I CAN’T EXPLAIN THIS…… Ryzen RAM Speed Testing

Purchasing medications from online medical stores is also helpful in terms of avoiding embarrassments and problems normally faced by the customers while communicating with local http://appalachianmagazine.com/2017/06/09/2017-print-edition-of-appalachian-magazine-available-for-purchase/ levitra online medical store. Those symptoms cause both disorder of physical function and psychological strain and may also aggravate disease conditions Prostatitis, result in painful diseases to so many patients, seriously affecting the patient’s health and their family life. buy tadalafil But, if you clutch the recommended site or reputed one, you will never ever be disputed with the quality,so try viagra discount as it is completely safe for most individuals without any serious side effects. A physician will always Get More Info levitra prices recommend for a non-invasive treatment, vacuum therapy system is your pick.

For the life of me, I can’t explain this. For the last few days I’ve been benchmarking this Ryzen 5600X system with different memory speeds to see how that affects performance. Now the gaming results mostly make sense, but where I’m stumped completely is the CPU specific benchmarks. Let me lay out what I’ll be testing, then for the sake of my sanity I’ll start with the gaming numbers.

First off, what are we testing? Well, we are testing different memory frequencies to see how they improve (or remove) performance. AMD claims the sweet spot for these Ryzen CPUs is 3600MHz, so that’s where I’ll start. I’ll also test 3400, 3200, 3000, 2666 and 2133, with the last one being the speed your memory will run by default until you enable XMP – this is a good point to open up task manager and check your memory is actually running at the speed you paid for! I’ll be testing a selection of games, and CPU specific programs like Cinebench, Blender and the Adobe CC Suite. Got it? Great. Let’s look at the gaming numbers.

Starting with Watchdogs Legion, all of the frequencies except 2133 are all pretty close. Excluding the slowest, it’s just 8 FPS between 2666, and 3600 with a nice linear progression down the ranks. The only outlier is 2133 which is a further 10FPS slower than even 2666. You get 18FPS more by enabling XMP on a 3600MHz kit, or over 25% more performance. Enable XMP NOW!

In Cyberpunk it’s pretty similar, with 3600 taking the lead with 89FPS, then a bit more noise in the numbers, but well within margin of error for Cyberpunk. You can be driving through a tunnel and get 50 FPS and then have over 100 the split second you exit, so even with multiple runs it’s safe to say you wouldn’t notice a difference in your playing experience too hard, except for 2133 where again it’s around 8 FPS slower than the second slowest, and 20 FPS slower than the fastest. Or, to put it another way, setting your memory to 3600 instead of 2133 nets you nearly 30% more performance! Mental!

CSGO doesn’t seem to care much for memory speed, perhaps since it’s a relatively lightweight game compared to the rest I’m testing. The fastest was 3400 by 2 FPS, then 3600, then 2666, with 3200, 3000 and 2133 all sharing the 330 mark. There isn’t much in it between these especially in terms of percentage, so not a big deal.

Fortnite does hold a bit of an advantage for our sweet spot 3600 kit, offering 6 FPS more than the next best, with 3400, 3200, 3000 and 2666 offering almost identical performance, then 2133 coming plum last with almost 10 FPS less performance than second worst. It’s impressive just how much you lose with the base rate frequency!

And finally, in Microsoft Flight, it’s back to a nice linear progression. 44FPS isn’t much to shout about, but considering you get almost 10 FPS less using 2133 instead over 3600, that’s 25% more performance potentially for free if you’ve already paid for, but haven’t enabled it. The rest follow a smooth 1 or 2 FPS drop each time.

So, in games on this Ryzen 5600X the faster memory you have, in general, the more performance you will get. It’s not an insane amount, and if you’ve got a system like this and ‘only’ have 3000 or 3200 you can rest easy knowing you aren’t losing out on a significant amount of performance. If you are building new today, 3600 isn’t (currently) drastically more expensive than 300. or 3200 so it’s likely worth splashing out for. But, if you already have a build it’s definitely not worth upgrading unless you are also upgrading your capacity, say going from 8GB to 16GB, or 16GB to 32GB.

Now we get to the confusing part. I should make it clear I’ve run all these tests multiple times, over multiple days. While I haven’t been able to control ambient temperature – it’s sweltering here at the moment – I’ve tried to mitigate that by running across multiple days at different points in the day as well. Right, enough excuses, let me show you what’s wrong here.

In Cinebench R20 single threaded, it all makes sense, they are pretty much the same which considering it’s just a single thread workload you aren’t really going to be able to utilise the memory all that much. That’s fine. What’s not fine are the multi-threaded results. Yes, you are reading that right. 2666 and 2133 are a full 200 points higher than 3600. That’s 5% faster, with SLOWER memory. I cannot fathom why. Temps, power usage, everything were all within margin of error. Like I said I’ve ran these number multiple times, but for whatever reason the slower speeds are… faster.

This isn’t exclusive to Cinebench either, Blender has the exact same problem. In the BMW scene 3600 ties with 3000 for the slowest, with 2133 coming out on top, a full 16 seconds faster than 3600. At least in Gooseberry it’s a little closer, although technically 2666 is the fastest again, although only by 2%. I don’t know what kind of black magic is going on at 2666 and 2133 for these tests, if you have any ideas please do let me know.

Using the Puget bench suite for Premiere Pro the results just about make sense. All bar 2133 are pretty tight scoring, with 3400 taking the lead, and 2133 dropping off a touch at the back. It’s not a massive difference, and unlikely to be one you’d notice, but still. In After Effects it’s a touch more sensible, although again offers a very tight grouping so little in the way of benefits to any particular kit, and Photoshop is the most sane with a nice linear progression down, although only a 7.5% spread so hardly significant.

So it seems, from my testing anyway, that RAM speed doesn’t have much of an effect on performance in CPU specific benchmarks. It should, but apparently it doesn’t. Yeah, no idea. At least in gaming there is a clear correlation between faster RAM speed and better performance.

The main point to note here is that I didn’t alter the memory timings at all between runs. They were what the XMP profile set, so if you did have a say 3000MHz kit you may be able to get a lower CAS latency than you would at 3600 which might make up for a touch of performance on that sort of kit.

If you’d like to see this exact test on a new Rocket Lake CPU from Intel, specifically an 11400F, then go check that video out in the cards above or on the end cards – that makes a lot more sense than this data does so go check it out!