Intel Nova Lake increasing core counts

[LukeTbk]

The threading would become more important once the PS4 and XBone were both released by 2013 with 8 cores/thread,

Xbox 360 having a 6 thread 3 cores processor was not that far from that (I think Ubisoft montreal before they got their hands on dev kit were making multiCpu Xeon machine), PS3 with 7 or 8 core CELL processor really do not know what was going with that.

Back in the 2005 days, the talk around those Xbox was that The other important thing to keep in mind here is that porting between multi-core PC platforms and the Xbox 360 will be fairly trivial.

I think one general issue is mistaken or talking about core versus multithreaded performance, even thought those console cpu had a lot of threads since the Xbox360 days, they did not had a lot of cpu multithreaded performance versus PC cpu gamers use and you want the most multithreadperformance on the least core usually (40k passmark on a 4core>40k on a 16 core for games usually, easier to use all the performance)

 

[Zarathustra[H]]

It is not 2010, and increasing cores improves performance in nearly all workloads at this point.


AMD was about 5 years too early back in 2011 with the debut of Bulldozer and their CMT architecture.
At the time the integer performance was about 55% clock-for-clock what Sandy Bridge was, and the FX 8-core CPUs were the only ones with 4 FPUs.

The threading would become more important once the PS4 and XBone were both released by 2013 with 8 cores/thread, and games and software started becoming more optimized with more than 4 cores/threads.
AMD was too early with CMT's larger thread counts for software to be optimized for it at the time, and Meltdown/Spectre/Foreshadow/etc./etc./etc. hadn't become public knowledge yet giving Intel an artificial boost with their now-known garbage CPU security implementations.

Had nearly all software been optimized for 6-8 threads (like it has been for the last decade) and had Intel actually implemented the proper security designs in their CPU architectures, the performance would have been much closer.
I'm not going to white knight for AMD, though at the same time lets not rewrite history.

I'd challenge your assertion that "nearly all" workloads benefit from more cores.

This just isn't theoretically possible (unless you are talking about performance benefits when running many different things at the same time, which is different than multi-threading)

Per computer science research it is only possible to multithread a minority of code. And even with code that can be multithreaded without thread locks and crashes, Amdahls law puts a pretty clear limit on overall performance benefits (which are pretty low). What you are suggesting is essentially not theoretically possible, with any amount of effort or any amount of time by programmers with any amount of skill, on any computer today or on any computer at any time in the future.

(Yes, I am aware of some people challenging Amdahl's law, and showing greater performance increases than it predicts, but in all of those cases, that just comes down to the benefits of greater caches when dealing with more cores, not due to multithreading itself)

Programming is essentially logic, and that simply violates basic logical structures.


And it checks out in real life too.

For the basic desktop tasks I do on them, my 24C/48T Threadripper 3960x is no faster than the little all e-core 4C/4T Intel Processor N97 in my Odroid H4.

There are some workloads that benefit greatly from parallelism, but the vast majority of them do not. That was not just a fact in 2010, it is also a fact today, and as long as there are computers it will be a fact. it is a logic limitation, not an architecture or programming limitation, and there is nothing anyone can ever do to change it.

 

[DukenukemX]

Let me know when they boost IPC because core count doesn't matter unless you run a server. Most modern CPU's have cores for days.

 

[Balkroth]

IPC/W (that's really a dumb thing anyway, ipc, as there seems to be no standard) is gonna be the thing most people are focused on honestly, as there are limited gains coming up soon, and it will be per application based.

 

[Red Falcon]

For the basic desktop tasks I do on them, my 24C/48T Threadripper 3960x is no faster than the little all e-core 4C/4T Intel Processor N97 in my Odroid H4.

"Basic desktop tasks" are not the same as running numerous applications in enterprise, and I'm not even talking about HPC or compute.
You must be running some real basic programs or scripts if your 3960X is no faster than an N97.

Try loading up Microsoft Teams on that N97 and compare it to your 3960X, it will be night and day.
Yeah, Microsoft Teams, hardly a "powerhouse" of a productivity program.

