Discussion Interesting test - GPU vs CPU upgrade

rumplestilts

Honorable
Feb 12, 2019
33
2
10,535
:LOL:
I'm not sure if this will provide any useful information but, perhaps, there's some sort of useful takeaway to this.

I have a decent system (HP Envy with Core i7-4790@3.6GHz, 24GB RAM, SATA SSD) that's served me well for at least four years. I upgraded the video card from the GTX 960 that it came with from the factory to a GTX 1660 Super (based upon some extremely helpful advice from a couple of TH members). The improvement in photo editing and Handbrake performance was tremendous. Okay; that's the back story.

I recently helped a client migrate from an old bottom of the line iMac to a PC and, based upon her budget, went with a nice HP with a Ryzen 7 4700G@3.6GHz, 16GB RAM, M.2 SSD and a GTX 1660 Super.

Maybe you can see where I'm heading. I figured why not test both machines with a Handbrake transcoding task? so that's what I did.

The task took just under 10 minutes on the Core i7 while it took about 6.5 min on the Ryzen. That's a pretty healthy improvement. I'm not inclined to think the 8GB of extra RAM on the Core i7 machine made any substantive difference. As Handbrake utilizes the GPU (if you tell it to) and that choice makes a massive improvement over not using the GPU, and Handbrake's task is, generally, read the next frame from the source, do any necessary conversion, drop the converted frame into the destination container, rinse and repeat until done (plus the audio conversion which, presumably, is more CPU than GPU directed - but I'll let others opine as I simply don't know), the improved performance of the Ryzen machine should be due to a number of things:
  1. The Ryzen's 8 cores over the Core i7's 4 cores. Handbrake's docs say 6 cores is optimum with diminishing returns for more cores.
  2. The Ryzen can ramp up to 4.4GHz while the Core i7 can do 4.0GHz.
  3. The Ryzen's single-core performance is about 20% faster than the Core i7 (in Passmark) so that may carry over to multiple cores in some fashion.
  4. The Ryzen's CPU Mark score is about 285% compared to a 100% rating for the Core i7. (Is this multi-core?)
  5. The Ryzen machine has an M.2 SSD while the Core i7 machine has an older 2.5" SATA SSD. I think this makes a big difference as I remember upgrading from a spinner HD to a SATA SSD improved everything regardless of whether it was my Core i7 PC or an old 2005 white MacBook. Handbrake's job is to read, process, and write over and over again so drive speed is critically important; the M.2 is at least 2X faster than the SATA SSD, IIRC.
Generally speaking, the Ryzen machine boots faster and is just snappier in everything I tried (web browsing, LibreOffice, Handbrake, even Windows Explorer). But where do you, dear members, think the improvement in performance resides?

Would next year's Ryzen 7 be worth the wait? (I know the answer here: Next year it will always be faster.)

Of course, there's always some budget consideration. Spending 1/3 the cost of the entire box for the video card seems ludicrous (especially since the 1660 Super seems to be providing excellent performance for LR and Handbrake). If you're a gamer, then you won't agree with the previous sentence.

Bottom line: This post isn't a cry for help (as I've been provided excellent help in choosing that 1660 Super for the upgrade I did). I had an opportunity to run a test with one important component being a constant in two machines built five years apart and I guess this shows how a combination of radically improved storage speed and CPU efficiency can make a huge difference given identical GPUs. It was an interesting test that cost me nothing but some time (as my client paid for the machine and it's now purring on her desk).

If you'd like to contribute to this thread, please do so. I'm not making any definitive pronouncements, just trying to provide a bit of clarity in comparison-shopping. :LOL:

Stay safe, everyone.

Barry
 

g-unit1111

Titan
Moderator
Of course it would boot faster, you're comparing a 2020 system to a 2014 system, things improve and change in 6+ years of computing. Everything will get faster and better in that length of time. But some other observations:

The Ryzen's 8 cores over the Core i7's 4 cores. Handbrake's docs say 6 cores is optimum with diminishing returns for more cores

That's certainly debatable. The thing is that video-based benchmarks like Handbrake are synthetic and they don't test the real-world hardware usage scenarios that a comprehensive benchmarking program like 3DMark would. You can certainly use those numbers as a basis for comparison but in reality they don't stress test the hardware the way that a professional benchmarking suite would. You'd probably get way different benchmarks depending on which software suite you would use.

The Ryzen can ramp up to 4.4GHz while the Core i7 can do 4.0GHz.

You're comparing the Ryzen 7 4600G, which is a laptop CPU, to a desktop i7-4790. That's a 6 year old desktop CPU vs a current laptop CPU. That's not really a very favorable comparison.

The Ryzen's single-core performance is about 20% faster than the Core i7 (in Passmark) so that may carry over to multiple cores in some fashion.

Again, you're basically comparing apples to oranges here, there might be some truth to that but I wouldn't read too much into it.

The Ryzen machine has an M.2 SSD while the Core i7 machine has an older 2.5" SATA SSD. I think this makes a big difference as I remember upgrading from a spinner HD to a SATA SSD improved everything regardless of whether it was my Core i7 PC or an old 2005 white MacBook. Handbrake's job is to read, process, and write over and over again so drive speed is critically important; the M.2 is at least 2X faster than the SATA SSD, IIRC.

There's not a whole lot of difference in read - write times between a standard issue 2.5" SSD and a 1st generation M2 drive. But current SSD read - write times have improved significantly since then. Try running an OEM M2 drive against something like a WD Black 750 or a 4th gen PCI-E drive like the Sabrent Rocket or Corsair MP600.
 

rumplestilts

Honorable
Feb 12, 2019
33
2
10,535
Thanks for your contribution to the discussion. I'll answer a couple of your points:

Handbrake: Synthetic? I don't think so; that was a real video. It's truly only one test but it's as real as you can get. 3DMark may cover many more testing situations but how many of us even know the list of specific tests included? Many of us use Handbrake so my test was valid for Handbrake. That's it. I wasn't intending my test to be any sort of comprehensive review of anything.

2.5" SATA vs M.2 data transfer speed: Two years ago, M.2 (NvME) was (theoretically) about 3500MB/s read and 2700MB/s write. Faster today, of course. SATA3 is about 600MB/s (taking into account overhead). The Sabrent Rocket you mentioned is spec'd at 3400/3000 read/write so it's similar enough to what HP is providing. Probably a bit faster since HP wouldn't be using high-end hardware. (Ah, wait! You were thinking the M.2 in the new HP wasn't NvME? I didn't specifically include "NvME" when I mentioned "M.2" so that may be it.)

One thing I neglected to mention was the RAM speed: The RAM in my old HP is PC3-12800 while the new one used PC4-3200 (IIRC). That's about 50% faster. Again, this is totally expected with newer hardware and, as that Handbrake frame has to be processed through that pipeline, it will be faster.

Cheers!