facebook rss twitter

AMD FX processors won’t get Steamroller cores in 2014

by Mark Tyson on 18 November 2013, 12:16

Tags: AMD (NYSE:AMD)

Quick Link: HEXUS.net/qab5cv

Add to My Vault: x

After closer inspection of the latest AMD roadmap November 2013 update published last week we notice that there is something conspicuous by its absence – AMD FX processors featuring Steamroller cores.

We covered the new entries on the AMD roadmap last week; namely ‘Beema’ and ‘Mullins’ APUs, aimed at the low-power portable and ultra-low fanless mobile markets respectively. These new mobile orientated chips offer impressive performance per watt and will be seen in devices on the market within H1 2014. Later in the week we even saw a first leaked device using the 2W SPD ‘Mullins’ APU, a tablet codenamed ‘Project Discovery’.

From the roadmap for desktops we can see the AMD Kaveri APUs with 2-4 Steamroller cores are an imminent arrival; the roadmap indicates they are already here but AMD has named an official retail release date of 14th January, a week after the CES2014 show. However, as far as is indicated, the ‘performance’ range of FX CPUs will stay with its 32nm Piledriver architecture until the end of 2014 and possibly beyond that.

You could read this as a shift of focus away from performance desktop efforts or a stronger emphasis on mobile processors, either way it looks like the FX processor range isn’t being changed for a while. Meanwhile AMD is still planning to develop a server chip called ‘Berlin’ for its 1P Web/Enterprise services clusters which features 4x Steamroller cores, GCN and HSA features, a lot like ‘Kaveri’ minus TrueAudio.

Is AMD’s lack of planned advancement in its performance FX CPU range and its push to develop low-power chips like ‘Beema’ and ‘Mullins’ the most prudent use of its resources?



HEXUS Forums :: 90 Comments

Login with Forum Account

Don't have an account? Register today!
Well that just about wraps it up for top end prices. With no real competition Intel will innovate little and keep the prices high.
Quite disappointed but I think on the desktop side, AMDs strategy is to optimise game engines to better use more cores and to offload as much to the graphics card as possible. (Which is why I think the console wins were important). Then offer value to compete favorably on the price points.

Would have still liked to see a new FX generation though.
Does this also mean no further development in the higher end 2p server space?, would be a shame to see them fall even further behind there, pricing on there boxes makes then very attractive for Virtual environments.
I wonder if they're moving to an Intel like roadmap of releasing their APUs first, then the server parts a gen later - we've only just seen the server versions of IVB, after the desktop Haswell release. It's unfortunate for us, but it does make sense - they build the server parts from proven cores on a mature process, especially important when you're dealing with larger die sizes if you want reasonable yields. Launching a large die with a brand new, ground-up uarch on a cutting edge process didn't go exactly go smoothly for AMD with BD. Intel's tick-tock methodology makes a good amount of sense for that reason, taking on one challenge at a time.
I think a new FX core will have to come eventually, as they'll need to refresh the server range before too long, but I also think they don't want to develop new chips on an aging platform. They're obviously positioning FM2+ as the 1P successor to AM3 in entry-level servers, which presumably means they have no real interesting in increasing the shelf-life of AM3+, so no interesting in releasing new FX CPUs or AM3+ Opterons.

I suspect they will release new FX CPUs when they decide on how to unify the APU and CPU platforms, which I've theorised previously might be when they decide to adopt DDR4. The fact that the server chips use HT for interCPU communication means they'll have to do a fairly significant silicon redesign to fulfill the multi-socket server chips (including MCMs for high core count) and bring all the current northbridge logic onto the CPU die. Makes me wonder if the server side of the business might be a major target for the semi-custom team…