Discussion Future ARM Cortex + Neoverse µArchs Discussion

[SarahKerrigan]

Server IP's have generally lagged consumer IP's. I don't think it's X5.

 

[soresu]

Server IP's have generally lagged consumer IP's. I don't think it's X5.

Hoping you are right 🫰

 

[FlameTail]

Let's get a Cortex X5 hype train going.

I'll throw in the first dibs:

30% IPC gain and 30% performance per watt improvement.

 

[FlameTail]

https://x.com/FelixCLC_/status/1760341406441181325?s=20

Microsoft's next Azure Cloud chip uses Neoverse N3 it seems.

 

[Nothingness]

I also thinks V3 is based on X4: both are ARMv9.2 while V2, based on X3, was ARMv9.0. And I guess X5 will go beyond

Let's get a Cortex X5 hype train going.

I'll throw in the first dibs:

30% IPC gain and 30% performance per watt improvement.

30% IPC or 30% performance?

 

[Tigerick]

View attachment 94242

D9400 score 2700 @ GB6 ST showing improvement around 20% better than Cortex X4.

It seems 8 Gen4 is using Oryon core NOT Cortex-X5, so far there is leak about 2850 ST scores, better than D9400..

 

[FlameTail]

D9400 score 2700 @ GB6 ST showing improvement around 20% better than Cortex X4.

20% isn't revolutionary. If this is X5, it no Custom-core killer.

The salt pill is that those D9400 are prolly from early engineering samples, so final product might score better.

It seems 8 Gen4 is using Oryon core NOT Cortex-X5, so far there is leak about 2850 ST score

Yep. Oryon Phoenix-L @4.0 GHz it seems.

 

[FlameTail]

Nvidia today just added almost $300 billion to their stock, thanks to the AI Boom.

What if.... What if Nvidia had succeeded in buying ARM? What difference would it have made today?

 

[igor_kavinski]

What difference would it have made today?

Accelerated the adoption of RISC-V. Jensen wastes no time in milking his customers. Think royalties per chip sold. License only if you will make a minimum number of millions of chips. Etc. Etc. Custom core? You better have deep pockets to be able to do that. And pay even more royalties per chip.

 

[poke01]

What if Nvidia had succeeded in buying ARM?

I would rather ARM be owned by SoftBank. It’s no longer possible to buy it anyway. Nvidia can try and make own chips and couple it with their own GPU IP instead

 

[Saylick]

Nvidia today just added almost $300 billion to their stock, thanks to the AI Boom.

What if.... What if Nvidia had succeeded in buying ARM? What difference would it have made today?

I mean, Nvidia announced they want to go into semi-custom. I think the conclusions are self-evident if they were successful in acquiring ARM: you know, just more lasso'ing of companies into the Nvidia ecosystem. If people don't like Apple, wait until they get a load of Nvidia.

 

[Shivansps]

I mean, Nvidia announced they want to go into semi-custom. I think the conclusions are self-evident if they were successful in acquiring ARM: you know, just more lasso'ing of companies into the Nvidia ecosystem. If people don't like Apple, wait until they get a load of Nvidia.

If Nvidia would have brought ARM, people would have tried to jump ship to RISCV ASAP. And ARM would have become like x86. At least for future IPs. Even MIPS would have had a chance of coming back to life. Ill say that Nvidia was the responsible of the sudden peak of interest in RISC-V, or at least one of them. China trying to become independient from the world is another one.

 

[DavidC1]

If Nvidia would have brought ARM, people would have tried to jump ship to RISCV ASAP. And ARM would have become like x86.

It took many years and a revolutionary product like the iPhone for ARM. What's on the horizon that'll make RISC-V like that?

They won't switch to new ISA overnight. They could try, but it'll end up like the Itanium.

 

[igor_kavinski]

What's on the horizon that'll make RISC-V like that?

The Aphone powered by the Tenstorrent Sonic Boom CPU!

 

[soresu]

It took many years and a revolutionary product like the iPhone for ARM

ARM was a thing long before iPhone.

It was just that they shifted focus in core development significantly after that point.

 

[Shivansps]

It took many years and a revolutionary product like the iPhone for ARM. What's on the horizon that'll make RISC-V like that?

