AMD Releases More Carrizo Details at ISSCC

New Page

AMD Releases More Carrizo Details at ISSCC

AMD Releases More Carrizo Details at ISSCC


AMD had released more details on their Carrizo APUs at AMD's  ISSCC 2015 presentation. At present Carrizo looks like it will only be released for mobile applications and as such focuses on increased efficiency despite using the ageing 28nm process. 

Carrizo will be released as two pin-compatible SoCs, Carrizo which will be using AMD's latest Steamroller CPU cores and Carrizo-L which is AMD's extremely low power variant which will be using AMD Puma+ CPU cores. Excavator will be the final iteration of AMD's Bulldozer architecture, and the integrated graphics will be the latest iteration of AMD's GCN graphics architecture. 


AMD Releases More Carrizo Details at ISSCC  

AMD's Carrizo APU will be released in 15-35w variants, which promise to be smaller, more power efficient and faster than their Kaveri equivalents. Carrizo will also be updated with newer instruction sets, two times the amount of L1 cache and will integrate the southbridge controller into the chip.

These changes from AMD will hopefully make the chip more appealing for system makers, which AMD hopes will allow them to sell more chips for Laptops and x86 tablets. Carrizo will also be the first chip from AMD that will comply with the HSA 1.0 standard, making it one, if not, the first CPUs in the world that comply with the standard.


AMD Releases More Carrizo Details at ISSCC   AMD Releases More Carrizo Details at ISSCC   


One of AMD's reasons for purchasing ATI was their ability to produce very dense chips, GPUs use a High Density Cell Library which allows GPUs to have smaller, less leaky designs which can do a lot of work per cycle. CPUs are usually a larger, faster, more leaky chip designs, so AMD gambled that ATI's High Density Cell designs could be applied to their CPUs, which could dramatically reduce the die size of an APU.

With the implementation of High Density Libraries in  Excavator, CPU core size has been reduced by 23% compared to Kaveri/Steamroller CPU cores. This is an impressive feat considering that AMD's excavator cores use more transistors and the AMD are still using a 28nm process.

AMD did not say if the use of High Density Libraries will affect the potential maximum speed of Excavator, but given that Carrizo will likely never make it's way onto desktops I think that Excavator/Carrizo would be a very difficult chip to scale to high frequencies. The greater chip density shortens all the interconnects between all the functional parts of the APU, which lowers the power consumption of the chip.

Considering that this APU is designed for mobile environment high frequencies are not a priority, so the power savings at low frequencies are more important than raw clock speed. Compared to Kaveri, Carrizo will either have a lower power consumption at the same clock speeds or a run faster at the same TDP. Sadly the scaling problem that I mentioned makes a desktop release highly unlikely. 


AMD Releases More Carrizo Details at ISSCC   AMD Releases More Carrizo Details at ISSCC   


The GPU portion of Carrizo will have AMD's latest GCN GPU design, which will feature many of the improvements seen in AMD's Tonga GPU. AMD have also been able to improve the clock speeds of the GPU and the power consumption, allowing AMD to reduce power consumption by 20% compared to Kaveri while maintaining the same clock speeds.

This change in GPU power consumption will allow them to enable all their GPU cores on mobile APUs, which should give them a great boost in GPU performance, without increasing the TDP.

AMD will again be using Voltage Adaptive Operation, which is a technique which gets around the problem of voltage droop or vDroop. vDroop occurs when the CPU requires more power, but the voltage/power delivery system cannot provide it quickly, CPU makers usually get around this by providing around 20% more power than is actually needed so that any vDroop will not affect system stability. Instead of providing a voltage about 20% higher than what the CPU needs in order to prevent vDroop, AMD have decided to make their CPUs downclock momentarily when vDroop occurs which will provide a 19% and a 10% power saving on the CPU and GPU sides respectively. 


AMD Releases More Carrizo Details at ISSCC   AMD Releases More Carrizo Details at ISSCC   


 AMD have also added a new low power state, called SOi3, which achieves the same thing as the older S3 standby power state which required the OS to create backup files and then turn off the silicon to save power. This requires the OS to create backup files, making the S3 state unpopular. The SOi3 power state can turn off a lot of the silicon while keeping system memory and I/O activated, which means that no backups or OS interaction is required for the system to go on standby. This state can be activated in less than a second, and provides a very low power consumption.


AMD Releases More Carrizo Details at ISSCC   

AMD are very hopeful for their Carrizo and Carrizo-L APUs, while in terms of IPC gains, Excavator achieves only a meagre 5% gain over Steamroller but has achieved a lot of power consumption savings. To put it bluntly I wouldn't be surprised to see similar efficiency from these chips as some of Intel's older 22nm CPUs, but they will still be behind Haswell and Broadwell. These chips are also a good chance for AMD to get HSA compatible hardware out there, which will hopefully tempt some developers to take advantage of it. 

AMD are really being let down by the fact that they are still on the 28nm process, but with many foundries moving to 16nm or 14nm AMD have a good chance to catch up to Intel in about a years time.   


You can join the discussion on AMD's Carrizo APUs on the OC3D Forums.


«Prev 1 Next»

Most Recent Comments

24-02-2015, 07:41:49

Very interested in these chips. I really hope they'll be any good.Quote

24-02-2015, 07:49:54

Lets see whether those chips are suitable for stream boxes 15W of power requirements is some serious business there AMDQuote

24-02-2015, 14:09:42

Sounds promising. I think moving down the node processes will really benenfit them. It would benefit them in many ways. Reason they haven't moved down to say 22nm or lower, may be because the GPUs(from amd or nvidia) are still in 28nm and there may be an industry thing going on we don't know about such as true yields and time to implement.

Wonder what they plan on doing after this in regards to architecture. Hopefully they release something great.Quote

Register for the OC3D Newsletter

Subscribing to the OC3D newsletter will keep you up-to-date on the latest technology reviews, competitions and goings-on at Overclock3D. We won't share your email address with ANYONE, and we will only email you with updates on site news, reviews, and competitions and you can unsubscribe easily at any time.

Simply enter your name and email address into the box below and be sure to click on the links in the confirmation emails that will arrive in your e-mail shortly after to complete the registration.

If you run into any problems, just drop us a message on the forums.