Wild_C (@_wildc) header
Wild_C (@_wildc) profile picture

Wild_C

@_wildc

| I mainly talk about tech | *No exclusive leaks* | I occasionally make renders and diagrams | Pronouns: He/Him | TOOT: @wild_c@graphics.social

Joined July 2016
  • Tweets5.1K
  • Following212
  • Followers1.3K
318Media
Photo by Wild_C (@_wildc)Photo by Wild_C (@_wildc)Photo by Wild_C (@_wildc)Photo by Wild_C (@_wildc)Photo by Wild_C (@_wildc)Photo by Wild_C (@_wildc)Photo by Wild_C (@_wildc)Photo by Wild_C (@_wildc)Photo by Wild_C (@_wildc)Photo by Wild_C (@_wildc)
The core counts are odd. Guess: "Ryzen AI 7 550" - Full Medusa 1 0CCD 10 cores: 4 Classic + 4 Dense + 2 LP 16MB L3 "Ryzen AI 5 540" - Full Medusa 2 8 cores: 2 Classic + 4 Dense + 2 LP 8MB L3 "Ryzen AI 5 530" - Cut Medusa 2 / Full 3(?) 6 cores: 1 Classic + 3 Dense + 2 LP 4MB L3
VideoCardz.com (@VideoCardz) profile pictureVideoCardz.com
@VideoCardzNov 6
AMD Zen 6 Medusa Point 2/3 spotted with 6 and 8 cores in Lenovo’s Geekbench leak 🔗videocardz.com/newz/amd-zen-…
Image from VideoCardz.com (@VideoCardz)
1
2
314
GB7 is probably grouping Dense and LP cores into 1 cluster, explaining why we see 4+6, 2+6, and 1+5 MDS1 "550" ≈ KRK1 350 / GRG2 450 config MDS2 "540" ≈ KRK2 345 / GRG3 435/445 config MDS2/3 "530" is similar to KRK2 330 / GRG3 430, but w/ halved L3
1
76
Wild_C retweeted
Jaykihn (@jaykihn0) profile picture
Jaykihn@jaykihn0Oct 17
Replying to @_wildc
Razor Lake -AX uses NPU7 with 6 tiles. Nova Lake and Razor Lake non-AX use NPU6 with 4 tiles.
2
3
32
986
Of course, I'm making some assumptions about clock speeds and the NPU being the same as was leaked for Nova Lake's NPU 6, but even if I'm wildly wrong, it doesn't change the overall picture 4/5
2
3
584
2X GPU matrix throughput is the only way I see to get to (a likely conservative!) 700 TOPS Otherwise there's a ~300 TOPS gap The only other solution I can come up with is something like a 12 NCE NPU at high clocks (4X Nova Lake-H). Not impossible, but very unlikely imo 5/5
2
231
Correct me if I'm wrong, but I think this suggests that the version of Xe3P used in Razor Lake-AX has 8-deep systolic arrays This means that each Xe Core doubles matrix throughput vs Xe3 Note: Xe3P in Crescent Island uses 16-deep systolic arrays 1/5
VideoCardz.com (@VideoCardz) profile pictureVideoCardz.com
@VideoCardzOct 17
Intel confirms Razor Lake-AX for 2028 with 700 TOPS AI performance 🔗videocardz.com/newz/intel-co…
Image from VideoCardz.com (@VideoCardz)
1
15
1,390
more replies
For Razor Lake-AX32Xe: CPU: (8P * ~5.1GHz + 16E * ~3.8GHz + 4LPE * ~3.7GHz) * 512/8 FMA/c * 2 = ~15 TOPS GPU: 32XC * ~2.32GHz * 4K/8 FMA/c * 8 deep Systolic Array * 2 = ~610 TOPS NPU: 3NCE * ~1.51GHz * 64K/8 MFMA/c * 2 = ~74 TOPS Total ~700 TOPS 3/5
1
1
1
2
1,004
The Math: For Panther Lake-H12Xe: CPU: (4P * 5.1GHz + 8E * 3.8GHz + 4LPE * 3.7GHz) * 512/8 FMA/c * 2 = 8.4 TOPS GPU: 12XC * 2.5GHz * 4K/8 FMA/c * 4 deep Systolic Array * 2 = 122.9 TOPS NPU: 3NCE * ~2.05GHz * 32K/8 MFMA/c * 2 = ~50 TOPS Total a little over 180 TOPS 2/5
1
1
313
Wild_C retweeted
Jaykihn (@jaykihn0) profile picture
Jaykihn@jaykihn0Oct 16
Nova Lake Mobile Die Sizes Figures are rounded from higher precision. Thank you @mooreslawisdead for the package shots.
Image from Jaykihn (@jaykihn0)
Moore's Law Is Dead (@mooreslawisdead) profile pictureMoore's Law Is Dead
@mooreslawisdeadOct 16
Want to see annotated Die Shots of #Intel Nova Lake, and see the specs of their #AMD Medusa Halo Mini Killer? Well, check out my latest video!🤠 Oh, and I also analyze #Nvidia's N1X APU that just launched, and leak how easy it will be to find one... youtu.be/VR4pGjq1drY
Intel Nova Lake Die Shots Leak | Nvidia RTX Spark N1X Analysis

