At Computex, Intel Demos A 28-Core Processor Clocked At 5GHz (Updated)
EDIT: According to this image we sourced from Engadget's compressed keynote video on YouTube, Intel was apparently running some sort of closed-loop cooling that required insulating material around the tubing. This could be connected to a multi-stage phase cooler (sub-zero cooling), or possibly a more mundane water chiller, under the table. We also spot more shielding over the long rectangular waterblock and what appears to be six sticks of RAM flanking the processor on each side. That implies the platform is based on the (until now) enterprise-class LGA3647 socket. This means the processor could be a variant of the $8,700 Xeon Platinum Scalable processor we reviewed here, albeit with an unlocked multiplier. (That doesn't bode well for pricing). We'll dig for more details and update as necessary.
EDIT #2: We tracked down Intels's test system, which confirms the company was using a chiller with the test setup. Head to our Intel's 28-Core 5GHz Processor And Test System Breaks Cover article for more details.
Intel's keynote address at Computex 2018 came with a few surprises, but none were more surprising to the crowd than the demonstration of a 28-core workstation-class processor running at 5GHz. Intel says the new chip will come to market in Q4 of 2018. Before the event, Intel pre-briefed us about the 28-core model, but the company didn't disclose that it would come bearing a beastly all-core 5GHz frequency. The impressive display of multi-threaded performance probably consumed a hideous amount of power, which wouldn't lend itself well to a reasonable TDP rating for the processor. As such, we assume the processor was overclocked for the presentation.
Intel's 14nm++ process has already proven that it can sustain 5GHz easily, even with all cores active, on six-core models. Even novice overclockers can achieve a 5.0 GHz overclock on many standard off-the-shelf K-series Coffee Lake processors. But processors with higher core counts, such as the 28-core model Intel had on display, often can't reach such high frequencies on all cores, although dedicated tuning can certainly squeeze out impressive single-core frequencies.
For instance, we tested the 16-core Core i9-7960X and could only reach up to 4.3GHz before the processor was overwhelmed by the intense thermal output and 400W of power consumption. In fact, we've pushed past 350W of power consumption during some of our overclocked tests with the smaller 10-core Core i9-7900X. We've heard reports of up to 800W with the 18-core model and LN2 cooling, so its hard to speculate how much power Intel's new 28 core processor was pulling during the demonstration. It's likely that the new model has indium solder to help dissipate heat from the die, but Intel hasn't shared specifics.
Intel's 28-core processor scored 7,334 CB Marks in an all-core CineBench test. For perspective, our eighteen-core Core i9-7960X at 4.2GHz reaches up to 4,136 CB Marks. We included a mock-up chart above with Intel's results superimposed over results from our labs, but be aware that we haven't confirmed Intel's numbers and the systems were tested under different conditions. In either case, the results are incredibly impressive compared to AMD's overclocked Threadripper processors and the 16- and 18-core Intel models.
Unlike AMD, which pushes the voltage/frequency curve to the limits with its 2000-series Ryzen processors, Intel typically leaves itself plenty of room for higher frequencies. We often see this even on Intel's fastest models, which only boost up to 4.7GHz on a single core and 4.3GHz on all cores, but often overclock up to 5.1GHz on all cores.
Intel's sudden push to wring the utmost performance out of its processors is obviously designed to one-up AMD before its rumored 32-core Threadripper 2 launch tomorrow. Intel also unveiled its new 40th-anniversary Core i7-8086K processor that comes with a 5GHz boost clock, and given the company's 28-core demo, we could see the company push the limits with its future mainstream desktop processors, as well.
Intel also announced that it would have new S-Series models coming this year. These are obviously the Whiskey Lake processors, but Intel left us hanging on the details. It would be nice if Intel threw a curveball and deployed a completely new microarchitecture for the first time in four years, but that doesn't seem to be likely. For now, it appears that Intel is leaving its new microarchitectures locked behind the next step on the process ladder. It's puzzling that Intel hasn't simply brought a new microarchitecture to the existing process, thereby offering larger performance benefits than we see with the rewarmed Skylake architecture.
In either case, Intel has a proven process and microarchitecture that still has some gas in the tank. If Intel carries this new approach over to its S-Series processors, it will set the stage for a much more interesting battle with AMD as Intel rolls out new mainstream desktop chips with higher core counts and record-setting stock frequencies.