They won't switch to new ISA overnight. They could try, but it'll end up like the Itanium.

They would have not switched ISA for fun, thats clear. But Nvidia was not trying to buy ARM for fun either, they were trying to secure an arch for their future processors.

 

[FlameTail]

Another company will start making SoCs for their smartphones.

Xiaomi is working with ARM to make its own chip: Report

Xiaomi is reportedly working on its own ARM based chip suggests a new Counterpoint Research report. Learn more here.

www.gizmochina.com

XIAOMI

 

[naukkis]

It took many years and a revolutionary product like the iPhone for ARM. What's on the horizon that'll make RISC-V like that?

They won't switch to new ISA overnight. They could try, but it'll end up like the Itanium.

Itanium failed because overly complex ISA didn't made possible to make competitive hardware. Risc-V doesn't have that problem - pretty much same hardware can run ARM and risc-V.

 

[Nothingness]

Itanium failed because overly complex ISA didn't made possible to make competitive hardware. Risc-V doesn't have that problem - pretty much same hardware can run ARM and risc-V.

Itanium also failed because Intel thought it was a great idea to rely on compiler improvements to extract performance from a brain dead ISA.

 

[naukkis]

Itanium also failed because Intel thought it was a great idea to rely on compiler improvements to extract performance from a brain dead ISA.

They made ISA to implement everything compiler could do to made their core forward-proof without recompilation and to make execution hardware as simple as possible - strictly in-order-design. They failed in both - code was not optimized for future without recompiling and executing hardware wasn't simple but much more complex than competing out-of-order designs.

 

[igor_kavinski]

It's funny thinking that companies can put their weight behind ideas that may look like a disaster waiting to happen from the start. I think the meetings they have, some engineers are bored and want to get out to meet their kids or wife or mistress, others disagree with the idea but don't know if anyone else disagrees so keep quiet rather than face opposition alone in front of bigwigs and the very knowledgeable ones who don't care because the bigwigs promoted the fool talking about the crappy design and passed them over for promotion so they really want to enjoy watching the train wreck in all its glory. And those Chief Architects that scarred the company with their designs? Probably retired now and enjoying the view of their acres of property and living the good life.

 

[naukkis]

It's funny thinking that companies can put their weight behind ideas that may look like a disaster waiting to happen from the start. I think the meetings they have, some engineers are bored and want to get out to meet their kids or wife or mistress, others disagree with the idea but don't know if anyone else disagrees so keep quiet rather than face opposition alone in front of bigwigs and the very knowledgeable ones who don't care because the bigwigs promoted the fool talking about the crappy design and passed them over for promotion so they really want to enjoy watching the train wreck in all its glory. And those Chief Architects that scarred the company with their designs? Probably retired now and enjoying the view of their acres of property and living the good life.

Basic principle should be KISS - keep it simple stupid. At Intel they time after time develop exact opposites of that - IAPX432, Itanium and so on, which is exactly opposite. Probably there's no-one really on top of design so they put everything ever invented in designs list and try to implement that - and of course are destined to fail miserably but only after shoveling truckloads of money into project. Risc-V is fresh air to ISA wars - it is designed pretty much to be as simple as possible to give hardware designers free hands to implement whatever they want.

 

[igor_kavinski]

IAPX432

Very interesting. So that was Gordon Moore's failed project, no less! What other famous failed projects are there? Any not mentioned here: https://www.extremetech.com/computing/274650-the-worst-cpus-ever-made

 

[DavidC1]

Itanium comparison might be extreme, but in the long run, ARM has a huge established software base. RISC-V won't replace ARM until something revolutionary happens like Smartphones in the computing space.

 

[naukkis]

Itanium comparison might be extreme, but in the long run, ARM has a huge established software base. RISC-V won't replace ARM until something revolutionary happens like Smartphones in the computing space.

Legacy support ain't so important nowadays - it's pretty much enough that there is emulation layer to support existing software. Google is building risc-V version from Android and when it's ready and there's competitive hardware RV will be pretty much viable alternative for ARM for phones. Though we are not there yet, lacking both software and hardware support but it sure looks that things might change pretty fast - still years away though.