Review AMD Ryzen 5 3400G Review: First-Gen Zen Gets Refreshed

[Admin]

AMD's Ryzen 5 3400G comes with four cores and eight threads paired with Radeon Vega graphics for a mere $150.

AMD Ryzen 5 3400G Review: First-Gen Zen Gets Refreshed : Read more

 

[yeti_yeti]

i3-9350K on top in pretty much all of the gaming tests with a discrete gpu? From what I have seen, it is considered a terrible purchase at its MSRP. Didn't expect to see it perform this well in this review.

 

[alextheblue]

i3-9350K on top in pretty much all of the gaming tests with a discrete gpu? From what I have seen, it is considered a terrible purchase at its MSRP. Didn't expect to see it perform this well in this review.

Yeah these are only a decent buy if you're looking for something with a strong iGPU. Even then, these are stopgaps until the Zen 2 powered APUs hit - although that might be a while, since they're wisely targeting mobile first. They've also been around for a while now, so it's not like a hot new release.

 

[Darkbreeze]

Yeah these are only a decent buy if you're looking for something with a strong iGPU. Even then, these are stopgaps until the Zen 2 powered APUs hit - although that might be a while, since they're wisely targeting mobile first. They've also been around for a while now, so it's not like a hot new release.

Which is exactly what makes it surprising that we're seeing a review of it at this late date. But I guess ANY review of actual hardware is a good thing. Better than another "this is on sale" or "check out these deals" piece of clutter.

 

[domih]

I happen to have i5 9600K, R5 3400g, R3 3200g as generic Linux servers.

Using InfiniBand FDR (56Gbps) translating into theoretical max ~48Gbe for IPoIB (Internet over InfiniBand) I get the following the iperf3 results:

• 44.5 Gbits/sec for the i5 9600K ($243 on NewEgg) on ASRock Z390M-ITX-ac, 32GB mem @ 3200.
• 46.1 Gbits/sec for the R3 3200g ($95 on NewEgg) on ASRock B450M Pro4, 32GB mem @ 3200.
• 46.5 Gbits/sec for the R5 3400g ($150 on NewEgg) on ASRock B450M Pro4, 32GB mem @ 3200.

Kind of statistically similar with a 4% advantage to the R5 3400g over the i5 9600K.

So while the i5 9600K has 6C/6T, the R5 3400g with 4C/8T can rival the i5 9600K in specific tasks.

Network cards: Mellanox ConnectX-3 VPI MCX354A-FCBT (about $50 on eBay)
Switch: Mellanox InfiniBand SX6790 (about $200 on eBay)
Cables: Mellanox QSFP+ DAC or AOC (for about $25 to $90 unit depending on type or length on eBay)

CONCLUSION: If you want a fast 40Gbe NAS (SMB IPoIB, NFS IPoIB or RDMA) you must go with recent hardware BUT you do not need to spend thousands of $$$.

OTHER CONCLUSION: I can hardly wait for the R5 4400g.

 

[linuxdude]

I don't get it - the 3400G has been released to market over a year ago - why review it now?

 

[Nick_C]

I don't get it - the 3400G has been released to market over a year ago - why review it now?

Planting a "refreshed AMD APUs aren't that much better than the older version" message in the readers' consciousness might me seen as an attempt to detract a little from the upcoming release of the 4000 series APUs - which are expected to be rather good....

 

[piquard4567]

is it true that due to the quirks of on chip memory bandwith, an overclocked R3 3200G pretty much matches an overclocked R5 3400G in integrated gaming performance? is that why AMD WON'T LET BIG REVIEWERS PIT OVERCLOCKED R3 APUs AGAINST R5 APUs?

please, for us budget gamers out here, will you PLEASE TEST AN OVERCLOCKED RYZEN 3 APU!!!!

 

[piquard4567]

WHAT'S GOING ON WITH TOM'S HARDWARE'S REVIEWS?

The whole freaking point of these APUs in this class of processors is all about gaming on the built in graphics, without having to buy an additional graphics card. So, what does Tom's do?

This is what they do: they only give us just a few games on the iGPU. And then proceed to bore us to tears with a ton of predictable ray tracing and compression and encoding benchmarks. What, a 6c/12t will smoke a 4c/4t in Handbrake?!?! COLOR ME SHOCKED

at LEAST run the same tests on the iGPU as you did on discrete cards.

