Tuesday, February 22nd 2022

Intel Details Ponte Vecchio Accelerator: 63 Tiles, 600 Watt TDP, and Lots of Bandwidth

During the International Solid-State Circuits Conference (ISSCC) 2022, Intel gave us a more significant look at its upcoming Ponte Vecchio HPC accelerator and how it operates. So far, Intel convinced us that the company created Ponte Vecchio out of 47 tiles glued together in one package. However, the ISSCC presentation shows that the accelerator is structured rather interestingly. There are 63 tiles in total, where 16 are reserved for compute, eight are used for RAMBO cache, two are Foveros base tiles, two represent Xe-Link tiles, eight are HBM2E tiles, and EMIB connection takes up 11 tiles. This totals for about 47 tiles. However, an additional 16 thermal tiles used in Ponte Vecchio regulate the massive TDP output of this accelerator.

What is interesting is that Intel gave away details of the RAMBO cache. This novel SRAM technology uses four banks of 3.75 MB groups total of 15 MB per tile. They are connected to the fabric at 1.3 TB/s connection per chip. In contrast, compute tiles are connected at 2.6 TB/s speeds to the chip fabric. With eight RAMBO cache tiles, we get an additional 120 MB SRAM present. The base tile is a 646 mm² die manufactured in Intel 7 semiconductor process and contains 17 layers. It includes a memory controller, the Fully Integrated Voltage Regulators (FIVR), power management, 16-lane PCIe 5.0 connection, and CXL interface. The entire area of Ponte Vecchio is rather impressive, as 47 active tiles take up 2,330 mm², whereas when we include thermal dies, the total area jumps to 3,100 mm². And, of course, the entire package is much larger at 4,844 mm², connected to the system with 4,468 pins.
Intel chose TSMC's N5 node for compute tiles, while the Xe-Link tiles use the TSMC N7 node. For RAMBO cache and Foveros base tiles, Intel 7 process is used. The entire chip is designed for maximum efficiency and performance and has a TDP of 450 Watts for air cooling, while the water cooling enables it to boost TDP to 600 Watts. Ponte Vecchio is designed for 63-81°C operation—a standard requirement for this type of product used in HPC sector.
Source: Hardwareluxx.de
Add your own comment

9 Comments on Intel Details Ponte Vecchio Accelerator: 63 Tiles, 600 Watt TDP, and Lots of Bandwidth

#1
Vya Domus
This totals for about 47 tiles. However, an additional 16 thermal tiles used in Ponte Vecchio regulate the massive TDP output of this accelerator.
If you're making a processor that has so many dies such that some of them need to be dummies for thermal dissipation you may be starting to defeat the purpose of the design in the first place.
Posted on Reply
#2
TheoneandonlyMrK
Vya DomusIf you're making a processor that has so many dies such that some of them need to be dummies for thermal dissipation you may be starting to defeat the purpose of the design in the first place.
And any cost savings smaller tiles present.
Seems like Intel are literally building pontevechio from as many dies as possible yet is it beating anyone at anything, that we need to know.
Posted on Reply
#3
Tomorrow
Cant even imagine the yield and packaging issues this thing has with that many tiles. They could have named it Ponte Frankenstein. An intresting experiment but not very viable for mass production.
Posted on Reply
#4
Richards
Most technological advanced gpu in the world.. nvidia and amd can dream of building this masterpiece
Posted on Reply
#5
TheoneandonlyMrK
RichardsMost technological advanced gpu in the world.. nvidia and amd can dream of building this masterpiece
I will heartily agree on the technical masterpiece but as for the fanboi trite bit your confused , neither Nvidia or AMD make any package they design IP for use with a foundries PDK and have it packaged and such as needed but Tsmc have Absolutely done as complex a chip and obviously produced EMIB compatible chiplets Soo they even have that proprietary stuff sussed, obviously they can't use it elsewhere but they aren't behind anyone either.
Posted on Reply
#6
TheGuruStud
1000W eh? LOLtel is still at it with this failed crap? You lost the supercomputer contracts, deal with it!
Posted on Reply
#7
ghazi
TheoneandonlyMrKI will heartily agree on the technical masterpiece but as for the fanboi trite bit your confused , neither Nvidia or AMD make any package they design IP for use with a foundries PDK and have it packaged and such as needed but Tsmc have Absolutely done as complex a chip and obviously produced EMIB compatible chiplets Soo they even have that proprietary stuff sussed, obviously they can't use it elsewhere but they aren't behind anyone either.
Didn't TSMC recently announce a technology of their own very similar to EMIB? AMD will need it soon enough. But not for the purposes of constructing a monstrosity like this... "chip" :laugh: Just to catch up on cross-die latency.
Posted on Reply
#8
TheoneandonlyMrK
ghaziDidn't TSMC recently announce a technology of their own very similar to EMIB? AMD will need it soon enough. But not for the purposes of constructing a monstrosity like this... "chip" :laugh: Just to catch up on cross-die latency.
I know, 5800X3d used it ,but as a foundrie they're also privy to Emib designs, they're making chiplets that comply to Intel's design.
And are already in pontevechio.

I'm fairly sure Intel would be fine sharing Emib with everyone, with a license:p :)
Posted on Reply
#9
trsttte
TheoneandonlyMrK5800X3d used it
Not really, X3D is more akin to Foveros stacking chips (CoWoS - Chip on Wafer on Substrate is the name TSMC uses)
Posted on Reply
Dec 23rd, 2024 22:28 EST change timezone

New Forum Posts

Popular Reviews

Controversial News Posts