AMD Ryzen 5 2400G Review: Zen, Meet Vega

[larkspur]

You also made mention of the difference of the CAS14 vs CAS16 for the FlareX vs Ripjaws V... but, in gaming usage, would that have anywhere near as pronounced an effect as what was demonstrated on the various RAM speeds? More? Less? I don't know, but am genuinely curious.

Going back to the techspot memory scaling, you can see that he overclocked the flareX kit to DDR4 3466 but had to relax the timings to CAS16. The results showed basically no difference between DDR4 3466 @CAS16 vs. DDR4 3200 @CAS14. It's not a perfect comparison - it would be better to see the differences in RAM timings @3200 and @2933. But it does give you some idea as to why RAM timings are also important.

 

[rhysiam]

You also made mention of the difference of the CAS14 vs CAS16 for the FlareX vs Ripjaws V... but, in gaming usage, would that have anywhere near as pronounced an effect as what was demonstrated on the various RAM speeds? More? Less? I don't know, but am genuinely curious.

Going back to the techspot memory scaling, you can see that he overclocked the flareX kit to DDR4 3466 but had to relax the timings to CAS16. The results showed basically no difference between DDR4 3466 @CAS16 vs. DDR4 3200 @CAS14. It's not a perfect comparison - it would be better to see the differences in RAM timings @3200 and @2933. But it does give you some idea as to why RAM timings are also important.

See my comment above. Graphics workloads care far, far more about RAM bandwidth than latency, which is why GDDR5/5X/6 RAM exists. We'd have to see a suite of benchmarks to prove it conclusively, but there's good reason to speculate that low latency RAM will have little to no benefit for graphics workloads on these APUs.

 

[larkspur]

Graphics workloads care far, far more about RAM bandwidth than latency, which is why GDDR5/5X/6 RAM exists. We'd have to see a suite of benchmarks to prove it conclusively, but there's good reason to speculate that low latency RAM will have little to no benefit for graphics workloads on these APUs.

Yes I agree with you. But given that these are new very powerful APUs it would be nice to see it tested. The techspot results are interesting and hint at a performance ceiling @3200. I'm curious how much of that was timing related and how much was just the APU being the limiting factor, not the memory bandwidth.

 

[TMTOWTSAC]

There are a number of comparisons between IGP's with Single vs Dual Channel RAM so you can get a sense of how bandwidth starved they really are. In severe cases, going with Dual Channel can actually improve fps by over 100%, when the CPU and GPU are fighting over limited bandwidth.

 

[FD2Raptor]

There's evidence of lower CL resulting in higher memory bandwidth overall:

On Intel IMC we know that it prefers 4 ranks of memory, which is why the Team Dark ROG Dual Channel kit (two dual rank DIMMs) and the Gskill TridentZ Quad-channel kit (four single rangk DIMMs) provides quite a bit more bandwidth than their two single rank DIMM counterparts. A CL advantage of 2 equates to ~1GB/s bandwidth on the TridentZ kit at the same data rate.

But the testing has mostly been done on the Intel IMC and not AMD; what we do know is that the AMD IMC, at least on Ryzen, has more difficulty reaching the same data rate if it has to deal with four ranks of memory (so the Team Dark ROG kit, and Gskill TridentZ kit in the test may have more trouble achieving the same data rate/latency on AM4 platform). What we also know is that the AMD IMC seems to prefers the Samsung B-die memory chips and as I understand the FlareX lineup is using this memory chips exclusively, unlike the Ripjaws/TridentZ lineup, or other lineups from other companies, which could have Micron/Hynix memory chips instead and could prove more difficult to achieve the same data rate/latency.

 

[rhysiam]

There are a number of comparisons between IGP's with Single vs Dual Channel RAM so you can get a sense of how bandwidth starved they really are. In severe cases, going with Dual Channel can actually improve fps by over 100%, when the CPU and GPU are fighting over limited bandwidth.

Absolutely. I don't believe anyone is disputing the benefits of faster (i.e. higher bandwidth) RAM for these APUs. Dual Channel improves effective bandwidth, not latency. The question here is whether low latency matters too. I'm very much of the opinion that for any graphics workloads RAM latency is going to be basically irrelevant, but benchmarks to confirm would be interesting.

 

[Gillerer]

Of course, existing boards need a firmware update to recognize the new models, while newer platforms will include a "Ryzen Desktop 2000 Ready" badge signaling drop-in compatibility.

So I have a question: If people on the forums are looking to buy one of these today (or over the next few days), how do we find cheap boards that we *know* will support these CPUs out-of-the box. [...]

CPUs are for sale and can be purchased right now. But compatible boards? Am I missing something?

Where I live, many online computer part stores offer a BIOS update at no cost, if you purchase both the motherboard and the CPU from them.

When selecting the motherboard, you just need to check the manufacturer's website and verify that they have a BIOS update available that enables Ryzen Desktop 2000 compatibility.

 

[Olle P]

IGPs are going to get really interesting when AMD and Intel decide to put HBM2/3 on-package: ... A ~$280 APU with an IGP approaching the 3GB GTX1060 should be feasible in the near future. ...

While feasible I don't see much of a market for it except in laptops.
If you buy a desktop computer and spend $2500 on the CPU you either want a discrete graphics card or make do with less.

... If I wanted to make a cheap PC with Intel to play in 1080 in low I would ... take G4560 + GTX 1050 vanilla + 8 GB of Ram. Surely I would have more FPS.

... and lousy overall performance for anything else.

