I think if you want something more than a 16 core ryzen you can enter the thread ripper market. I certainly think higher core ryzen chips are going to come at some point but I just think its the 4core and maybe 6 core chips will disappear first
AMD Execution Thread [2024]
- Thread starter del42sa
- Start date
Intel’s 4-core chips are still pretty popular for a good reason. For general purpose computing you don’t need more than that. I hope they keep making small power efficient cpus.
This is the same conversation that always happens. X is still popular and for some reason that always has to be true but never is lol. I remember hearing this with single core cpus and then dual core and now we are at quad cores and even some people saying 6 core cpus are the sweet spot.
Bondrewd
Veteran
No, he means the big chungus premium APU.
That's really not desktop stuff, you can do dGPU on DT just fine.
in laptop, oh yes, APUs have very real benefits.
even the 'fanless' LNL is 8c rofl.
Quads are very-very dead.
The 2025 'low-end' KRK1 chop is 2+4, after all.
ah but it seems like if you want graphical power its much easier to just buy a gpu. There is only so much gpu you cna fit on a cpu
You can have a nice and quiet 8 /16/32/64 core htpc also.
What is funny is that no matter where we are on the time line people continue to say that x amount of cores is more than enough. Would you put a single core cpu in your htpc ? what about a dual core?
it seems like Microsoft are ruining AMD CPUs on laptops. Microsoft forced AMD to remove the SLC (System Level Cache), a key Component from Its Zen 5 Laptop CPUs.
forums.anandtech.com
Page 350 - Discussion - Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)
Page 350 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
forums.anandtech.com
Frenetic Pony
Veteran
It's not really about "more CPU cores". More about a giant APU. I can easily see AMD putting giant APUs on desktop one way or another, even if it's just higher TDP binned parts meant for larger cooling systems. Apple charges $2000+ for a Mac Studio with all of 12 CPU cores. If Microsoft wanted to try similar I can see AMD happy to provide such an APU.
I’m struggling to understand why this is relevant. Clearly today in 2024, 4 cores is enough.
Yes I remember.
Bondrewd
Veteran
This is for premium laptops, where emasculated console APU very much works.
It's not that they forced, just that the initial two Strix RFPs had a 16meg MALL, which was cut in the third one for cost reasons (and needing a bigger AIE. blergh).
Mind that the OG Strix from like 2021 was tiled. That client-specific CCD with USRs was intended to be used across the lineup. Now, -halo is the only survivor.
Not happening outside of NUCs.
DT can do with just dGPUs.
MS is retarded, and obsessed with WoA.
-halo is for other, less retarded, OEMs.
Yes, let us all be asking people riding horses if they want automobiles or not.
That may not be too far off the series s performance wise . the S is 20CU @ 1.57 ghz. They'd still be missing out on the bandwidth however
What role do you see local AI inference having?
The way I see it, all the interesting near future AI products will run on models that are way too big to fit on a smartphone or a laptop. They will run on huge inference server farms instead. This also makes more economic sense, because batching means that you can more effectively utilize the memory bandwidth when you are doing inference for more problems at the same time. The cost of doing centralized inference for a million clients is dramatically lower than those million clients each doing their own inference at equivalent quality.
My phone does local AI photo processing right now, no need to wait for near future. Laptops are running DLSS and XeSS right now too.
But it all depends on what is "interesting" to you I suppose.
I'd expect the vast majority of "interesting" AI applications to run 100% locally because of latency and privacy concerns.
The rest which will be able to run remotely in DC are likely what we have running already. They may improve or be used in new ways but it won't be revolutionary.
Bondrewd
Veteran
none of the tiny ML model stuff is interesting outside of "it's a party trick which I used exactly twice".
who the fuck gonna source the membw and DRAM capacity?
Like Apple still ships 8GB laptops for as high as 1.6kilobucks.
Similar threads
