AMD's ancient Bulldozer and Piledriver platforms getting new open source firmwares in 2025 — update delivers 15-second boot-up times with 256GB memory setups
AMD's FX series and Opteron CPUs based on Bulldozer and Piledriver are still being supported through a third-party on a handful of motherboards.
Believe it or not, AMD processors based on Bulldozer and Piledriver are receiving firmware updates in 2025 through third-party open-source firmware developers. Providing these updates is an independent software project called 15h.org, which supports a handful of server and desktop boards.
One of the developers, Mike Rothfuss, wrote to Phoronix about several major updates the project has already published. These included bug fixes and performance updates such as bug free RAM initialization, an increase in memory support to 512GB, "fast and reliable boots" with the platform now booting consistently within a 15 second window for systems boasting 256GB of RAM, bug free IOMMU support, full support for AMD's speculative execution patches, support for up to four PCIe devices and more.
The project is developing firmware updates for these processors because they represent the final high-performance x86 CPUs that are not hamstrung by newer UEFI security features such as firmware signing. This makes these chips some of the newest CPUs that can be harnessed to build a fully open-source system, from the operating system all the way down to the firmware level.
Phoronix reports that 15h.org uses existing AMD open-source firmware code to implement these updates. Back when Bulldozer was new, AMD reportedly promoted Coreboot and made its AGESA code open source.
Bulldozer and Piledriver are renowned for their terrible single-core performance and earned a reputation as some of AMD's worst ever CPU designs. Regardless, both architectures still provided competitive multi-core performance for their time, making them ideal for heavily multi-threaded applications.
The 15h.org officially supports several server and desktop motherboards that run either FX or Operon Bulldozer/Piledriver processors, featuring the Asus KGPE-D16, SuperMicro H8SCM, and SuperMicro H8SGL. There is also beta support for the Asus F2A85-M, 2A85-M Pro, F2A85-M Le, SuperMicro H8DGi, H8DG6, H8QGi+-F, and H8QG6+-F.
Bulldozer and Piledriver CPUs might not be the fastest CPU around, even by 2011-2015 standards, but they are the fastest chips capable of running completely open-source firmware from the ground up. This will inevitably be appealing to Linux enthusiasts who want to create their own desktops and servers from scratch while having complete control over their system's BIOS.
Get Tom's Hardware's best news and in-depth reviews, straight to your inbox.
Follow Tom's Hardware on Google News, or add us as a preferred source, to get our latest news, analysis, & reviews in your feeds.

Aaron Klotz is a contributing writer for Tom’s Hardware, covering news related to computer hardware such as CPUs, and graphics cards.
-
qxp So I disagree with description of Opteron's as "worst ever CPU design". At their time they were solid 64-bit chips, with better price/performance than comparable Intel offerings. Nice to hear about open source firmware. Control over such firmware can have an advantage in real-time control applications, as closed-source commercial firmware can result in periodic latency spikes due to firmware code running in competition with the OS (not sure if all motherboards do it).Reply -
usertests Reply
The article doesn't say Opteron, it says Bulldozer/Piledriver:qxp said:So I disagree with description of Opteron's as "worst ever CPU design". At their time they were solid 64-bit chips, with better price/performance than comparable Intel offerings.
Bulldozer and Piledriver are renowned for their terrible single-core performance and earned a reputation as some of AMD's worst ever CPU designs.
Technically, there are Opterons using older cores, Jaguar cores, and even ARM Cortex-A57.
The Bulldozer family has a horrible reputation from having low single-thread performance during a time when 8 "cores" were hard to utilize, terrible efficiency exasperated by being pushed towards 5 GHz in the enthusiast parts, and having effectively fake core counts using modules with shared resources, which they were sued over.
The fastest FX-9590 is steamrolled by the i7-4770K and later parts using half the energy.
