Anirudh RegidiAug 12, 2021 15:26:09 IST
As we saw in part I, AMD’s Zen 3 microarchitecture has finally, and definitively, overtaken Intel as the more efficient and performant option for high-end desktop PC users, especially if productivity is important. Let’s now look at the performance of AMD’s various Zen 3 CPUs and zero in on the chip that’s best for your needs.
Before we dive into the test results and analysis, here’s how Team Red’s (AMD) best stacks up against Team Blue’s (Intel) best:
|Model||Cores/Threads||Turbo Freq. (GHz)||TDP (W)||Price (INR)|
|Intel Core i5-11600K||6/12||4.9||125||25,150|
|AMD Ryzen 5 5600x||6/12||4.6||65||27,250|
|Intel Core i7-11700K||8/16||5.0||125||38,100|
|AMD Ryzen 7 5800x||8/16||4.7||105||39,300|
|AMD Ryzen 9 5900x||12/24||4.8||105||54,750|
|Intel Core i9-11900K||8/16||5.3||125||60,900|
|AMD Ryzen 9 5950x||16/32||4.9||105||73,899|
What’s immediately obvious is that Team Blue’s chips are all rated at a higher TDP than AMD’s, i.e. they draw more power. And when it comes to core counts, Team Red is far ahead.
Team Blue’s only advantages here are clock speed and a slight edge in pricing with the i5 and i7 parts. However, as discussed in part 1, those blistering speeds (5.3 GHz is a big deal) come at a high cost and will require an expensive cooling setup.
The same goes for pricing. AMD’s entire lineup will run just fine on mid-range B550 motherboards that cost around 15k. The more expensive X570 boards add more features and support more PCIe Gen 4 lanes for additional GPUs and NVMe storage, but that’s about it. X series AMD CPUs can be overclocked on B550 boards, and you’re allowed to use fast RAM.
While Intel’s K series CPUs can run on B560 motherboards that also retail around 15k, and are the first of Intel’s x60 series chipsets to support memory overclocking (thank you, AMD), you will not be able to overclock your CPU, which is the whole point of getting a K series CPU to begin with. To do that, you’re forced to buy premium Z590-chipset based boards that costs an additional 10-20k. From a value perspective, this makes little sense.
With that out of the way, here’s how my test rig was configured:
Motherboard: For the motherboard, we used the MAG B550M Mortar from MSI. Retailing at around 15k, this board supports 4 sticks of DDR4 memory at up to 4,400 MHz, comes with dual PCIe x16 slots (x16 + x4) and two M.2 slots (one of each is PCIe Gen 4) and supports USB 3.2 Gen 2. The board held up well in testing and allowed all CPUs to run at full tilt without issues.
RAM: We went with four sticks of Corsair Vengeance Pro RGB RAM for a total of 32 GB. The RAM was rated at 3,200 MHz C16. The RGB was, of course, only for bling (And why not?).
PSU: Our trusty Corsair AX850 powered this rig. Rated for 850 W of power and an 80+ Titanium efficiency rating, this modular PSU didn’t have any trouble keeping our hot, power-hungry components running normally.
GPU: For the GPU, AMD sent their best, i.e. the 6900XT for testing. This beast of a GPU comes with 16 GB of VRAM and a whopping 5,120 stream processors, not to mention 128 ROP and 320 texture units. A recent update adding support for DXR (a Microsoft API for ray-tracing) means that these GPUs are finally at par with Nvidia’s RTX cards on that front.
Storage: We had to go with Corsair’s MP600 here. This PCIe Gen4 compatible NVMe drive is one of the fastest drives you can get and is capable of hitting read/write speeds in excess of 5 Gbps. All apps and games were installed on this drive. A Samsung 870 QVO and a pair of WD Red NAS drives were used for storing captured data and footage.
Cooling setup: To keep these CPUs cool, I used my NZXT Kraken x53 CPU liquid cooler. For thermal paste, I went with Arctic Silver MX4, and chassis cooling fans included two 120 mm Noctua NF-F12s and a single NF-P14 140 mm fan. The Noctua’s were configured as exhaust and the x53 as an intake at the front of the chassis.
Chassis: The hardware was mounted in a Corsair Obsidian 750D full-tower cabinet. This beast of a cabinet is an old design, but it’s still one of the roomiest, airiest cabinets you can get today. It also happens to be one of the few cabinets that I can mount a RAID 5 HDD array in.
Monitor: I used my BenQ EX2780Q for the display. This 2K 144 Hz panel supports G-Sync and FreeSync and is rated for HDR400. It’s fast, accurate, and more importantly, supports 4K 60 Hz when necessary.
How we tested
- We used a fresh install of Windows 10 with all necessary drivers and updates installed.
- Apps and games were installed on the Corsair MP600 SSD to eliminate any storage bottlenecks.
- Chassis fans and the x53 cooler were configured for performance mode.
- All tests were looped three times to ensure consistency.
- Gaming benchmarks were logged via CapFrameX. This was the only app explicitly allowed to run in the background when testing games. All games were tested at 1080p since our focus was on CPU performance and not the GPU.
Performance analysis: Gaming
Metro Exodus Enhanced Edition: This is the only AAA game so far that has dumped raster graphics in favour of a fully ray-traced environment. The game looks incredible, and understandably, brings even the best GPUs to their knees. Since this is a GPU-bound game, we don’t expect to see much of a difference in performance between the CPUs. This was exactly what we saw in our results.