Again, it isn't 2010, and your "basic desktop tasks" do not match what the rest of the world is doing, especially in enterprise and even for most consumers beyond using tablets and phones - of which those are still heavily multi-threaded.
Even smartphones with 8-core SoCs aren't doing so because apps are all single-threaded, the world has moved on and your opinions on it are seriously outdated by about 15 years.

There are some workloads that benefit greatly from parallelism, but the vast majority of them do not. That was not just a fact in 2010, it is also a fact today, and as long as there are computers it will be a fact.

Yeah, I get that, my point was that in 2010 only a few mainstream programs for home or enterprise were starting to use more than a few threads.
Today in 2025 that absolutely isn't so, and even basic applications will utilize 8+ threads easily.

it is a logic limitation, not an architecture or programming limitation, and there is nothing anyone can ever do to change it.

Technology and programming have changed.
While I do agree with the basic concept, the fact that so many day-to-day applications (even mobile apps) utilizing heavily multi-threaded capabilities are only expanding, and this has been old hat since 2015 at least.


You know, I'm starting to remember you stating in other threads that you needed lots of PCIe lanes, but not necessary lots of CPU and that you could probably get away with a C2D if it had enough PCIe lanes.
Your workflows are a bit of an outlier where you need lots of I/O but little CPU, and that is not indicative of daily operations for nearly anyone else at this point.

I had a coworker with very similar requirements that you have, and it makes me laugh when he complained that he couldn't understand why anyone needs a SSD when his workflows are just fine on standard SATA HDDs with his light-weight Linux distros, scripts, and light-weight programs.
Like, that's great you can get away with using HDDs with your operating systems and tiny programs... meanwhile the rest of the world has moved on with NVMe, PCM, flash, eMMC, etc.

It is starting to feel that same way with everything you post about.
I'm glad you can get away with using an Intel N97 on a SBC for your workflows, but many of us would not be able to do so for our very modern workflows in the 2020s.

 

[DukenukemX]

IPC/W (that's really a dumb thing anyway, ipc, as there seems to be no standard) is gonna be the thing most people are focused on honestly, as there are limited gains coming up soon, and it will be per application based.

IPC does still matter, but this is mostly towards gaming. You could argue that it matters for day to day tasks but realistically your CPU will under clock and even move tasks to lower performing but energy efficient CPU cores, and people still think their computer is "snappy". This has been the case for a while now. This is why I argue that IPC is mostly for gaming since gaming is what benefits it the most. This is why the 9800X3D is such a sought after CPU while the 9950X can easily be found. The 9950X costs the same as a 9800X3D while still being available. But yea, we have terrible ways of testing IPC. Phoronix tests shows that Intel's 285K has really good IPC, but in gaming it's a dog. Same goes for Apple as Geekbench and Cinebench R24 show M4 is the king of the hill in IPC while it's also a dog in gaming. This doesn't make sense.

Yeah, I get that, my point was that in 2010 only a few mainstream programs for home or enterprise were starting to use more than a few threads.
Today in 2025 that absolutely isn't so, and even basic applications will utilize 8+ threads easily.

This right here. We're at a point where applications have finally caught up to modern CPU design to the point where having more cores is beneficial. But I'd also argue that we have a lot of cores in modern CPU's. We're far past the days of Intel's quad core hell. The Intel 265F has 20 cores, with eight of them being performance cores, and it's only $340. The crazy thing is this CPU is mostly sold out. Amazon doesn't have any in stock and neither does B&H. Either Intel didn't have a lot in stock or there's a lot of demand for these chips. Intel needs to worry about IPC and efficiency. I think Intel might want to walk back on removing hyper-threading, since SMT is working very well for AMD. You don't see AMD having efficiency and performance issues with SMT. AMD on the other hand seems to be really comfortable with eight cores, as most of their desktop CPU's seem to be a variant of eight cores. They're really good eight core designs, but Intel's 265F shows that there's a demand for more cores for less money.

