Thursday, January 5th 2023
AMD Shows Instinct MI300 Exascale APU with 146 Billion Transistors
During its CES 2023 keynote, AMD announced its latest Instinct MI300 APU, a first of its kind in the data center world. Combining the CPU, GPU, and memory elements into a single package eliminates latency imposed by long travel distances of data from CPU to memory and from CPU to GPU throughout the PCIe connector. In addition to solving some latency issues, less power is needed to move the data and provide greater efficiency. The Instinct MI300 features 24 Zen4 cores with simultaneous multi-threading enabled, CDNA3 GPU IP, and 128 GB of HBM3 memory on a single package. The memory bus is 8192-bit wide, providing unified memory access for CPU and GPU cores. CLX 3.0 is also supported, making cache-coherent interconnecting a reality.
The Instinct MI300 APU package is an engineering marvel of its own, with advanced chiplet techniques used. AMD managed to do 3D stacking and has nine 5 nm logic chiplets that are 3D stacked on top of four 6 nm chiplets with HBM surrounding it. All of this makes the transistor count go up to 146 billion, representing the sheer complexity of a such design. For performance figures, AMD provided a comparison to Instinct MI250X GPU. In raw AI performance, the MI300 features an 8x improvement over MI250X, while the performance-per-watt is "reduced" to a 5x increase. While we do not know what benchmark applications were used, there is a probability that some standard benchmarks like MLPerf were used. For availability, AMD targets the end of 2023, when the "El Capitan" exascale supercomputer will arrive using these Instinct MI300 APU accelerators. Pricing is unknown and will be unveiled to enterprise customers first around launch.
The Instinct MI300 APU package is an engineering marvel of its own, with advanced chiplet techniques used. AMD managed to do 3D stacking and has nine 5 nm logic chiplets that are 3D stacked on top of four 6 nm chiplets with HBM surrounding it. All of this makes the transistor count go up to 146 billion, representing the sheer complexity of a such design. For performance figures, AMD provided a comparison to Instinct MI250X GPU. In raw AI performance, the MI300 features an 8x improvement over MI250X, while the performance-per-watt is "reduced" to a 5x increase. While we do not know what benchmark applications were used, there is a probability that some standard benchmarks like MLPerf were used. For availability, AMD targets the end of 2023, when the "El Capitan" exascale supercomputer will arrive using these Instinct MI300 APU accelerators. Pricing is unknown and will be unveiled to enterprise customers first around launch.
44 Comments on AMD Shows Instinct MI300 Exascale APU with 146 Billion Transistors
Cash in equals more cash for rnd.
And I would love to mess with this.
This unironically is the future, and the one Nvidia is most worried about.
And Intel is also working towards.
A 4090 has a die size of 608 mm^2, wheras each chiplet die on a Zen 4 has an area of 70 mm^2. Not to mention all the power delivery circuitry, RAM, PCB and cooling that goes along with producing a GPU.
Of course AMD must keep a hand in the GPU market to stay relevant in that sector and other sectors, but I don't believe it is a big money maker for them.
Nvidia makes 58% of it revenue from Compute & Networking and 42% from Graphics. Graphics include Gaming GPU and streaming, professional Quadro/RTX GPUs and Visualization software, Automotive. So if we remove the non gaming stuff, the revenue just from Gaming is even less, it's about 26% from NVidia Q3 2022 Financial Results. Why would AMD aim for a market that brings only 26% of NVIDIA revenue when it can go for the bigger one?
The PC could be based on a DUAL Processor that supports CPU/GPU without using an external heater core (GPU).
From my vantage point, CPUs have a longer life cycle. Sad, the 5090 is on the way, and the 4090 just showed up.
This is exciting.
Back on the topic of the article, I REALLY want something like this to trickle down to consumer. I know it probably won't because the ability to upgrade components separately trumps the advantages offered by combining CPU/GPU/Memory, but I'd love if AMD offered a desktop APU with 8 Zen4 cores, at least 24-32 RDNA3 CUs, and 8-16GB of HBM2e/3 (to be buttressed by additional DDR5 slots).....man, the SFF builds would be EPIC!!!
