Some sleuthing by CPU-World has uncovered the list of to-be-released Kaby Lake single-socket quad-core Xeons. As it to be expected, these are incremental updates from Skylake-based Xeons using the newer 14nm Plus node from Intel. In our consumer Kaby Lake reviews, our results showed that the new design offers a better voltage/frequency profile than previous generations, affording more frequency at the same voltage. Another big change from the previous generation is the TDP: what used to be 80W is now listed as 73W if it has integrated graphics, or 72W if it does not.

The list from CPU-World, who in turn discovered a QVL (qualified vendor list, or ‘CPUs which we confirm work in this board’) posting from a motherboard manufacturer whom accidentally included the new Xeons. The posted list features eight Xeon processors altogether. The two at the bottom of the stack are quad core parts without hyperthreading, and the others do have hyperthreading. The main differences between the processors will be frequencies and the presence of integrated graphics.

Intel E3-1200 v6 CPUs (Kaby Lake)
  C/T Base Freq L3 Cache IGP IGP Freq TDP
E3-1280 v6 4/8 3.9 GHz 8 MB - - 72 W
E3-1275 v6 4/8 3.8 GHz 8 MB P630 1150 MHz 73 W
E3-1270 v6 4/8 3.8 GHz 8 MB - - 72 W
E3-1245 v6 4/8 3.7 GHz 8 MB P630 1150 MHz 73 W
E3-1240 v6 4/8 3.7 GHz 8 MB - - 72 W
E3-1230 v6 4/8 3.5 GHz 8 MB - - 72 W
E3-1225 v6 4/4 3.3 GHz 8 MB P630 1150 MHz 73 W
E3-1220 v6 4/4 3.0 GHz 8 MB - - 72 W

Most of these numbers come direct from the motherboard validation lists, with some such as core count being derived from L2 cache listings. All the parts listed have a full 8MB of L3 cache, indicating they run closer to the Core i7 design rather than a Core i5 (even those that have hyperthreading disabled).

On the integrated graphics models, i.e. those ending in '5', are all running Intel HD P630 graphics and run up to 1150 MHz. This is the ‘professional’ version of the HD630 we see on the consumer parts, using Intel’s latest Gen9 graphics architecture and supporting H.265 encode/decode. Our Kaby Lake review piece goes into more detail.

Not listed are the turbo frequencies of the CPUs, as these are currently unknown. Neither is the pricing, however given previous launches we would expect the tray price (OEM batches of a thousand CPUs) to have parity compared to previous generations.

Intel E3-1200 v6 and v5 CPUs
IGP v6 Model v5 IGP
- 3.9, 72W E3-1280 3.7/4.0, 80W -
+ 3.8, 73W E3-1275 3.6/4.0, 80W +
- 3.8, 72W E3-1270 3.6/4.0, 80W -
- - E3-1260L 2.9/3.9, 45W -
+ 3.7, 73W E3-1245 3.5/3.9, 80W +
- 3.7, 72W E3-1240 3.5/3.9, 80W -
- - E3-1240L 2.1/3.2, 25W -
- - E3-1235L 2.0/3.0, 25W +
- 3.5, 72W E3-1230 3.4/3.8, 80W -
+ 3.3, 73W E3-1225 3.3/3.7, 80W +
- 3.0, 72W E3-1220 3.0/3.5, 80W -

For the most part, the new processors are ~200 MHz faster than the v5 parts while still being rated at the lower TDP. Memory support is expected to be the same as the consumer parts (DDR4-2400), and it is not yet confirmed if the v6 processors will support Transactional Synchronization Extensions (TSX) given issues in previous revisions, so we will wait on future Intel announcements on this front.

It is still worth noting that for LGA1151 based Xeons, Intel adjusted the requirements such that Xeon processors require a server grade chipset on the motherboard. For Skylake E3 v5 parts, this was either a C232 or C236 chipset – we reviewed a few motherboards with these on (ASRock E3V5 Gaming, GIGABYTE Z170X-Extreme ECC). With a BIOS update, these C232/C236 motherboards should support the new v6 processors. However, we currently do not know if there will be a second generation of chipsets for these CPUs in line with the consumer updates. On the consumer side the new chipset has additional PCIe lanes and Optane Memory support, so we stand in wait for a new desktop chipset to support these. There is a new mobile chipset, CM238, for mobile E3 v6 Xeons, but no equivalent in the desktop space yet.

