• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

Dolphin benchmark weirdness

WildW

Senior member
I'm looking to replace my HTPC, currently an old Ivy Bridge 3570k on a slowly dying H61 motherboard so not even overclocked. I use the Dolphin emulator and would like better performance, and have heard it runs better on newer architectures. It's also supposed to only really use two threads so was considering a higher clocked pentium like the G4600 (Kaby Lake).

I looked at the benchmarks at http://www.anandtech.com/bench/CPU/1027 and the numbers look all screwy to me. It's showing i3 7350K @ 4.2 GHz as slower than a Skylake i5 6600K @ 3.5 GHz, and even slower than Broadwell 5775C @ 3.3 GHz. In general 4C/8T does better than 2C/4T, which I do not expect to see with Dolphin. Is there a real advantage to having more than two cores here?
 
Last edited:
Dolphin uses instructions not available before Haswell.

So it has a much bigger boost in performance than most other software.
 
I'm looking to replace my HTPC, currently an old Ivy Bridge 3570k on a slowly dying H61 motherboard so not even overclocked. I use the Dolphin emulator and would like better performance, and have heard it runs better on newer architectures. It's also supposed to only really use two threads so was considering a higher clocked pentium like the G4600 (Kaby Lake).

I looked at the benchmarks at http://www.anandtech.com/bench/CPU/1027 and the numbers look all screwy to me. It's showing i3 7350K @ 4.2 GHz as slower than a Skylake i5 6600K @ 3.5 GHz, and even slower than Broadwell 5775C @ 3.3 GHz. In general 4C/8T does better than 2C/4T, which I do not expect to see with Dolphin. Is there a real advantage to having more than two cores here?
Those 3 have almost the same score. 7350K tops out at 4.2Ghz, 6600K tops out at 3.9Ghz, and 5775C tops out at 3.7ghz, and they have cache differences per core, and the 5775C has 128mb of L4 cache which the other two lack.

Here is another more recent set of benches. I can't see building a new system today with less than 4 full cores. If it were my system, for longevity, I wouldn't go less than 4C/8T.

http://images.anandtech.com/graphs/graph11083/85507.png
 
I hadn't considered the turbo clocks, I guess that does make them closer. Still, it seems more than 2 cores does make a difference - I checked Dolphin's FAQs and it looks like it does use 3 cores after all.

I guess I'm in that familiar situation that I have a perfectly good Intel CPU that I wouldn't consider replacing if the motherboard wasn't giving me trouble, but nobody has made new motherboards for it in years.
 
I guess I'm in that familiar situation that I have a perfectly good Intel CPU that I wouldn't consider replacing if the motherboard wasn't giving me trouble, but nobody has made new motherboards for it in years.
The selection of good, affordable 1155 motherboards is really small, but that's to be expected this many years later. I have several 1155 systems that I would like to shrink into mITX builds, but no boards are available (for reasonable prices).

If you're changing motherboards anyway, look at a Ryzen 1400/1500/X 4C/8T or 1600/X 6C/12T. Dolphin scores are excellent, you get tons of cores/threads, and they're cheap (relative to Intel).
http://www.anandtech.com/bench/CPU/1617
 
I have several 1155 systems that I would like to shrink into mITX builds, but no boards are available (for reasonable prices).

$30 reasonable?
http://www.ebay.com/itm/Intel-DH61D...511174?hash=item51e2729586:g:c7sAAOSwBnVW-yoB

MATX Lenovo boards are $18..

http://www.ebay.com/itm/Lenovo-Thin...392397?hash=item44043584cd:g:QKwAAOSw~CFY7S80

Ascendtech has new B75 Acer MATX boards for $46

http://www.ebay.com/itm/ACER-DESKTO...547940?hash=item3ac2937964:g:a5UAAOSwTM5Y3pVw
 
Last edited:
I guess if you trust used eBay sellers? No IO shield or other accessories included. I've had mixed results that often end up costing me more money and time, but YMMV.
Obvious screw up? NO?! You would expect ryzen to be close to broadwell-e that has similar clocks not well above it.
What's an obvious screw up? I'm confused by your post.
 
Dolphin uses instructions not available before Haswell.

So it has a much bigger boost in performance than most other software.

The reason why Dolphin gets a huge jump with post-IB CPUs, is that Haswell brought a much faster cache. About 2x faster than Sandybridge's.

The reason why that's important, is that emulators reuse instructions like no tomorrow. Small number of instructions, huge number of offsets. Faster cache = faster instruction & offset retrieval.
 