According to press release from the government, it has 296 Ponte Vecchio accelerators, and 5120 Sapphire Rapids HBM cores paired with 1.66 PB of RAM (?)
The result was AMD to go from 30-40% to just 8%. Whatever AMD was offering wasn't good enough for people. People where always trying to justify paying for Nvidia hardware and Intel, to justify ignoring AMD's offerings. AMD's offerings where good enough only to force Intel and Nvidia to drop their prices, never good enough to be a real alternative.
That's the mentality of people for at least the last 10-15 years. It just got much worst lately.
So. Why invest a gazillion of dollars in a market where people just care about shiny stickers? The university/country/corporation, or the professional/scientist will care about performance not the sticker. So money invested in staff like MI300 will bring back profits. Money invested in gaming GPUs will just make a hole in AMD's financials. Even if the products are better value than the competition. So?
As an old and current AMD shareholder (and lately also Intel and Nvidia shareholder), I would say ***beep*** gamers and retail GPUs. They want Nvidia? Let them pay what Huang asks them. Keep investing in GPUs for products that will end up in servers, keep investing in GPUs for products that will go in laptops, keep investing in GPUs for APUs and products that will go in handheld gaming devices, keep investing in GPUs for securing the next generation of XBOX and PlayStation. But stop losing money in retail market. Sell directly to OEMs if necessary, as Intel is doing, but IGNORE THE ***BEEP*** retail market. It's a black hole. They are celebrating everytime Nvidia wins, they are celebrating everytime AMD is having a difficulty/bad luck.
Let's see what gamers will get from Huang if Nvidia is the only option in the market. And let's see if GPU hardware reviews will be getting the same numbers of viewers/readers if there is only ridiculously expensive options in the market from just one company.
What I am saying is AMD needs to focus on the market in which there are a lot of possibilities for growth.
If the graphics market is so undesired, then AMD leave it !This is only AMD to blame. They have big partners like Apple which are inclined to use them instead of nvidia but outside they have very little presence. It's AMD's problem in creating the infrastructure and ecosystem for deliveries.
The average joe doesn't know - if you enforce on them AMD Radeon, they will buy it.
Today, they are enforcing nvidia, so yeah, there is no choice.
Don't forget how EVGA (or XFX years ago) which hold around 40% of the US market, simply jumped off ship.
These are the opportunities that AMD must focus in..
Aside from the gramatical error, this sounds pretty awesome. No normal person will ever own one of these, but the fact that AMD is producing this means it's possible and feasible.
Why invest in stuff only to hear "it's your fault I am not buying your stuff"?
Enjoy the Huang Era of gaming. You might need to get a consumer loan first.
PS. Read it. Apple is only a minimal part of the market, especially now they are moving to internal solutions. Other big companies use AMD as a secondary option. Nvidia and Intel are their primary options. Remember how OEMs where sabotaging their own AMD products a few years back? The average Joe will go to Nvidia because everyone sells/promotes Nvidia, no matter what. EVGA is just an AIB. It's not a huge retailer like for example Amazon, to push AMD's products to it's customers. EVGA's sticker value is directly connected with Nvidia's power in the market. EVGA selling AMD cards would had less success than Sapphire for example. For AMD EVGA's contribution would have been small.
PS2. Keep laughing.
I said AMD is to blame because most of the time it doesn't deliver up to the expectations. The users want the fastest. AMD never delivers the fastest.
Actually, if RDNA 1 and before Vega 64, etc garbage, they simply ignored the top of the line market - the halo which sells the whole product lineup beneath it.
:DExactly. AMD must concentrate on working relationships with those "OEM"s.
Or build its own ecosystem without them - direct sales, etc.
That's AMD to blame - because nvidia and Intel concentrate on buying the "OEM", so the result which you see in the stores is 99% Intel-nvidia exclusives.
PS I am expecting a laughing smile, nothing less.
And ... This thing will also require 3D power delivery electronics all around.