We currently have all the Skylake E3 v5 Xeons in for testing on our new benchmark suite soon, and we’ll make similar moves to acquire the Kaby Lake E3 v6 models when they are released. Currently there is no word on release date or pricing, however we typically see the E3 Xeons release very shortly after the consumer processor release.

Source: CPU-World

POST A COMMENT

31 Comments

View All Comments

  • BrokenCrayons - Tuesday, January 24, 2017 - link

    You're right, 8 watts doesn't seem like a lot, but it's a 10% reduction which, for a mild CPU update, is a notable difference. I doubt any current Xeon owners will feel pressured to upgrade, but I think Intel's focus on slowly driving TDP downward is a far better approach than the GPU half of the industry that has been creeping upward lately. Reply
  • flgt - Tuesday, January 24, 2017 - link

    When you put it in those terms I'm thinking that companies running datacenters will be happy when the remaining Xeon variants are release. Reply
  • BrokenCrayons - Tuesday, January 24, 2017 - link

    They might indeed. There's other factors besides CPU TDP to consider that'll play a role in the total cost of ownership analysis. I don't know of many companies that will undertake upgrades outside of planned lifecycles so businesses running 3-5+ year old equipment are most likely to pick up Kaby Lake parts almost regardless of power and cooling considerations. Companies with newer hardware probably aren't going to be tempted into making capital investments to keep up with generational processor upgrades unless there's greater incentives like platform upgrades as well that'll offer a really compelling argument to justify upending current hardware. The TDP is important, but its more like the icing rather than the entire cake. Reply
  • Molbork - Tuesday, January 24, 2017 - link

    For individuals it's not a big deal, but the improved perf/(cost*Watt) value scaled across thousands of servers\workstations over a couple of years is significant.
    8W savings for iso-perf across 1000 machines for 1 year at 10¢ per kWh(ha I wish, but businesses might be able to negotiate that low) assuming 66% up time under max load...
    8W*16h/day*365days/year*1000cpus/(1000kW/W)*0.1$/kWh =4672$/year for 1k CPUs with my probably flawed calc which doesn't include cooling $$ savings. And if the performance went up, you are getting more done with less energy/cost.
    Or $4.6/year out of your personal stations pocket, ya not much!
    Reply
  • SharpEars - Tuesday, January 24, 2017 - link

    Xeon and only quad core - makes sense (not). Reply
  • jardows2 - Tuesday, January 24, 2017 - link

    Xeon E3's have always been quad core parts. What you add across the board is ECC memory support. Older models also had more cache, though that doesn't seem to be the case in the current lineup. You also can get parts without integrated graphics, and you can get close to or equivalent performance to an i7 for less.

    For example, the Skylake Xeon E3-1230 v5 has the same base clock as the i7 6700, but a 200mhz lower boost. Intel's price is about $50.00 less than the i7. If you don't overclock, and don't need the integrated graphics, it could have been a great deal for a home computer, before Intel killed support in all but the C series chipsets. If you are a professional working with large datasets and can't risk memory errors, the ECC support is a huge plus.
    Reply
  • tomaz4 - Tuesday, January 24, 2017 - link

    Skylake E3 v5 has functional TSX or not? Reply
  • MTEK - Tuesday, January 24, 2017 - link

    I believe TSX was disabled for most Haswell models and fixed sometime during Broadwell. Reply
  • BrokenCrayons - Tuesday, January 24, 2017 - link

    I think it's notable there's only a 1 watt TDP difference between KB Xeons with graphics versus those without since they share the same CPU clock speed and arguably identical processor performance. If those leaked numbers are accurate, then there's likely significant binning between a part with and one without graphics. After all 1 watt TDP can't account for all the difference in those iGPU execution units being disabled or not. Reply
  • close - Wednesday, January 25, 2017 - link

    The TDP number is not an absolute for power consumption. It's actually more useful for cooling. So the parts with IGP might be able to sustain boosted clocks for shorter times than the parts with no IGP.
    Also you will see the TDP stays unchanged even with a 100MHz increase. That would warrant ~2W.
    Reply

Log in

Don't have an account? Sign up now