Hyper Threading vs Turbo Boost: Do translators need them?

Today we’re trying to give an indicative answer to the question “should I get an i7 processor if I’m a translator and intend to use my computer mostly to work with CAT tools?” This has been a common staple of translator communities all over the world, with people debating if Intel’s top offerings are worth it or not for the workloads a translator throws at them. I wanted to do this test for a while now, but only recently I finally upgraded my computer with a Xeon E3 1231-v3 processor from a Core i5 4570. My other computer was a laptop with a Core i5 6600U, which supports both Hyper Threading and Turbo Boost, but its BIOS didn’t allow me to disable them, preventing me from running them. For those who don’t know, entry-level Xeon E3 processors are basically i7s without an integrated GPU sold at the price of an i5, which makes them (or rather, used to) a viable upgrade for those who want i7 performance and don’t care about the integrated GPU because they already have a dedicated graphics card. However, since the release of sixth generation Core processors, Intel has made it a requirement to use Xeon processors with specialised motherboards with the C230 chipset family.

I mentioned that these results are only indicative because I don’t have access to every single CAT tool in existence, but they should give us some insight as to whether higher-end processors are worth purchasing if all we intend to use our computers for translating files.

What are Turbo Boost and Hyper Threading?

We should spend a few words explaining what these two technologies are and how they affect performance, because this website is meant for people who aren’t very tech-savvy and Intel advertises both technologies rather aggressively.

Turbo Boost is a technology that lays its foundation on the previous (and still used) SpeedStep technology, introduced in 2005. SpeedStep dynamically regulates the voltage provided to the processor, which in turn determines the clock speed, or how many times the information contained within the processor changes in a second. This allows for decreased power consumption when the system is idling or doesn’t need much processing power. But if you can dynamically decrease the voltage, this also means you can dynamically increase it. This is what led to the creation of Turbo Boost, first released with first generation Core i5 and i7 processors in 2008. When a computer with a processor that supports Turbo Boost requires more processing power, it’s allowed to run at higher frequencies than it normally would, provided its temperature doesn’t go too high. If the heat dissipation method in your computer allows it, a Core i5, i7, or Xeon processor could theoretically work at turbo frequencies indefinitely, but in reality there is no 100% efficient dissipation method, meaning that, sooner or later, your processor’s clock speed will have to go down to allow temperatures to fall below a certain threshold.

Turbo Boost is very useful in portable devices like laptops and tablets, where it isn’t uncommon to see a processor with low clock speeds but very high turbo speeds, which come in handy when more computing performance is needed in short bursts, such as when launching an application.

Hyper Threading is Intel’s implementation of Simultaneous Multi Threading (SMT), which allows two threads (basically, a sequence of instructions) to run at the same time on the same processor core. These two virtual cores are slower than a physical one, but in workloads that benefit from the execution of multiple instructions at the same time, they can help with performance. Think about the simple mathematical formula:

A+B+C+D

One way we can solve it, is adding the values sequentially. But we can also divide it into three operations, two of which are executed at the same time, like this:

Thread #1: A+B
Thread #2: C+D

Thread #1 + Thread #2

This way, we saved a little bit of time, even though the two threads are a little bit slower.

Hardware specifications and testing methodology

All the test in this roundup were performed on a computer with the following specifications:

CPU: Intel Xeon E3 1231-v3 @ 3.40 GHz
Motherboard: Asus B85M-G
RAM: 16 GB DDR3 @ 1600 MHz
Storage: Crucial MX100 256 GB SSD + 2 x WD Blue 1TB hard drives in RAID 1 configuration
Graphics Card: MSI GTX 960

I’m mentioning the graphics card not because it directly influences the performance we can expect from the tests, but because two of the programs we used, MemoQ and IATExtract, do not show the time taken to perform operations. To work around this, I recorded what was happening on the desktop using NVIDIA Geforce’s Share functionality, which has a negligible impact on the CPU, since it’s the graphics card that’s doing most of the work.

We’ll be running the following tests:

  • Translation memory files (in TMX format) merging using Rainbow, a program that belongs to the Okapi Framework. The memories we’re merging are the 30088 files found in the DGT translation memory, that contains the whole corpus of EU laws.
  • Termbase (in TBX format) extraction. We’re going to extract the terms for English, German, French, Italian, Spanish, and Portuguese from the IATE glossary using a tool developed by the EU itself called IATExtract.
  • Termbase import in MemoQ from the file generated by IATExtract to MemoQ’s own termbase format.

Results

TMX Merging in Rainbow

The TMX merging test yielded the following results:

TMX Merging test results. Lower is better.
TMX Merging test results. Lower is better.

It’s clear that Okapi doesn’t benefit much from Hyper Threading, whereas Turbo Boost gives a very meager performance improvement in this particular task. It completed 20 seconds earlier with both options enabled compared to running it with both disabled. I can’t say that an Hyper Threaded processor is worthwhile for this kind of task.

I had a chat with Okapi developers and they admitted that Rainbow executes its pipeline inside a single thread, meaning that performance benefits derived from Hyper Threading are minimal. What is more important here is raw processing power.

TBX Extraction

The extraction of the terms for 6 languages was actually quite surprising to me. It’s common knowledge that two threads running on two virtual cores on a Hyper Threaded processor are slower than a single thread running on a physical core, but these results defied all of my expectations.

IATE Glossary extraction in 6 languages. Lower is better.
IATE Glossary extraction in 6 languages. Lower is better.

That’s right. With Hyper Threading enabled and Turbo Boost disabled, the extraction was slower, compared to every other scenario. I thought this was an erratic value and to make sure I re-ran the test twice, yielding the same result. We can conclude that Hyper Threading is actually detrimental in this scenario.

TBX Import

Now, we’re importing the TBX we created with IATExtract into MemoQ. The results of this test are comparable to that of TBX extraction, with our artificial “Core i3” processor being slower compared to the other scenarios, even though only slightly.

IATE glossary import in MemoQ. Lower is better.
IATE glossary import in MemoQ. Lower is better.
Conclusion

This test is clearly not all-encompassing. It was spur primarily out of my curiosity to see if Hyper Threading directly impacts performance on the workloads I subject my machine to. If you use your computer to do things like graphics design, video editing, desktop publishing, CAD design, then an Hyper Threaded processor is going to be a definite no-brainer for you to purchase. But if you intend to use your machine primarily for processing text files, like a translator would, then a Core i5 would be a better value for your money.

As always, my suggestion for upgrading your processor or purchasing one for your new computer build is to always get what you can comfortably afford without breaking the bank. Whatever model you choose, just know that Intel processors are very good products that will prove useful for many years to come.

About Andrea Luciano Damico 137 Articles
Andrea Luciano Damico is a freelance translator from Italy. Among his interests are linguistics, technology, video games, and generally being a chill guy. He runs Let's Translate.it and Tech4Freelancers.net.