 

[piquard4567]

please fix your POV-RAY Single Core charts already!

you are using different versions of POV-RAY without noting it, which is a no-no in benchmarking

 

[bit_user]

Which is exactly what makes it surprising that we're seeing a review of it at this late date.

Exactly. Seeing this review, with no apparent acknowledgement of its lateness, sent my head spinning.

As I thought, it seems the Ryzen 5 3400 G has been available since July, 2019. Retail boxed, no less.

https://pcpartpicker.com/product/XP...core-processor-yd3400c5fhbox?history_days=365

 

[piquard4567]

i3-9350K on top in pretty much all of the gaming tests with a discrete gpu?

where you been man? the unlocked i3 is a friggin BEAST for gaming!

 

[bit_user]

I happen to have i5 9600K, R5 3400g, R3 3200g as generic Linux servers.

This is a bad idea, since none of these CPUs supports ECC memory. Running any sort of file or database server without ECC memory is a recipe for data corruption.

For that, you either need Ryzen Pro APUs, which are not sold via retail or even in individual OEM packaging (as far as I've been able to find) or an Intel E-series Xeon or i3. My choice is i3, with MSRPs ranging from $122 to $184 (depending on clock speed), and are now available with 4 cores.

https://ark.intel.com/content/www/u.../9th-generation-intel-core-i3-processors.html

Otherwise, you can use a non-APU Ryzen, but then you need a dGPU or a motherboard with a separate graphics chip (which some have, in the form of a BMC).

Unfortunately, a bigger cost item than the CPU is likely to be the motherboard. Most mobos that support ECC memory tend to charge a premium. In the case of my i3 system, I paid more for the ASRock Rack mini-ITX board than the CPU that I plugged into it.

 

[piquard4567]

an attempt to detract a little from the upcoming release of the 4000 series APUs

why would tom's detract from the 4000 APUs? that's quite the accusation

 

[bit_user]

why would tom's detract from the 4000 APUs? that's quite the accusation

Yeah, I assumed this review just got put on the "back-burner", at the time, and it basically took them until now to get around to finishing it off.

I've also read some authors complaining that they submitted completed pieces that have sat in the queue for weeks. Although, I'd be surprised if anything that was really finished would get held up for 6 months, barring major staffing issues.

Here's the most recent Ryzen 5 3400G coverage I can find, on the site:

• https://www.tomshardware.com/news/amd-ryzen-3-3200g-ryzen-5-3400g-specs-pricing,39619.html
• https://www.tomshardware.com/news/amd-ryzen-3000-3400g-apu-picasso,39555.html
• https://www.tomshardware.com/news/ryzen-5-3400g-ryzen-3-3200g-b550-chipset,39366.html

So, it seems they really never did post a proper review, until now.

 

[piquard4567]

So, it seems they really never did post a proper review, until now.

nor have they done a review of the budget Ryzen 3 3200G either, which would show that when overclocked, it performs as well as an overclocked Ryzen 5 3400G.

that might sound counter-intuitive, because it is. but the truth will expose the issue with all these APUs; that there is alot more going on when you don't have a dedicated card with it's own highspeed memory. there will be all sorts of bottlenecks that will let the ryzen 3 catch up to the ryzen 5.

 

[bit_user]

that might sound counter-intuitive, because it is. but the truth will expose the issue with all these APUs; that there is alot more going on when you don't have a dedicated card with it's own highspeed memory. there will be all sorts of bottlenecks that will let the ryzen 3 catch up to the ryzen 5.

I'm sure it will depend on the game - whether it's more compute or memory bottlenecked. But, it's true that the APUs are a lot more memory-bottlenecked than dGPUs. So, anything to help that will tend to have an out-sized effect on APU performance.

I agree that benchmarks would be nice. Data is king.

 

[PaulAlcorn]

Exactly. Seeing this review, with no apparent acknowledgement of its lateness, sent my head spinning.

