Over the past several years, the processor market has become a more exciting but confusing place for the average buyer looking to keep their facility at the cutting edge.
Unfortunately, ‘confusing’ could be the very least of a buyer’s worries, as the wrong purchase for a particular type of work could see your super-fast processor running much more slowly than expected, and affecting the whole workflow.
We thought we’d take the opportunity to look at what’s currently going on in the battle of cores and clock speeds, and how it will affect the choices you’re likely to make in the future.
In the blue corner
Intel’s i7’s and i9’s were originally conceived as ideal processors for the gaming market. Running games requires a high clock speed, but isn’t a core-intensive operation, and i7’s and i9’s hit this need in the domestic market perfectly.
Over time though, these processors became better and better. Their initial severe memory-capacity limitations became a thing of the past, thanks to the CoreX processors being able to accommodate 128 and even 256GB ram. Additional cores were added, and soon these consumer processors were performing as well as, and in some cases better than, the professional Xeon-based workstations. At one point we had Intel’s Core X series processors running with 18 cores at a base of 3.00Ghz and a turbo speed of 4.6Ghz.
As you can imagine, this brought products meant for the domestic market into sharp focus for the professional market, with the two overlapping.
The situation wasn’t helped with the introduction of W-Series Intel processors designed for workstations, which actually had slower clock speeds and lower core counts in some instances than the domestic Core X processors. As a result, we saw Tier 1 manufacturers like HP offering systems that contain both i7 and i9 Intel-core workstations, which further blurred the lines between Core and Xeon processors.
Intel has moved to address the situation, looking to differentiate clearly between the consumer and professional CPUs. The development from the 10th Gen to 11th Gen Core X was halted, and consumer i7’s and i9’s saw their cores reduced from 10 down to eight.
And in the red corner
Intel’s CISC (Complex Instruction Set Computer) design has been around for a long time, and is one side of the core-count escalation we’ve seen recently. The other is AMD, which also helped to muddy the waters with its Threadripper, a chip also aimed squarely at the gaming market but boasting 32 cores. The AMD high core-count chiplet design can actually accommodate a mind boggling 96 cores on a single chip, and helped usher in the modern super socket era, with everything operating from a single chip plugged into one socket on the motherboard.
Tier 1 manufacturers began using this chip at the heart of workstations too, with Lenovo being the first to offer the Threadripper Pro, soon to be followed by Dell and HP.
Cores for alarm
However, what’s been clear for some time is that simply adding cores leads to more heat generation. As a rule, people don’t like their processors melting, so clock speeds had to be throttled back to reduce the heat.
That decision has clearly had an impact on the Xeon range of CPUs, where we’ve seen that a higher core count always leads to a lower base frequency. Even with performance-boosting features like AVX2 and the newer AVX512, only certain processors in the Xeon range operated across all cores at the listed base clock and turbo speeds.
Is there an answer to this Catch 22 situation? For some time ARM has been developing its big.LITTLE CPU architecture. Originally tried and tested in the mobile phone sector, this design couples slower, more energy-efficient processors with quicker and more powerful processors. This balance within the chip itself was originally driven by the need to save battery power in smartphones while still being able to run high-performance apps. This kind of thinking can also be applied to the workstation core vs clock speed conundrum too, as we’ll see in Intel’s Alder Lake and Raptor Lake machines below.
Choosing the right processor
The job of selecting the right CPU for any given situation has become extremely complicated. Do you base your decision on the number of cores? Clock speeds? The applications you’ll be using? The power needed to spin up all these cores? Price? The only way to make the wise decision is to consider all of these factors: not an easy task.
VFX workflows have always demanded a balance of cores vs clock speed. With multi-threaded tasks such as simulation, rendering and lighting/look development, the power of multiple cores is usually the priority, whilst for single-threaded applications including modelling, texturing and animation, clock speed is the focus.
But what about a smaller facility that needs workstations that can do both effectively? These tend to have machines that are generalised, with as many cores and as high a clock speed as possible. The key is balancing its all-round performance with the right price point.
For a larger facility, we’re now in a position where we can make a very focused decision, where the workstation is absolutely a specialist machine, excelling at either core or clock-speed intensive tasks.