Maybe the MCM 16-"core" Opterons were a great value compared to Intel Xeon offerings, but they were less efficient. -
palladin9479 Replyqxp said:So I disagree with description of Opteron's as "worst ever CPU design". At their time they were solid 64-bit chips, with better price/performance than comparable Intel offerings. Nice to hear about open source firmware. Control over such firmware can have an advantage in real-time control applications, as closed-source commercial firmware can result in periodic latency spikes due to firmware code running in competition with the OS (not sure if all motherboards do it).
Much of Bulldozer / Piledrivers hate isn't justified. AMD had moved over to a new modular chip design methodology and was very much still developing and refining it. Much of BD/PD was built using automated chip design software that essentially built the design using lego like modules. After much refining and development, that software is now responsible for Ryzen and why AMD can build custom CPU's so cheaply. By contrast Intel has to do all the tuning and optimization work by hand based on whatever process node they are using for that CPU. This advantage has allowed AMD to go through design cycles much faster then Intel with a smaller staff.
Or to put it another way, if AMD never attempted Bulldozer, we wouldn't have Ryzen today. -
qxp Replyusertests said:The article doesn't say Opteron, it says Bulldozer/Piledriver:
Technically, there are Opterons using older cores, Jaguar cores, and even ARM Cortex-A57.
The Bulldozer family has a horrible reputation from having low single-thread performance during a time when 8 "cores" were hard to utilize, terrible efficiency exasperated by being pushed towards 5 GHz in the enthusiast parts, and having effectively fake core counts using modules with shared resources, which they were sued over.
The fastest FX-9590 is steamrolled by the i7-4770K and later parts using half the energy.
Maybe the MCM 16-"core" Opterons were a great value compared to Intel Xeon offerings, but they were less efficient.
Actually, at the very top the article says "..Opteron CPUs based on.." and I was remembering the 8-core ones. For me they were priced right and run great - just put Linux on them and start 8 separate processes. Really don't see what is so hard to utilize 8 cores. -
palladin9479 Replyqxp said:Actually, at the very top the article says "..Opteron CPUs based on.." and I was remembering the 8-core ones. For me they were priced right and run great - just put Linux on them and start 8 separate processes. Really don't see what is so hard to utilize 8 cores.
I believe it's more that the consumer workloads of that time were not heavily parallel. Your application can spawn 20 threads but if 90~95% of the work is being done in the main logic loop, then extra cores are not doing much for you. That was the landscape of that era and how all processors were evaluated. BD wasn't so great as a server CPU architecture due to how new and unoptimized the process was at that time, PD (which I still have some) was much better but by then people had a bad taste in their mouths. The Opteron's based on PD were pretty good workstation and server CPUs, but the CPU is one of the less important components of those systems. Intel had a near monopoly on the server / workstation platforms of that era, these are the motherboard + NIC + storage + management components. Enthusiasts could definitely "build" a cheap workstation / server from AMD components, but most customers just purchased these items as singular units from vendors who provided complete support and validation for the product. Most of those AMD offerings were packaged as "cheap" and frequently were sold with cheaper lower performing components or with less capacity.
AMD realized how important those vendors were, and have since formed very strong relationships with them to ensure good premium products are made available to the market. -
qxp Replypalladin9479 said:I believe it's more that the consumer workloads of that time were not heavily parallel. Your application can spawn 20 threads but if 90~95% of the work is being done in the main logic loop, then extra cores are not doing much for you. That was the landscape of that era and how all processors were evaluated. BD wasn't so great as a server CPU architecture due to how new and unoptimized the process was at that time, PD (which I still have some) was much better but by then people had a bad taste in their mouths. The Opteron's based on PD were pretty good workstation and server CPUs, but the CPU is one of the less important components of those systems. Intel had a near monopoly on the server / workstation platforms of that era, these are the motherboard + NIC + storage + management components. Enthusiasts could definitely "build" a cheap workstation / server from AMD components, but most customers just purchased these items as singular units from vendors who provided complete support and validation for the product. Most of those AMD offerings were packaged as "cheap" and frequently were sold with cheaper lower performing components or with less capacity.
AMD realized how important those vendors were, and have since formed very strong relationships with them to ensure good premium products are made available to the market.
