Vega 20 performance data

[imrazor]

So I'm considering getting an iMac and pondering the costs vs. benefits of the 21.5" model and the 27". One piece of info caught my eye; the 21.5" is available with a Vega 20 GPU. The problem is I can't seem to find any performance info for this GPU. Usually I rely on PC benchmarks for comparison's sake. (I know that under Windows the hardware performs differently, but to comparing the relative performance of GPUs usually works. That is, unless they're brand spanking new with poor driver support - Radeon VII.)

Since I can't find benchmarks for the Vega 20, I can't meaningfully compare it to a 560X, 570X, 580X or a Vega 48. Can someone point me to some benchmarks (PC or Mac) that will help me get an idea of the Vega 20's relative performance?

 

[fpenta]

So I'm considering getting an iMac and pondering the costs vs. benefits of the 21.5" model and the 27". One piece of info caught my eye; the 21.5" is available with a Vega 20 GPU. The problem is I can't seem to find any performance info for this GPU. Usually I rely on PC benchmarks for comparison's sake. (I know that under Windows the hardware performs differently, but to comparing the relative performance of GPUs usually works. That is, unless they're brand spanking new with poor driver support - Radeon VII.)

Since I can't find benchmarks for the Vega 20, I can't meaningfully compare it to a 560X, 570X, 580X or a Vega 48. Can someone point me to some benchmarks (PC or Mac) that will help me get an idea of the Vega 20's relative performance?

I have a MacBook Pro 2018 with Vega 20 and a brand new iMac with Vega 48. If you want I can compare them both
What benchmark would you want to see?

 

[imrazor]

Thanks for the info. The exact benchmark depends on if you have Bootcamp on both machines. If you're limited to OS X, one of the Unigine benchmarks (Heaven or Valley) would probably be best. Given the age of both benchmarks, just run the "Extreme HD" preset.

One potential problem is that the MacBook may be more thermally constrained than a 21" iMac. But it'll still be a useful point of comparison.

 

[AndreeOnline]

Are you looking specifically for gaming benchmarks?

If compute numbers are of value, you have Geekbench 5 filtered on 'Metal':

Geekbench 5 Metal

You can't compare these number to Geekbench 4.

A Radeon Pro Vega 48: 45800
A Radeon Pro Vega 20: 23300

Refer to link above for more comparisons.

 

[Sam Marks]

This is a benchmark of the Radeon Pro Vega 20 vs Vega 48.

Pro Vega 48 vs Pro Vega 20 [3-Benchmark Showdown]

Comparing Pro Vega 20 with Pro Vega 48: technical specs, games and benchmarks.

technical.city

This is the passmark benchmark, and comparison with other GPUs.

PassMark - Radeon Pro Vega 20 - Price performance comparison

 

[imrazor]

I was specifically looking for gaming benchmarks, but I decided to take another route. I bought an old school Mac Pro for significantly less money, and I can drop in any GPU I want. Right now I have an RX 580 in it - if anyone's interested I can try a few benchmarks for comparison.

 

[Sam Marks]

I was specifically looking for gaming benchmarks, but I decided to take another route. I bought an old school Mac Pro for significantly less money, and I can drop in any GPU I want. Right now I have an RX 580 in it - if anyone's interested I can try a few benchmarks for comparison.

Excellent decision. I need a 2019 i9 iMac but I am deciding between Vega 48 and Radeon Pro 580X, so these RX580 benchmarks would be very interesting for me, thanks.

Here you have the Vega 20 gaming benchmarks:

AMD Radeon Pro Vega 20

A thorough insight into technical specs and benchmarks of Pro Vega 20.

technical.city

 

[imrazor]

Funny, just after I posted that I ran Geekbench's compute bench. I got just a hair under 36000. Not as fast as the Vega 48, but not too far off, and quite a bit quicker than the Vega 20.

To compare another number, in the Fire Strike graphics benchmark, my RX 580 gets about 13500 (though that's in a gaming PC rather than the Mac Pro.) Quite a bit quicker than the Vega 20, which seems to sit around 8000. That surprises me some - for some reason I was expecting the Vega 20 and RX 580 to be relatively close.

One caveat - for some reason the Mac Pro recognizes my RX 580 as an RX 480, and the RX 580X is likely clocked faster than my GPUs 1266MHz default speed.

 

[Sam Marks]

Thank you for the information on the RX 580, very interesting numbers. I also expected that the Vega 20 would be close to the RX 580. That is a nice surprise.