As I thought, it seems the Ryzen 5 3400 G has been available since July, 2019. Retail boxed, no less.

https://pcpartpicker.com/product/XP...core-processor-yd3400c5fhbox?history_days=365

Yeah, it is a bit after launch, but you'll notice a lot of sites did not test these, and this was obviously a lower priority, as we have published reviews of all the other 3000-series models and Intel chips in the interim (a dozen or so). The samples for these came amidst the rush of the 3000-series chips, which we obviously prioritized. A string of other NDAs, coupled with the normal travel and other things (five articles on AMD boost investigation that were definitely much more important at the time), led to this one being late. Which would you rather have, the 3400G review, or the exclusive article (and work that went into discovering and proving it) that exposed that AMD had changed binning tactics (without telling anyone) and that, by design, not all cores are capable of reaching the maximum rated boost frequencies? I know which one I, and all the people who couldn't figure out why they weren't hitting clocks, would rather have. Or how about choosing between the world's first look inside Intel's overclocking lab or the 3400G review? That's why we prioritize.

However, we won't see new versions of these APUs for quite a while (AMD releases on yearly cadence), so this review is still relevant. We're also finishing rounding out the full testing of all models from top to bottom because we switched to a new test image and OS version/video drivers, much of which was necessitated by an unrelenting string of security patches/etc, and a whole string of new BIOS revisions that corrected AMD's issues with boost clocks. It's noteworthy that early BIOS revisions also had serious issues with the 3400G, which didn't help with timeliness.

If only you knew how many times we've had to retest entire test pools over the last six months to ensure that we have the most reliable and accurate results possible. This means we didnt have usable historical data for the downstream chips. Because of security patches and BIOS fixes, we have to test top to bottom to refresh the hierarchy. This is the final price bracket for retesting, so we might as well share the testing.

 

[PaulAlcorn]

is it true that due to the quirks of on chip memory bandwith, an overclocked R3 3200G pretty much matches an overclocked R5 3400G in integrated gaming performance? is that why AMD WON'T LET BIG REVIEWERS PIT OVERCLOCKED R3 APUs AGAINST R5 APUs?

please, for us budget gamers out here, will you PLEASE TEST AN OVERCLOCKED RYZEN 3 APU!!!!

In my experience, this is inaccurate. 3200G doesn't overclock as high, at least our sample doesn't. We can only hit 4.0 on the CPU, 3200 on the RAM, and 1550 MHz on the graphics. Still having stability issues with the chip at these much lower settings (well, lower than 3400G). Not to mention it has 8 CU as opposed to 11. Serious performance delta there. I'm still trying to nail down some of the issues we're having with our sample, which could be either BIOS or silicon. The 3200G also comes with thermal paste and not solder TIM, which might be part of the reason for much lower OC capability. The reason you haven't seen reviews of that one is simply time: These chips were released during a mudslide of other consecutive releases, and many sites didnt even bother hitting them at all.

 

[PaulAlcorn]

ould be nice

Planting a "refreshed AMD APUs aren't that much better than the older version" message in the readers' consciousness might me seen as an attempt to detract a little from the upcoming release of the 4000 series APUs - which are expected to be rather good....

That's silly. We're just calling it how it is. Ryzen 3000-series chips dominate our recommendations in almost all categories and have received multiple awards. Remember, the first round of APUs came during the GPU shortage due to mining, so they were clear-cut winners then. The market is different now. This is a good chip, just not an uncontested leader in its price class due to the reasons cited in the article. Also, pricing on AMD's previous-gen chips.

 

[bit_user]

First, thanks for the review.

Second, thanks for the reply.

Yeah, it is a bit after launch, but you'll notice ...

All good reasons. My only real complaint was that the article didn't acknowledge that the chips launched ~6 months ago. So, I spent a couple minutes scanning the article, trying to see if this was a different 3400G than we had before (since the article talked about it being a refresh, which I get was a reference to the 2400G, but I wasn't initially sure) or if some update got added that popped the article back onto the front page.

So, all I would ask is that there be something in the first couple paragraphs to alleviate that confusion among those of who already knew about the product, and inform other readers that it's not actually a newly-launched product, the way the headline & article sort of makes it sound (or, at least the parts I read).

If only you knew how many times we've had to retest entire test pools over the last six months ...

Paul, I'd never accuse you of being lazy!

P.S. Enjoy the game! Go Chiefs!

 

[Nick_C]

That's silly. We're just calling it how it is.

I think you'd have to agree that the timing is a bit odd though, given the release date of the APU under review.