Average frame-rates varied between 88.6 fps on the 5600x and 92.6 on the 5900x. 1 percent lows, an indicator of stability, were steady at 56 fps.
As can be seen from the frame-time plot above, the 5900x and 5950x showed some slight very instability compared to the 5600x and 5800x. Perhaps this can be attributed to some sort of latency issue when using dual CCXs.
Dropping DXR quality to medium, we see frame rates jump to 130+ on all CPUs. The 5600x scored the lowest here, managing to pump out a 131-fps average. The 5950x managed 133, and the remaining CPUs managed to push 139. It’s not an insignificant difference, but 1 percent lows were consistent across all CPUs and to a normal user, that 8 fps difference will go unnoticed.
Red Dead Redemption 2: Another GPU bound game this, but one that supports both DirectX 12 and Vulkan APIs. Since the other games were using DirectX 11 and 12, we used this game to test Vulcan performance.
Again, there’s no perceivable difference in performance between these CPUs. The 5600x managed the lowest avg fps of 107 while the 5950x managed 114. Interestingly, the game seems to prefer 8-core CCXs for some reason, with the 5600x and 5900x (using 1x 6-core CCX and 2x 6-core CCXs respectively) hitting 1 percent lows of 60, while the single and dual 8-core CCX-toting 5800x and 5950x hit 65. The difference is insignificant when gaming, of course.
F1 2019: This is an old game, but it’s one that’s heavily dependent on IPC and clock speed. There’s a clear “loser” in this test, but no clear winner.
The “loser”, the 5600x, managed 239 fps to the 5800x’s 256. The other two CPUs managed to get above 250 fps as well. At worst, this is a 7 percent difference in performance, but we’re also talking about 200+ fps here. Looking at 1 percent lows, we see the same discrepancy, with the 5600x managing 160 while all other CPUs stayed above 170.
Is it worth paying Rs 12,000 for a 7-10 fps boost in performance in a game that’s already running at 240 fps? I think not.
With Horizon: Zero Dawn, Battlefield V and Counter Strike: Global Offensive, we see a similar story. The 5600x is slower in CPU-bound titles, but not by much. In GPU-bound games, there’s next to no difference in performance. At higher resolutions, where the GPU is more important, this difference disappears.
Conclusion: If you’re gaming, don’t waste your money on more cores
- If you only intend to game on this setup, go with the 5600x and invest the money you’ll save into a more powerful GPU and/or faster RAM.
- If you intend to stream while gaming, I’d recommend the 5800x. The four additional threads and additional thermal overhead will leave you with enough headroom for high-resolution streaming.
This is where things finally get interesting, and these CPUs come into their own.
In PCMark 10 and Speedomoter 2.0, benchmarks that indicate performance in general computing tasks and web browsing respectively, there’s next to no difference in performance. The same goes for image editing in Photoshop and Lightroom. The 5900x and 5950x do show a 13-14 percent increase in performance over the 5600x in Lightroom, but you’ll end up paying an additional Rs 27,000 for that privilege. You could literally buy a second 5600x for that price. Is it worth it? You decide.
If you’re editing video, compiling code, or working with 3D objects, performance scales linearly with core count. When using Blender – an advanced 3D modelling and animation tool – the 5950x is 2.5x faster than the 5600x. When transcoding video in handbrake or compiling code, you can expect a 70 percent uplift in performance with the 5950x. The 5600x is a good performer, but it’s outclassed by everything else in the line-up.
Thermals, power draw, and overclocking performance
On average, these CPUs all tended to operate in the 40°C to 80°C range depending on whether I was gaming or stress-testing. Temperatures never ventured into the 90s, and unlike Intel and it’s power-hungry 14 nm node, power draw remained relatively low with only the 5950x came close to crossing the 200 W barrier.
Do note that I didn’t manually overclock these CPUs. With Intel, overclocking is as simple as adjusting the CPU multiplier and making minute voltage tweaks till you get stable performance. With Zen 3, it’s not that straightforward. You’re dealing with many more cores and performance needs to be optimized based on workload. You’re basically looking at a vast chart of clock speeds and voltages that need to be tweaked on a per-core basis. It’s a nuanced process that takes a great deal more time and effort than it does on Intel chips.
As someone who’s lived with Intel CPUs his whole life, AMD’s options are daunting. If, like me, you’re not yet comfortable enough with manually overclocking Zen 3 chips, AMD’s Ryzen Master software comes with an Auto OC option that does a decent enough job of pushing these chips higher. Auto OC pushed the turbo frequencies of all CPUs up by 100 MHz on average, with the 5800x seeing the biggest boost of 150 MHz.
Bottom line: Are you a gamer or a content creator?
I’ll keep this simple:
- If you need a powerful CPU for gaming and general work, you don’t need anything more than a 5600x.
- If you’re dabbling in video editing, 3D rendering, coding, or livestreaming, it’s worth picking up a 5800x. The additional cores and higher clock speed will significantly speed up your work. Unless of course you only work with photos, in which case the 5600x is enough.
- If you’re serious about video editing, 3D rendering and other CPU-intensive tasks, pick a CPU that fits your budget. Just remember: More cores=more performance. Don’t bother with the 5600x.
Editor’s note: We’d like to extend our thanks to AMD, Corsair and MSI for providing the hardware that made this review possible.