MSI MS-7971 - Geekbench

Benchmark results for a MSI MS-7971 with an Intel Core i7-7700 processor.

browser.geekbench.com

 

[imrazor]

Looks like my guess was a bit off. Here are some technical specs on the Radeon Pro 580X.

AMD Radeon Pro 580X Specs

AMD Ellesmere, 1200 MHz, 2304 Cores, 144 TMUs, 32 ROPs, 8192 MB GDDR5, 1710 MHz, 256 bit

www.techpowerup.com

If you go to the comparison chart, you'll see that the RX 580 is actually rated 12% faster than the Radeon Pro 580X. Clock speeds for the Radeon Pro 580X are shown to be 1100 - 1200 MHz which is actually slower than my reference RX 580 which runs at 1266MHz stock.

Hope that helps...

EDIT: One other note, I was wrong above. The Vega 20 gets about 9000 in Fire Strike, not 8000. Still way short of the RX 580.

 

[Sam Marks]

Yes, you are right. The Pro 580X is a workstation grade GPU not a gaming one, so it is clocked lower than its gaming equivalent and it has less TDP, so it is slower but more efficient and realiable for long rendering processes and similar pro things.

Probably I will choose the Vega 48 in my new iMac i9 and use it for gaming and deep learning modeling.

 

[cynics]

Something to just keep in mind. For productivity task that are heavy in CPU and GPU use (video editing/rendering/transcoding) just having a dedicate GPU is the largest performance increase, doesn't matter which one. Since certain task are only limited by the CPU's RAW performance when you move the GPU task off the CPU package and memory it frees up a lot of performance.

From there it just depends on what you are doing. If you don't have a specific GPU requirement or workload I just get whatever your budget can justify. Benchmarks show and overall representation of the GPU's performance however they are testing everything from video rendering to cryptography so its not a realistic metric to use when applying that data to real world usage.

For example, if you just touch up a photo every now and then its unlikely you would notice the difference between the 20 and 48. If you data mine or something then a dedicate GPU is mandatory as performance is hardly comparable (dedicate GPU vs integrated GPU).

 

[Sam Marks]

I totally agree with you.

My only question for my new iMac (after having a previous new iMac i9 with Vega 48 fatally failing during its first days of usage and was returned), is if I should buy the optional Vega 48 or maintain the Radeon Pro 580X and then buy a TB3 eGPU (Razor X or Sonnet) with something like the Radeon Vega VII for a total cost of €1,000 vs the €540 of the Vega 48. I know that the TB3 connection limits the performance results of the eGPU and that not all the apps support the internal iMac screen with an eGPU. Difficult question.

 

[cynics]

I totally agree with you.

My only question for my new iMac (after having a previous new iMac i9 with Vega 48 fatally failing during its first days of usage and was returned), is if I should buy the optional Vega 48 or maintain the Radeon Pro 580X and then buy a TB3 eGPU (Razor X or Sonnet) with something like the Radeon Vega VII for a total cost of €1,000 vs the €540 of the Vega 48. I know that the TB3 connection limits the performance results of the eGPU and that not all the apps support the internal iMac screen with an eGPU. Difficult question.

I think this comes down to your workload. A long time ago I had to tell myself I just don't need the best as much as I wanted it (and could afford it) just wasn't beneficial.

However if you rely heavily on GPU for what ever it is you do and your software supports it the initial purchase could be more of an investment because you can now upgrade the GPU next year and the year after and then use the eGPU on a new Mac or with other Macs.

Once the price drops a bit more I'm going to get one so I can keep using 13" MBP's (I prefer the smaller size) without needing to get a high end iMac for the stuff I need a dGPU for.

 

[imrazor]

@Sam Marks I just wanted to add a couple of notes to the discussion. Firstly, I also tried to add a Vega 56 to the Mac Pro, and it simply would not boot. I think this is due to the Vega 56 wanting more than 150w of power. I've seen a couple of PSU hacks to address the issue, but didn't feel skilled enough to attempt them. At present, it seems I'm limited to GPUs with no more than 2x6-pin PCIe power connectors.

My use for this machine will probably be simply to stay abreast of OS X developments, and some gaming. I work in IT, and enjoy OS X, and also want to keep my finger on the pulse of the Apple ecosystem. Thus it's hard to justify spending thousands of dollars on a true pro system. Oddly enough, it's Apple Keychain that's dragging me more firmly into the Apple sphere. Too many bloody passwords...

I had always heard that the limited bandwidth available to TB3 eGPUs didn't matter for compute workloads. It seems that Linus' video proves that theory wrong. Thanks for the link.