMD64 processors
Image made by September 2003 -> Image available to users by October 2003 -> First AMD64 Windows Beta
April 2005 -> First AMD64 Feature Complete Windows

And, the current Windows is always in BETA.

Microsoft is not going to launch a RISC-V build of windows anytime soon, on what SOC they plan people to run it on? the VIsionFive2 SBC? you cant run Windows on that, it would be just as slow as Windows 11 on A55 cores, it takes 10 minutes to open task manager, i tried,

SG2380 with Milk-V Oasis should be first one to get in by Q4 2024.
SG2044 variant with P670 cores which I think is Q1 2025 should be second. Since the SG2042/SG2044 main market with T-head cores is Domestic Datacenters:

Sophgo has like three cores in running: T-Head(Domestic), SiFive(Global), and their own shared RXU cores(draft below, no Windows target unlike SiFive cores):


P670(RVA22/RVA23/RVA24) chips are the bare minimum chips to support every other architecture(RVA23/RVA24) that comes after.

Ventana Client product is Veyron+Imgtec. Launch point will be after SG2380.

Tenstorrent has agreements to do client products with LG/Samsung: Laptops to "Gaming"/High-end TVs. Which translated to WoRV markets: Laptops to All-in-ones.

Rivos states only high performance RISC-V targeting integrated system solutions for Enterprise. While Akeana is leaning towards more a SiFive model with pre-existing SPARC-leaning architects; IP blocks for CPU cores, interconnects and accelerators. Elaborated further with high end laptop/server cores, and coherent and non-coherent interconnects and accelerators.

SOPHGO/Milk-V hardware availability and Microsoft starting image making process w/ 0x5064 and processors. Means that the above companies can immediately have the largest non-phone Operating System market share. No work on their part to grab exclusivity by 2016 then kill it by never launching a product sustained.

SiFive -> Top 4 Immediate -> Meabwhile, no need for ALA :: AMD swoops in 1:1 device operation between Windows on AMD64 and Windows on RV64. AMD's ARM license from Seattle/Cambridge days expired. So, they will need to operate against SoftBank/Nvidia hierarchy that do not want AMD.

While the low-hanging fruit is this:


"SMTS OPEN SOURCE SOFTWARE PARTNER ENGAGEMENT ENGINEER" <== AMD's "Opensource Solutions Group" is leaning heavily into RISC-V.

 

[FlameTail]

Who came up with the name Milk-V

 

[Nothingness]

RISC-V is more inline with AMD64/Itanium. They have no demand and interest heavily leveraging the designs toward phones. Rather, they are purely aimed at Datacenter/Desktops/Laptops/etc.

You're wrong. RISC-V current market is essentially as a deeply embedded micro-controller.

 

[FlameTail]

Do guys pronounce "RISC-V" as 'RISK-Vee' or 'RISK-Five' ?

 

[igor_kavinski]

Do guys pronounce "RISC-V" as 'RISK-Vee' or 'RISK-Five' ?

Interesting that it was initially an effort from just one university:

What Does RISC-V Stand For? | Semiconductor Engineering, Roddy Urquhart – RISC-V International

riscv.org

 

[NostaSeronx]

You're wrong. RISC-V current market is essentially as a deeply embedded micro-controller.

RISC-V's current market given reservations, pre-orders, long term agreeements say otherwise. Of which, most of the current RISC-V CPU $$$ development is heavily leaned towards Datacenters/Desktops/Laptops.

August 2000 => AMD64 Specification <=== This already happened for RISC-V on Apr 2, 2023 which was ratification day for RVA profiles.
April 2003 => AMD64 CPUs release <=== This hasn't happened yet.
September 2003 => AMD64 Windows image is made <== Microsoft requires Hardware.
October 2003 => AMD64 Windows image is downloadable
April 2005 => AMD64 Windows feature complete

"Availability and Pricing: The journey of the SG2380 SoC has officially begun. Expect the silicon to be at your fingertips in 9 months, and the Milk-V Oasis in just 10 months. An irresistible starting price of $120 awaits post coupon code application." - Oct 20, 2023 // Nine months from this date = July 20, 2024 ... what month/year is now? February going to March 2024.

Will bleed into Ventana's RISC-V, Tenstorrent's RISC-V, Rivos' RISC-V, Akeana's RISC-V. As well as AMD since AMD is a RISC-V member.

SiFive P670 -> Ventana Veyron V2 -> Tenstorrent Ascalon are all expected to be in 2024 with a late window into 1H25.