Intel Nova Lake Die Shots Leak | Nvidia RTX Spark N1X Analysis

See first leaked pics of Nova Lake, and also see how much N1X fails...

youtube.com
7
14
2
101
10,904
I kinda forgot to post this Snapdragon X2 Elite Extreme—Glymur die
Image from Reptalica (@Reptalicant)
Fried egg galaxy (@7742_friedegg) profile pictureFried egg galaxy@7742_friedeggSep 13
Piglin uploaded Snapdragon X2 Elite Extreme Dieshot!! @Reptalicant tieba.baidu.com/home/main?id…
Image from Fried egg galaxy (@7742_friedegg)
2
27
228
11,395
RDNA 5 (or other naming for the next gen graphics IP) Confirmed
Gotou_3rd (@Gotou_kai3) profile pictureGotou_3rd
@Gotou_kai3Sep 10
Medusa Halo is back….
2
2
26
3,311
Huh so they're going the Apple route of not naming the SKUs Thanks, I hate it
VideoCardz.com (@VideoCardz) profile pictureVideoCardz.com
@VideoCardzSep 3
NVIDIA confirms RTX Spark N1X 6144 and 5120 CUDA core specs, launching next month videocardz.com/newz/nvidia-c…
Image from VideoCardz.com (@VideoCardz)
4
518
Do we know why Intel is using UCIe-S for Diamond Rapids? Is it that EMIB can't make packages that big? Is it cost? Is it that yield makes such a complex product non-viable? Is it that EMIB isn't very good and UCIe-S is just as efficient?
7
1
28
2,979
Wild_C retweeted
HiEQ (@HiEq_2005) profile picture
HiEQ@HiEq_2005Aug 13
Image from HiEQ (@HiEq_2005)
Image from HiEQ (@HiEq_2005)
3
6
37
2,778
Starfire = Panther Lake-R (Rugged) ?
VideoCardz.com (@VideoCardz) profile pictureVideoCardz.com
@VideoCardzJul 13
Intel unveils Starfire space processor with 8 CPU cores and 4 Xe GPU cores videocardz.com/newz/intel-un…
Image from VideoCardz.com (@VideoCardz)
6
511
Intel3 Xe3んとこ
Image from テカナリエ清水 (@techanalye1)
1
24
1,118
So now the cleanly delineated 10/100/200 series has more ambiguity than the 7000 series AMD try not to have the most annoying naming scheme possible challenge: IMPOSSIBLE
VideoCardz.com (@VideoCardz) profile pictureVideoCardz.com
@VideoCardzJul 9
AMD expands Ryzen 200 and Ryzen 100 with 11 Hawk Point SKUs videocardz.com/newz/amd-expa…
Image from VideoCardz.com (@VideoCardz)
4
437
Well that wasn't what I ordered, I wanted PonteVecchio. Turns out some of the Arctic Sound HP tiled chips made it into engineering samples that got into the wild. Never thought I'd see one of these after they got memory holed.
Image from Chips by their layers (@ChipsByLayers)
Image from Chips by their layers (@ChipsByLayers)
7
14
3
127
14,187
Intel: DMR, Crescent island, WCL AMD: FPGAs, MI400 series Nvidia: Vera, Rubin, Spectrum, bluefield and VPU Google, Meta, Microsoft: TPUv8, MTAI 400, MAIA 200 Arm: AGI CPU Fujitsu: Monaka CPU Samsung: LP5X-PIM Sambanova: SN50 Broadcom: Thor ultra IBM: Z and Inference chip
Bionic_Squash (@SquashBionic) profile pictureBionic_Squash@SquashBionicMay 28
Damn this year's hotchips is packed, so excited for this one
6
19
128
14,012
Intel Arrow Lake dieshot annotated shot by : b23.tv/H2drEFh
Image from Reptalica (@Reptalicant)
5
21
171
9,333
Load more