Piledriver trickles down to Opteron 4300 and 3300 CPUs

AMD has released a fresh batch of Opteron processors with Piledriver inside. These chips are based on the same “Abu Dhabi” microarchitecture as the Opteron 6300 series that was announced last month. However, the Opteron 4300 and 3300 processors sport fewer cores, less cache, lower prices, and more conservative TDP ratings. Here’s how the various model stack up:

Model number Cores Base/Turbo clock speed TDP Price
4386 8 3.1/3.8GHz 95W $348
4376 HE 8 2.6/3.6GHz 65W $501
4340 6 3.5/3.8GHz 95W $348
4334 6 3.1/3.5GHz 95W $191
4332 HE 6 3.0/3.7GHz 65W $415
4310 HE 4 2.2/3.0GHz 35W $415
3380 8 2.6/3.6GHz 65W $229
3350 HE 4 2.8/3.8GHz 45W $125
3320 E 4 1.9/2.5GHz 25W $174

All of the chips have 8MB of L3 cache. The Opteron 4300 offerings can support up to six DIMMs and 192GB of memory per CPU, while the 3300 series is limited to four DIMMs and 32GB. Both families have dual memory channels that work with DDR3 memory up to 1866MHz. The press release also mentions support for ultra-low voltage 1.25V memory, but only in relation to the Opteron 4300 series.

The Opteron 4300 series’ dual HyperTransport links allow it to work in dual-socket systems. With only one HT interconnect, the 3300 family is limited to single-socket implementations. The new chips should drop into the same sockets and systems as their predecessors, facilitating easy upgrades. Incidentally, AMD claims the Opteron 4300 series offers 15% better performance than the previous generation in SPECint and a 24% boost in performance per watt in SPECpower.

While AMD doesn’t quote performance data for the Opteron 3300 series, it’s worth noting that the 3320 E breaks new ground with its 25W TDP. The most power-efficient chip in the old Opteron 3200 lineup has a 45W thermal envelope. The Opteron 4310 HE’s 35W TDP isn’t any lower than the previous generation, but the chip’s clock speeds are higher.

Comments closed
    • ronch
    • 7 years ago

    The Friday Night Topic last week got me thinking about AMD’s server lineup. It seems from the comments last week that some software companies charge on a per-core basis. If you were building a server farm or if you’re the IT dept.’s budget planner, how would you like to pay for more cores while getting less per-core performance? Does AMD’s generally lower pricing compensate for the added software and power costs?

    • Bensam123
    • 7 years ago

    So are these cores or threads? Cause a 8/16 processor $229 is a downright steal, but last I heard they cost ~$500 from AMD.

    I definitely see a lot of life in enthusiast AMD server processors, but they don’t have the boards or the prices to support them. Depending on what that 3380 is, that could change… OCing 8 cores would drive people bonkers. More so because the consumer market stagnated on four cores when everyone was going core crazy (except for hexa core Intels which are uber expensive).

    I’ve said before that they should move their server lineup down to the desktop. It would totally outclass Intel and AMD doesn’t have a huge chunk of the server market anyway so they wouldn’t lose a whole lot. They could keep uber cache and ram offerings for servers too.

      • mnecaise
      • 7 years ago

      [quote<]So are these cores or threads? [/quote<] AMD's definition of a core under Piledriver: 8 integer cores, 2 cores per module for 4 modules. One deep wide fpu per each of 4 modules. [quote<] I've said before that they should move their server lineup down to the desktop.[/quote<] The current FX processors are basically Opteron 33xx processors, or vice-versa. The Opterons at that level target low power consumption. The FX performance. That is reflected in the clock speeds. In general though, I agree. I might like to see the higher performance Opterons available in desktop boards (without having to pay the "workstation class machine" tax).

        • Bensam123
        • 7 years ago

        Well I don’t know, usually when AMD updates charts like this they include ’12 and 16 core’ processors, which aren’t present. I thought TR may have edited the chart to reflect actual cores instead of logical cores.

        A 2/4 core processor for a server seems unlikely for instance…

    • Arag0n
    • 7 years ago

    I would like to see that as an FX version

    3380 8 2.6/3.6GHz 65W $229

    • zdw
    • 7 years ago

    The 3320 E or 3350 HE would be pretty darn slick in a HP Microserver style box.