 

[LukeTbk]

Today in 2025 that absolutely isn't so, and even basic applications will utilize 8+ threads easily.

Which is different than being able to scale over core well or needing more than a 8 core / 16 threads cpu, maybe people could be talking over each other, the conversation was not about is having more than one core great value for regular user in 2025, that certain, but since the 8/16 bar was reached, how much was gained for regular folks for the core-threads count to continue to grow.

The basic Apple m4 come with 8 cores, no SMT, my feeling that it will do what regular people tend to do extremely well, better than a 5950x would.

Windows 98 had more than 8 threads running back in the days, the average users could have had 30-50 of them running on their single core-single thread cpu.

Just launched teams and in process explorer a see 156 threads running for its main process alone, many of the childs process has over 40.., even if heavily correlated thread count and able to take advantage/needing more core is still a little bit different, than how many threads they use.

My simple passwordmanager app use over 100 threads for its gui it seems.

 

[Lakados]

I’ve been coming across a lot of papers lately about how OS design is changing and how application deployment needs to change with it.

Lots of academic papers about code compiling to an intimidate binary that can then be executed by a platform specific compiler, code once run anywhere design.

I have also seen others about extreme sandboxing where each application and possibly services are self contained and operate within the OS as their own Virtual Machine of sorts.

The last interesting game changer was the use of signing keys for those programs and services. Something they were calling content based addressing with public keys.

How is any of this relevant?
For them to work CPU core counts need to increase, and so does memory availability if it’s ever going to be viable on consumer devices. IPC improvements help and is never a bad thing but in such an environment raw core count is more beneficial until you hit a point of diminishing returns.

Is it something we will ever see at a consumer level, who knows, but I suspect for business and enterprise the security improvements would be welcome given the costs involved with securing the current systems.

 

[LukeTbk]

Lots of academic papers about code compiling to an intimidate binary that can then be executed by a platform specific compiler, code once run anywhere design.

Will they call it Java 2.0 ?

I have also seen others about extreme sandboxing where each application and possibly services are self contained and operate within the OS as their own Virtual Machine of sorts.

Will they call it iOS ? Or Windows 8 store apps redo, that sound a bit like what UWP-WinRT was doing back then or Xbox do for the game running env vs the xbox os env since the Xbox One.

For them to work CPU core counts need to increase

OS on small phone do that, they had 1-2 core for a long time, what would be the distinction made between core count need to increase and multi-thread performance need to augment (and does it really need to, a 7900x is a lot of cpu already) ?

9700x passmark MT: 37,201 (16 threads)
3950x passmark MT: 38,681 (32 threads)
m4pro passmark MT: 38,368 (14 threads)

We have the 16 core 32T multithreading performance of yesterday with half the running threads now.

 

[Lakados]

Will they call it iOS ? Or Windows 8 store apps redo, that sound a bit like what UWP-WinRT was doing back then or Xbox do for the game running env vs the xbox os env since the Xbox One.

It’s more inline with how supercomputers have their OS’s designed around running multiple super lightweight kernels focusing on massive parallelism.
The majority of the papers were focusing around Linux.
I get the impression it’s based on the design principles being forced there.

The CPU does little more than serve as a coordinator for PCIE traffic and memory management.
Like what if your home CPU did just that which would favour ARM and RISC-V architectures greatly then any and all compute be it jobs normally done by the CPU or the GPU were handed off to accelerator cards. So x86 ceases to be the main CPU but an Accelerator Card, GPU functionality already essential is an accelerator card, but then you could add others as needed. It favours a scale out approach to system design.

It’s very interesting but would be a fundamental change in how consumer hardware is designed and consumer software deployed.

 

[uOpt]

This is relevant to my interests. I have an actual use for so many cores.

But buying the platform will be problematic since ECC memory will not be on consumer boards and we'll have to wait endlessly for boards with their "workstation" chipsets. Limited selection, too, and high price.

 

[Balkroth]

Not gonna lie, I'm surprised we haven't seen each service as a container in consumer stuff already.