If you're changing motherboards anyway, look at a Ryzen 1400/1500/X 4C/8T or 1600/X 6C/12T. Dolphin scores are excellent, you get tons of cores/threads, and they're cheap (relative to Intel).
http://www.anandtech.com/bench/CPU/1617

That set of benchmarks tells me a couple of interesting things:

(1) My Ivy Bridge i5 must be, at best, level with the 4960X, i.e. hopeless compared to anything modern.
(2) Anything modern with stock clocks is much better than what I have. Hmm, perhaps these two things are just one thing really.

Looking at Ryzen, the cheapest Ryzen 5 1400 is almost twice the price of a Pentium G4620 (160 vs 86 Great-British Currency Units), and motherboards are about the same (Intel slightly cheaper). Granted they're not equal in overall performance but this machine plays Youtube and Mario Kart.
 
That set of benchmarks tells me a couple of interesting things:

(1) My Ivy Bridge i5 must be, at best, level with the 4960X, i.e. hopeless compared to anything modern.
(2) Anything modern with stock clocks is much better than what I have. Hmm, perhaps these two things are just one thing really.

Looking at Ryzen, the cheapest Ryzen 5 1400 is almost twice the price of a Pentium G4620 (160 vs 86 Great-British Currency Units), and motherboards are about the same (Intel slightly cheaper). Granted they're not equal in overall performance but this machine plays Youtube and Mario Kart.
I think the difference is your initial Dolphin benchmark is older while the one I posted is v5.0 which is a phenomenally newer, optimized version. Either way, you're right about the price, but that's why I suggested the R5 1400 as an option because it will blow the doors off the i3 7350 for only a little more (depending upon location/stock/sales, I guess), but offers 4C8T performance for longevity. Then again, Kaby Lake has merit in a HTPC environment for some compatibility with Netflix 4K and UHD Blu-ray if you go that route...
 
That set of benchmarks tells me a couple of interesting things:

(1) My Ivy Bridge i5 must be, at best, level with the 4960X, i.e. hopeless compared to anything modern.
(2) Anything modern with stock clocks is much better than what I have. Hmm, perhaps these two things are just one thing really.

Looking at Ryzen, the cheapest Ryzen 5 1400 is almost twice the price of a Pentium G4620 (160 vs 86 Great-British Currency Units), and motherboards are about the same (Intel slightly cheaper). Granted they're not equal in overall performance but this machine plays Youtube and Mario Kart.
The Ryzen 5 even the 1400 are meant to be competitors with the i5 series. There are 4 core Ryzen 3 series coming out July or later that will be priced as low as an R3 along with later this year Zen+Vega based APU's.
 
The reason why Dolphin gets a huge jump with post-IB CPUs, is that Haswell brought a much faster cache. About 2x faster than Sandybridge's.

The reason why that's important, is that emulators reuse instructions like no tomorrow. Small number of instructions, huge number of offsets. Faster cache = faster instruction & offset retrieval.
Oh neat, I've definitely read it was due to AVX2 in the past but looking it up now it seems that was wrong. Blindly trusting non-dev posters on the dolphin forums was probably foolish.
 
Oh neat, I've definitely read it was due to AVX2 in the past but looking it up now it seems that was wrong. Blindly trusting non-dev posters on the dolphin forums was probably foolish.

Yeah, it's not so much new instructions. PCSX2 and other emulators that don't take advantage still saw large performance boosts on haswell.
 
Yeah, it's not so much new instructions. PCSX2 and other emulators that don't take advantage still saw large performance boosts on haswell.

AVX instructions would only be usable when the emulator is running a software renderer. And there's no reason to use a software renderer when there's a hardware renderer, barring a handful of problematic games that glitch out with HW rendering.
 
AVX instructions would only be usable when the emulator is running a software renderer. And there's no reason to use a software renderer when there's a hardware renderer, barring a handful of problematic games that glitch out with HW rendering.

I believe the PS2 had a pretty extensive SIMD instruction set. AVX might be useful somewhere in there. In fact, I just found a dev commenting on it, saying the same thing.
 
I believe the PS2 had a pretty extensive SIMD instruction set. AVX might be useful somewhere in there. In fact, I just found a dev commenting on it, saying the same thing.

The PS2 has two vector units (whatever that means), called VPU0 and VPU1. That's probably where AVX would help.
 
The PS2 has two vector units (whatever that means), called VPU0 and VPU1. That's probably where AVX would help.

I know about em lol. But pcsx2 doesn't use AVX for them, at least not yet. He also mentioned that there's something that makes it hard to apply there.
 
Last edited:
Back
Top