Intel has taken this decision-making a step further with its Alder Lake and (the more recent) Raptor Lake CPUs. As we mentioned earlier, these use cores which followed the ARM model seen in mobile phones, which not only allowed the right core for the right job, but also at the best power requirement. Alder Lake CPUs saw 12th Gen i7’s and i9’s with combinations of P-Cores (Performance-Cores) and E-Cores (Efficiency-Cores) within one socket design.
As you’d imagine, the P-Cores have full multi-threaded capabilities in line with what we’ve seen from other 11th and 10th Gen processors. These are supported by the E-Cores; not multi-threaded or as performant, but able to be targeted for workflows that don’t require the computing power of the P-Cores. Additionally, as the name suggests, they are exceptionally low power to run.
The 13th Gen Raptor Lake CPUs with Windows 11 and Rocky Linux are starting to show us the benefits of this new architecture. The i9-13900KS starts to hit 6.00Ghz in certain scenarios with max turbo frequency, but all the Performance Cores exceed 5.00Ghz in base frequency. The E-Cores are no slouch at a Turbo frequency of 4.30Ghz either.
With eight P-Cores and 16 E-Cores you have a chip that allows excellent clock speed for applications and a good amount of cores for multi-threaded applications. And with this design only just beginning to hit its stride, it will be further refined going forwards.
Beyond hardware
As with all things hardware, there is also an OS and software implication. And here there are some issues to be aware of too.
When it comes to OS, Intel has ‘Thread Director’ which allows Windows 11 to take full advantage of these new combinations of specialist cores, effectively directing software to run on the best cores. Unfortunately, things become complicated when it comes to Windows 10. Intel’s own website makes the statement that “Thread Director works with the Windows 10 Scheduler, but is not optimised for it.”
Reading between the lines, and with our knowledge of studios up and down the UK, there are plenty which are still committed to running Windows 10. For those people, there’s a very real danger that the top-of-the-line workstations they buy based on specs, won’t perform up to anywhere near their capabilities, due to Thread Director’s inability to see the difference between the P-Cores and E-Cores. As a result, tasks that should be running through the P-Cores will in fact be taken on by the E-Cores and vice versa.
HPC CPUs: Threadripper and Sapphire Rapids
As impressive as Intel’s Alder and Raptor Lake processors are, they’re not built to compete directly with AMD’s Threadripper 32/64/96 multi-threaded CPU processor when it comes to simulation, rendering and machine learning.
Intel now has its own direct competitor, with the high-core-count, multi-threaded CPU Sapphire Rapids. This 4th Gen Xeon scalable-CPU allows up to 56 full cores on a single socket. With certain manufacturers still supporting dual-socket motherboards, it’s possible to have a dual 56-core Sapphire Rapids system delivering 112 cores, or 224 threads.
This development also opens the way for PCi 5.0 and DDR5 memory support, as well as accelerators such as IFS, DSA, QAT, DLB and IAA. But whilst the potential for even faster machines is alluring, it has to be tempered by the fact that not all accelerators are available on all models of the CPU. So again, making a decision without all of the information could result in disappointing performance.
How many cores is too many?
But with all of this potential power comes a question: how many cores is too many? We’re currently seeing that 32 cores seems to be the upper limit choice for most machines. Whilst desirable, 64 cores is often seen as too high a price point for the performance gains. This is particularly true for facilities where 32 cores is more than enough for 70% of the work. Is it really worth paying a premium for the 30% of the time where you’ll notice the 64-cores benefit?
Additionally, there’s the issue of how much power (in terms of electricity) it takes to run this many cores. With energy prices continuing to be a cause for concern and with energy-use as a whole being something we all need to consider, this extra factor is often the reason machines stick to 32 cores.
Admittedly the Raptor and Alder Lake architectures with their mix of P-Cores and E-Cores should address this problem, but with this CPU still fresh to the market, it remains to be seen just how effectively it will be able to do it.
So whilst the development of the CPU arena continues to evolve and excite, the dazzling array of choices can often result in analysis paralysis, or worse, making the wrong choice. But from sectors as diverse as VFX to architecture, these new CPUs are rich in potential, and we for one, can’t wait to see what these specialised processor cores can do when used in the right way.
--
For more insights like this sign up for our newsletter (below) or if you want to speak to someone in our sales team about how we can help with your workflow This email address is being protected from spambots. You need JavaScript enabled to view it.