I really like the 2200g @ $99. That is a clear value winner and a great entry-level gaming recommendation - Great job AMD! But I don't see enough performance delta with the 2400g to justify the $70 price difference for the entry-level segment.

I also think the 70% price addition is a bit steep for including SMT and having a little better graphics.

The 2200g also makes the 1200 and 1300x irrelevant now.

The 1300X can still be the better option for an all-purpose budget build when paired with a half decent graphics card. Especially if things like many PCIe lanes are valued high.
The Pentiums don't have enough computing power and next step up from Intel will cost way more.

 

[Olle P]

... the GT1030 is the cheapest or second cheapest current-generation GPU money can buy.

The GT 1030 is, as shown, close to the Vega iGPU in performance and about as low as anyone can reasonably go if buying a discrete graphics card for gaming.
For lighter graphics still it's the GT 710 that is the latest generation, available at less than half the price of the GT 1030.

 

[Olle P]

The techspot review actually shows memory scaling on the Ryzen APUs very well and the differences appear to be significant. Not a deal-breaker, but significant nonetheless: https://www.techspot.com/review/1574-amd-ryzen-5-2400g-and-ryzen-3-2200g/page8.html ...

I see something totally different:
* Assuming the use of a cheap 60Hz monitor all framerates >60 fps can be seen as equal, since there's no visible difference.
* The test thus show that for CS:GO at 720p with dual channel memory speed doesn't matter at all! Even at 2133 MHz the 1% low stay well above the monitor's refresh rate.
* In single channel there are occasional glitches with 2133 MHz memory, but none at all with 2666 MHz.

So fast memory is not important for CS:GO in a low budget build!

 

[logainofhades]

IGP has always benefited from faster memory, especially when talking AMD APU's. Infinity fabric also benefits from faster ram. That has been the case since Ryzen released. The price difference is minor enough, that going with any ram below 3000 is kinda silly to do. Even going to 3200 isn't a massive leap in price.

PCPartPicker part list / Price breakdown by merchant

Memory: G.Skill - Ripjaws 4 Series 8GB (2 x 4GB) DDR4-3200 Memory ($108.99 @ Newegg)
Memory: Corsair - Vengeance LPX 8GB (2 x 4GB) DDR4-3000 Memory ($99.99 @ Amazon)
Memory: G.Skill - Aegis 8GB (2 x 4GB) DDR4-2133 Memory ($87.99 @ Newegg)
Total: $296.97
Prices include shipping, taxes, and discounts when available
Generated by PCPartPicker 2018-02-20 09:35 EST-0500

 

[TMTOWTSAC]

There are a number of comparisons between IGP's with Single vs Dual Channel RAM so you can get a sense of how bandwidth starved they really are. In severe cases, going with Dual Channel can actually improve fps by over 100%, when the CPU and GPU are fighting over limited bandwidth.

Absolutely. I don't believe anyone is disputing the benefits of faster (i.e. higher bandwidth) RAM for these APUs. Dual Channel improves effective bandwidth, not latency. The question here is whether low latency matters too. I'm very much of the opinion that for any graphics workloads RAM latency is going to be basically irrelevant, but benchmarks to confirm would be interesting.

We're in luck, Gamersnexus just did a comparison with different kits:

https://www.gamersnexus.net/guides/3244-amd-r5-2400g-memory-kit-benchmarks-and-single-vs-dual-channel

They include results from a CAS 14 and 16 kit both running at 3200MHz as well as single vs dual channel, and even a CS:GO at 1080p. Something for everyone.

 

[rhysiam]

We're in luck, Gamersnexus just did a comparison with different kits:

https://www.gamersnexus.net/guides/3244-amd-r5-2400g-memory-kit-benchmarks-and-single-vs-dual-channel

They include results from a CAS 14 and 16 kit both running at 3200MHz as well as single vs dual channel, and even a CS:GO at 1080p. Something for everyone.

Good find, thanks for the link!

That seems to raise more questions than it answers with regard to the above discussion about memory timings though. The 2933mhz CL16 kit consistently matches or even slightly outperforms the GEIL 3200mhz CL16 kit. If you just tunnel vision on the two 3200mhz kits you could conclude that it's the lower latency on the CL14 kit that's producing the extra performance, but that doesn't hold up when you compare the two CL16 kits. The large variation in scores between motherboards is interesting too, particularly with the GEIL kit. It's a shame they don't unpack that a little further in the article. I suppose this is all symptomatic of AMD's lower market share and the industry still getting used to Zen's memory controller. It does make it harder to provide reliable recommendations though when you see 5-10% or even higher variation based on motherboards, or RAM with otherwise similar specs.

 

[larkspur]

It does make it harder to provide reliable recommendations though when you see 5-10% or even higher variation based on motherboards, or RAM with otherwise similar specs.

Yeah, I agree - it's not really conclusive to our previous discussion. A lot of odd variation but the magnitude of variation is pretty low. Hard to know for sure which kits to recommend for Raven Ridge. Looks like we'll just have to stick to the QVL of the motherboard being used with an emphasis on higher-frequency, popular RAM kits. Maybe Tom's could explore this further in an article (Crashman you out there? ) Spending extra on the FlareX kits might save a headache or two but as long as they command such a high price premium its hard to recommend them for these tight budget builds.

 

[KaiserPhantasma]

Finally some solid 720p gaming for "lightweight" laptops not... gone will be the days where you need a purposefully "built" laptop for gaming... looking forward when 1080p becomes pretty much normal on almost all games on the integrated chips in the future =D