    • MadManOriginal
    • 7 years ago

    IT people – does anyone actually do in-place CPU-only upgrades on their servers? Just curious.

      • UberGerbil
      • 7 years ago

      I’ve never seen it, at least on equipment that was on a service contract, though “informal” servers (and workstations) might get it. Though I don’t have the “huge fleet” IT experience that some other folks here have, so it may happen.

      But the “in place upgrade” benefit is really at the OEM level, where they can offer the new CPU on the same boards / racks / blades with the same power supplies and cooling solutions without having to go through a whole new validation process. That saves significant time and money for the vendors thus allowing them to add it to their lineup relatively quickly and painlessly.

      • poohbah10
      • 7 years ago

      Outside of HPC, CPU swapping is fairly uncommon; most businesses will move to new servers rather than disassemble and reassemble production systems.

      The main advantage in working with existing infrastructure is that an organization can standardize on a platform, while upping performance or lowering power. This can make support of the infrastructure much simpler, while enabling a growth path.

      • just brew it!
      • 7 years ago

      I suppose it may be more common to upgrade CPUs in HPC applications. For more typical server use scenarios, I/O upgrades are probably a lot more common than CPU upgrades. Here where I work we’ve never done CPU upgrades, but we’ve upgraded the disks in the RAID arrays a couple of times.

      • Flatland_Spider
      • 7 years ago

      Not really, no. I’ve seen CPUs replaced, but they were faulty. Most of the time the server gets replaced out right.

      I have created a couple of Frankenservers for testing and what not. 🙂

      • DrakeRocks
      • 7 years ago

      I was interested in upgrading our servers to the 4200 opterons because of the SS4.2 string instructions that fly on some local text processing applications. I could not find any supplier with stock for any AMD Opteron 4200 parts in the UK so assume that was OEM only or just a paper product. Fingers crossed the 4300 will actually reach the channel.

      • Deanjo
      • 7 years ago

      On Intel based servers, I have never seen it done. On AMD servers however I have seen it done a few times but that is because their socket compatibility life is so long. The upgrades that I have seen on the AMD systems were primarily done going to cpu’s with more cores and were being used for virtualization setups.

      • Arvald
      • 7 years ago

      I have done it on a higher end machine. We got in another CPU module for an HP Itanium server and the vendor had sent the next model up instead (was upgrading 3 servers from 1 CPU to 2.)
      We instead in place upgraded 2 servers (vendor sent a 4th CPU as an apology later once the issue was explained) .

      when dealing with machines like this it is not as likely though, larger shops tend to value support contracts over upgradability.

      • derFunkenstein
      • 7 years ago

      I think this is more so OEMs can just keep existing server designs and sell them with the new CPUs.

      • designerfx
      • 7 years ago

      Anecdotally, no.

      Logically, why wouldn’t someone? it’s just a matter of cost to performance gain, and if it’s more beneficial to drop in a new CPU vs buy a new server then I’m sure they would.

        • ludi
        • 7 years ago

        [quote<]Logically, why wouldn't someone?[/quote<] Always the possibility of an undiscovered bug in the new design, and nobody wants to be the whipping boy for a production server going down. Here at work, prior to Big Corporate Acquisition a year and a half ago, we had a closet full of Frankenservers, an ancient pre-IP phone and voicemail system, and second-hand routers because cost was the bottom line on everything. It all worked, but there were bugs. Now, the most important factor is reliability, and everything is off-the-shelf Dell and Cisco.

      • crystall
      • 7 years ago

      I’ve seen it done in a past job but it was the exception rather than the rule. In that case it was done mostly because our vendor was charging so much for high-end CPUs over Intel’s list price that it made more sense to us to buy their entry-level offering and then pop in CPUs we had bought on our own. IIRC it was a ~500 machines deployment so not very large. You probably wouldn’t want to do something like that in a large datacenter.

      • Bensam123
      • 7 years ago

      Good question, have any of the people responding worked with AMD servers and had the option to do CPU only upgrades?

      • dmjifn
      • 7 years ago

      I haven’t seen that but I *have* seen our company buy a server with 1-2 empty sockets that they might fill later. Usually they don’t though.

Pin It on Pinterest

Share This