In my case, I got a Dell server first with Intel CPU. It was a power hog and run very slow. Next I got an Opteron system which was faster and cheaper. Needless to say I was only buying AMD for a year or two. All were prebuilt, with onsite warranty. Intel had a near monopoly with vendors like Dell because of the contracts. Dell still lists mostly Intel systems on their website. Thankfully, Supermicro was making perfectly good motherboards and cases, and if you bought systems from a good vendor they were rock-solid. Intel later released E5's and those were nice.
20-thread systems came later, and if your application is mostly single thread, you would want to configure the system to reflect that - higher frequency and CPU with smaller latency, which I think would imply AMD as well at the time (certainly at the time of Athlon 64). -
hwertz I wonder what that RAM costs? As we well know DDR4 pricing is apocalyptically high recently, picking one of these up and shoving 256-512GB RAM into it could be useful and possibly much less expensive than just the DDR4 or DDR5 alone (unless others already did this and demand drove the prices up already; I read about someone buying about that age Xeon to run like DeepSeek on since they could cram 512GB or maybe even 1TB RAM into it.)Reply -
rluker5 Those quad core: https://www.tomshardware.com/news/amd-fx-bulldozer-false-advertising-class-action-lawsuit-eight-cores-settlement,40256.html Bulldozers and Piledrivers are still good for low end use like old games and video streaming, like my Dell Venue 11 Pro 5130 with a 2w Atom chip that performed on par with the quad core Phenom 2 laptop I had. So it is nice that they can do modern boot times now, but I wouldn't waste money on extra ram for such a system, just get something better with that money like some used Zen 3, or Haswell or newer build. Haswell is slower, but it is so cheap on Ebay. 4770k+mobo+ram < 100 delivered and is still fairly nice for everything but workstation, new AAA games > 60 fps all the time, and W11 official support. Zen 3 has 2 of those covered but runs about 3x as much for a 5600x+mobo+ram.Reply
Edit: Also isn't CSM a thing for getting around UEFI signing? -
artk2219 Reply
Its a similar story for Intel and the Pentium 4, well at least in terms of what they gained from it. Intel learned how to deal with heat, high motherboard power requirements, what it takes to build a cpu to run at high frequencies, hyper threading, multi core processors, how to tune a power hungry architecture for lower power mobile requirements. Intel learned a ton from the design, even if it was a flawed product. Core 2, and especially Nehalem, benefited greatly from what they learned during this period, even if they were a bunch of cheaters :confused:palladin9479 said:Much of Bulldozer / Piledrivers hate isn't justified. AMD had moved over to a new modular chip design methodology and was very much still developing and refining it. Much of BD/PD was built using automated chip design software that essentially built the design using lego like modules. After much refining and development, that software is now responsible for Ryzen and why AMD can build custom CPU's so cheaply. By contrast Intel has to do all the tuning and optimization work by hand based on whatever process node they are using for that CPU. This advantage has allowed AMD to go through design cycles much faster then Intel with a smaller staff.
Or to put it another way, if AMD never attempted Bulldozer, we wouldn't have Ryzen today. -
qxp Reply
That's not my recollection. Pentium 4 much slower than Athlon 64 because Intel tried to ramp up the frequency by having a brand new design with extra deep pipeline that suffered horribly on missed branches.artk2219 said:Its a similar story for Intel and the Pentium 4, well at least in terms of what they gained from it. Intel learned how to deal with heat, high motherboard power requirements, what it takes to build a cpu to run at high frequencies, hyper threading, multi core processors, how to tune a power hungry architecture for lower power mobile requirements. Intel learned a ton from the design, even if it was a flawed product. Core 2, and especially Nehalem, benefited greatly from what they learned during this period, even if they were a bunch of cheaters :confused:
From what I heard, Core 2 was an evolved Pentium III design that was deveoped by a separate team in Israel and aimed at low-power applications. The design was brought forward when Pentium 4 flopped. I don't know of any further designs that were based on Pentium 4.