Phoronix has some benchmarks too. Decent improvment... The top model is close to the 64 core Genoa but still chugs way more power.
makes sense, per core perf isn't a problem for Intel this gen.
Phoronix has some benchmarks too. Decent improvment... The top model is close to the 64 core Genoa but still chugs way more power.
Intel castrated SPR so much that its Golden Cove cores were not able to strech its legs, IPC Wise Golden Cove/Raptor Cove and Zen4 are very close to each other. But AMD Has Node advantage so they can pack way more cores so Intel will not be able to match them at the Top end and their price should reflect that but that CPU is mord expensive than the 9654.makes sense, per core perf isn't a problem for Intel this gen.
With increasing demand for generative AI solutions, Intel expects to capture a larger portion of the accelerator market in 2024 with its suite of AI accelerators led by Gaudi.
where did you see this?It seems that Gaudi3 has 8 stacks hbm3(e).
I'm going to go way out on a limb and say that lightisgood counted the chips on the image that was specifically attached.where did you see this?
Bit of an interesting one that, given that Granite Rapids ups the core count to 120 (for GNR-AP). Actually implies that there's less L3 per core - a reduction from 5MB L3 per core to 4MB L3 per core.
There is a 2 chiplet (80 cores) version of GNR as well. If the 1.5x number refers to this configuration (do we know?), then it would be an increase in L3 per core.Bit of an interesting one that, given that Granite Rapids ups the core count to 120 (for GNR-AP). Actually implies that there's less L3 per core - a reduction from 5MB L3 per core to 4MB L3 per core.
That's some wortzusammensetzung.Clearwaterforest
The big jump should mostly be in Skymont, and Darkmont an 18A version of it. It would be a rabbit out of the hat surprise otherwise, since Darkmont on client is just a year after Skymont.There is news floating around that Clearwater forest has Darkmont cores on 18A. That should be a big jump from Crestmont plus in Sierra Forest even with same core count.
No, for the short term and for the forseeable future, getting a competitive DC CPU is more critical, because Intel already has a big presence, similar to Nvidia having big presence in DC GPU.I think getting a competitive DC GPU part is more critical as Cloud/Enterprise spend is moving more towards there. Intel dropped the ball big time on it. Falcon Shores should have come this year. Almost 2 year delay wont help them for sure.
The Core Ultra platform delivers leadership AI performance today with our next-generation platforms launching later this year, Lunar Lake and Arrow Lake tripling our AI performance. In 2025 with Panther Lake, we will grow AI performance up to an additional 2x.
This maybe? https://www.reuters.com/article/idUSKBN1YK1BU/Who or what is Intel tapping for their NPU designs?
Habana (Goya/Gaudi) is one of the possible options.
That answers that. And I find that interesting. Also it appears that Intel refers to their NPU as a VPU, I guess?The NPU is based of the movidius architecture, a company they acquired years ago.
VPU is old term and has been merged and called NPU now...That answers that. And I find that interesting. Also it appears that Intel refers to their NPU as a VPU, I guess?
Hmm okay. Wait wasn't Movidius the self-driving car company they bought out? That's hilarious.VPU is old term and has been merged and called NPU now...
No, that’s MobileyeHmm okay. Wait wasn't Movidius the self-driving car company they bought out? That's hilarious.
Looks like there was some talk of Movidius tech finding its way into self-driving cars and/or drones:No, that’s Mobileye
Yeah suppose it could’ve been used in cars but wasn’t Intels main aim AFAIK. It was used in drones quite a bit, also sent to space, in USB sticks, Keem bay to compete against nvidia in inferencing iirc but not sure it ever launchedLooks like there was some talk of Movidius tech finding its way into self-driving cars and/or drones: