AMD Ryzen 5 2400G and Ryzen 3 2200G APUs performance unveiled

[Gideon]

- RavenPI 1.1.0.1 (Bristol, Summit and Raven Ridge)
- PinnaclePI 1.0.0.0x (Bristol, Summit, Raven and Pinnacle Ridge)

Thanks for clearing that up, I was really confused between the two. For instance I though the latest Tomahawk B350 BIOS already had hidden support for Pinnacle Ridge (since 1.1.0.1 > 1.0.0.0x). Now it's quite apparent, it does not.

 

[Shivansps]

Then how would AMD have made a more powerful offering for Microsoft if that was the case? Seems to me if Sony did what you are implying then they would have been smart enough to include Microsoft in the terms....See how silly your scenario sounds.

Maybe because the Microsoft deal was already signed by then, or their did not include other semi-custom deal in the terms of the contract.
Sony makes its money out of games on a super closed platform, console price/features is its the only thing they have lure people in, semi-custom deals are not a threat, generally avalible hardware for a widely open and used playform it is.

His scenario is not silly, it's absurd. Next on Dallas we'll find out it was Nvidia who payed Sony to pay AMD to not launch Vega.

Its not absurd to think Sony is doing with AMD the exact same thing they do with the software houses, thats what Sony do

Specially if you have the huge coincidence that when Sony was launching its "4k gaming console" AMD 4K gaming dGPUs went missing for a year.

Or this deal with Intel to do a super high performance APU that compete with their own products, why AMD couldt do that themselves? They did not need Intel for that nor they are in need to accept such a deal.

 

[VirtualLarry]

However, at least in the the few "benchmarks" I've seen, 2200G seems to be a good 7-10% faster in games, this is pretty extraordinary and similar difference than between broadwell - skylake..(larger than 1200-1300x!) .how come nobody has picked up on this?....these results are repeatable across other "reviewers" ...although using slower GPUs differences are less pronounced (bottleknecked).

No more cross-CCX latency penalty? Remember, the 1200 / 1300X CPUs, have two CCXs, each with two cores disabled. The 2200G has a single CCX, with all four cores enabled.

 

[Hitman928]

Based on the CB15 ST score he posted, the clocks are < 4.1GHz.
Raven scores 160 at 3.9GHz.

He was running single channel memory, how much of an effect does that have on CB for Zen?

 

[epsilon84]

Yes it does seem off, interesting.

However, what I've not seen much in the reviews I saw, was raven ridge connected to a dgpu and then tested against summit ridge.
For instance...2200G Vs 1300x (both stock,1080ti)..
https://youtu.be/-uJ6LsWZJ8I
In theory although both are same threads and clock speeds, 1300x should be a little faster due to double L3 cache.
However, at least in the the few "benchmarks" I've seen, 2200G seems to be a good 7-10% faster in games, this is pretty extraordinary and similar difference than between broadwell - skylake..(larger than 1200-1300x!) .how come nobody has picked up on this?....these results are repeatable across other "reviewers" ...although using slower GPUs differences are less pronounced (bottleknecked).

Must be precision boost 2 and those cache latency improvements to Zen+.
What is the effect going to be with proper L3 and 12nm?...pinnacle ridge looks competitive for sure.

Interesting, this is the first that I've seen of the APUs being significantly faster than an equivalently clocked Ryzen 3/5 for gaming. Techpowerup, for example, has the 2200G ~2% ahead of the R3 1200 at 720P gaming, but ~3% behind at 1080P gaming: https://www.techpowerup.com/reviews/AMD/Ryzen_5_2400G_Vega_11/20.html

This actually puts it a bit behind clock for clock as the 2200G is clocked at 1300X speeds, but only matches a 1200 in dGPU gaming.

The 2400G exhibits the same trend compared to the 1500X, 1% ahead at 720P but 4% behind at 1080P.

 

[Abwx]

The 2400G exhibits the same trend compared to the 1500X, 1% ahead at 720P but 4% behind at 1080P.

That s a very good result for the 2400G considering the 1500X massive advantage provided by its 16MB L3, one could have expected a bigger difference.

 

[PeterScott]

Interesting, this is the first that I've seen of the APUs being significantly faster than an equivalently clocked Ryzen 3/5 for gaming. Techpowerup, for example, has the 2200G ~2% ahead of the R3 1200 at 720P gaming, but ~3% behind at 1080P gaming: https://www.techpowerup.com/reviews/AMD/Ryzen_5_2400G_Vega_11/20.html

This actually puts it a bit behind clock for clock as the 2200G is clocked at 1300X speeds, but only matches a 1200 in dGPU gaming.

The 2400G exhibits the same trend compared to the 1500X, 1% ahead at 720P but 4% behind at 1080P.

I trust the benchmarks more from the established web sites, rather than the guys who don't even bother with reviews, and just produce short videos are nothing more than benchmark videos with muzak.

 

[Mopetar]

Interesting, this is the first that I've seen of the APUs being significantly faster than an equivalently clocked Ryzen 3/5 for gaming. Techpowerup, for example, has the 2200G ~2% ahead of the R3 1200 at 720P gaming, but ~3% behind at 1080P gaming: https://www.techpowerup.com/reviews/AMD/Ryzen_5_2400G_Vega_11/20.html

This actually puts it a bit behind clock for clock as the 2200G is clocked at 1300X speeds, but only matches a 1200 in dGPU gaming.

The 2400G exhibits the same trend compared to the 1500X, 1% ahead at 720P but 4% behind at 1080P.

I think it would be safe to assume those differences are within the margin of error and that the two processors are essentially equivalent for the purpose of the tests being conducted. Maybe going past 1080p would start to show more of a gap, but it may not matter at lower resolutions.

 

[epsilon84]

That s a very good result for the 2400G considering the 1500X massive advantage provided by its 16MB L3, one could have expected a bigger difference.

I was responding to french toasts post with the linked Youtube video showing the 2200G being 5 - 10% faster than the 1300X when paired with a 1080 Ti. If anything, I would expect it to be slightly slower than the 1300X due to the smaller L3 cache, not 5 - 10% ahead.

Like PeterScott I would be more inclined to trust the tech site results as being 'accurate' rather than random YouTube videos.

 

[epsilon84]

I think it would be safe to assume those differences are within the margin of error and that the two processors are essentially equivalent for the purpose of the tests being conducted. Maybe going past 1080p would start to show more of a gap, but it may not matter at lower resolutions.

1% I would consider margin of error, but not a 4% difference averaged out over many games. I think its fair to say the APUs perform slightly worse than the equivalently clocked R3/R5 variants at 1080P gaming, though that shouldn't come as a surprise due to the smaller L3 cache.

 

[Markfw]

Meanwhile I'm waiting for Ryzen 3 CPUs and Linux support...

Linux support ? I am running linux mint on 2 Ryzen and 3 threadripper boxes.

 

[Kenmitch]

1% I would consider margin of error, but not a 4% difference averaged out over many games. I think its fair to say the APUs perform slightly worse than the equivalently clocked R3/R5 variants at 1080P gaming, though that shouldn't come as a surprise due to the smaller L3 cache.

....But in the end have more value.

Things to consider when cross shopping these offerings

Current budget vs gaming desires....Example want dGPU 1060 class or better but currently can only afford a 1050. iGPU could get you buy until you save some xtra cash up or prices stabilize.
What if your dGPU goes belly up and you have to rma it? Could take weeks, months.
You want to sell your rig and keep your current dGPU?

 

[Peter Watts]

The APU's are already out of fashion, we want 2600G next month.

 

[epsilon84]

....But in the end have more value.

Things to consider when cross shopping these offerings

Current budget vs gaming desires....Example want dGPU 1060 class or better but currently can only afford a 1050. iGPU could get you buy until you save some xtra cash up or prices stabilize.
What if your dGPU goes belly up and you have to rma it? Could take weeks, months.
You want to sell your rig and keep your current dGPU?

In terms of the 2200G vs R3 1200, I would agree. Essentially you get a free Vega 8 iGPU, so there is essentially no reason to opt for an equivalently priced R3 1200.

With the 2400G it isn't so clear cut, but not due to the 1500X but rather the 1600 which is very close in price, so unless you need the iGPU I would opt for the 1600 for the extra 2 cores / 4 threads/

 

[whm1974]

The APU's are already out of fashion, we want 2600G next month.

With 6c/12t and 8MB of L3 cache.

 

[Glo.]

But in the end its still a contract between Sony and AMD, not sure why you belive the contract cant "ask" something out of AMD just because the semi custom unit is another division.

Also its Sony!! They do exactly this stuff with the software houses, not allowing them to release the game for PC, some time as time exclusive, in others forever! Or do you belive some of the software houses dont want to make money out of the PC market?
I expect the worse out of Sony, and back on that moment AMD could have accepted anything.

I agree this is all speculation on my part, but if this thing with Vega and KBL-G happens again the coincidences will be piling up.

So let me give you a question. Did Sony-AMD contract stopped AMD from selling Console killer GPU: Vega? It is faster than any of Console APUs.

Again, nothing Sony can do to stop AMD from designing their own chips, even those that can kill Sony and Microsofts business.

 

[The Stilt]

He was running single channel memory, how much of an effect does that have on CB for Zen?

Makes no difference for ST.
159pts with 2933MHz single channel, 160pts with 2933MHz dual channel.

 

[alexruiz]

These are still Ryzen CPUs- maybe they would work fine if put into a W7 system that was installed with Summit Ridge?

It doesn't work. My win7 image was created on Summit Ridge (AsRock X370 kiler sli/ac with Ryzen 7 1700) That image has AMD chipset drivers 18.10... still crashes

My guess that the CPU part will work fine, but the iGPU wouldn't.

See quote above

If that is PinnaclePI-AM4 1.0.0.0a, its newer then RavenRidgePI-AM4 1.0.7.2. Version number got reset for 2nd gen Ryzens. So it would actually be a downgrade. Be warned, 1.0.7.x has been known to be buggy on some boards. My Crosshair VI included...

I am aware that the AGESA number resets, that is why and I proposed an older AGESA (even if higher number) so just in case that the older AGESA could still recognize Raven Ridge, but maybe not crash in win 7

Don't get confused by the AGESA version numbers, because they're different for each branch.

- SummitPI 1.0.0.6x (Bristol and Summit Ridge)
- RavenPI 1.1.0.1 (Bristol, Summit and Raven Ridge)
- PinnaclePI 1.0.0.0x (Bristol, Summit, Raven and Pinnacle Ridge)
- ThreadRipperPI 1.0.0.5 (Zeppelin SP3r2 MCM)

PinnaclePI is the most current development branch for AM4 platform.
Should be used for all AM4 parts, as soon as it reaches full maturity.

Thank you, you are the man!
This explains very nicely how they are being branched.
So then, AGESA 1.0.7.x belongs to SummitPI?

 

[The Stilt]

So then, AGESA 1.0.7.x belongs to SummitPI?

1.0.7.x is early RavenPI.

Once all of the ODMs get up to the pace, all AM4 motherboards will be using PinnaclePI.
Branches targeting newer designs are backwards compatible within the same infrastructure.

 

[alexruiz]

On the youtube reviews subject.
True, these are obscure reviewers, but I won't discredit them because of that.

IF they can provide very detailed breakdown of settings, software, patches and hardware used, then the numbers should be reproducible by the rest of us. I DO mean very detailed, not just "2200G at 4.0GHz"
Pictures of UEFI settings, details of the board and BIOS version, drivers version used for GPU and chipset, etc.

The established sites, with all respect due, many of them are just looking for hits. Many of them, as they have to work with several platforms, deploy the test beds from a single master image, that in most of the cases was created on intel. Many of them keep testing AMD with the generic Microsoft AHCI driver, when the AMD SATA driver is faster, but their mind is stuck in 2010 when AMD SATA was buggy...
My rant is that compared to most sites, my own AMD numbers are always higher, in many cases, not a trivial amount. But then, my master image was created on a promontory chipset, with all the latest AMD chipset drivers, SATA forced to AMD driver instead on MS AHCI generic, etc...
I wouldn't be surprised if many of these big sites that always show underwhelming AMD numbers have their AMD test beds fully patched for meltdown... (*cough* PClab.pl *cough*)

So, if the small reviewers show very promising numbers, let's find out why.
We don't know what these obscure guys are doing, maybe their fine tweaking is much better.
How much better? That is when common sense kicks in.
2400g faster than i5-8400? Possible. 50% faster? Can we see the settings for everything please?

 

[PeterScott]

The established sites, with all respect due, many of them are just looking for hits. Many of them, as they have to work with several platforms, deploy the test beds from a single master image, that in most of the cases was created on intel. Many of them keep testing AMD with the generic Microsoft AHCI driver, when the AMD SATA driver is faster, but their mind is stuck in 2010 when AMD SATA was buggy...

I see the opposite. Actual detailed reviews, in depth analysis from the established reviewers, vs the benchmark only youtube channels that do nothing but needlessly record a benchmark in video form for hits. No detail, no analysis, no speaking even. Just quick and dirty, undetailed benchmarks with Muzak. Minimum work to create a video. I see no reason to value this over actual reviews/analysis.

 

[epsilon84]

On the youtube reviews subject.
True, these are obscure reviewers, but I won't discredit them because of that.

IF they can provide very detailed breakdown of settings, software, patches and hardware used, then the numbers should be reproducible by the rest of us. I DO mean very detailed, not just "2200G at 4.0GHz"
Pictures of UEFI settings, details of the board and BIOS version, drivers version used for GPU and chipset, etc.

The established sites, with all respect due, many of them are just looking for hits. Many of them, as they have to work with several platforms, deploy the test beds from a single master image, that in most of the cases was created on intel. Many of them keep testing AMD with the generic Microsoft AHCI driver, when the AMD SATA driver is faster, but their mind is stuck in 2010 when AMD SATA was buggy...
My rant is that compared to most sites, my own AMD numbers are always higher, in many cases, not a trivial amount. But then, my master image was created on a promontory chipset, with all the latest AMD chipset drivers, SATA forced to AMD driver instead on MS AHCI generic, etc...
I wouldn't be surprised if many of these big sites that always show underwhelming AMD numbers have their AMD test beds fully patched for meltdown... (*cough* PClab.pl *cough*)

So, if the small reviewers show very promising numbers, let's find out why.
We don't know what these obscure guys are doing, maybe their fine tweaking is much better.
How much better? That is when common sense kicks in.
2400g faster than i5-8400? Possible. 50% faster? Can we see the settings for everything please?

No amount of tuning is going to allow a 2200G @ 4GHz to be up to 50% faster than an i3 8100 at 1080P medium/high settings on a 1050 Ti, because that is a GPU bound resolution and there should be virtually no difference between two (on paper) comparable CPUs. This is just common sense. If you have a result that is far, far from the expected norm, then those results are going to be scrutinised.

Actually I wouldn't mind if you 'named and shamed' such sites or channels, what numbers of yours do you find to be consistently a lot higher than is reported by review sites and tech channels?

 

[3DVagabond]

Interesting, this is the first that I've seen of the APUs being significantly faster than an equivalently clocked Ryzen 3/5 for gaming. Techpowerup, for example, has the 2200G ~2% ahead of the R3 1200 at 720P gaming, but ~3% behind at 1080P gaming: https://www.techpowerup.com/reviews/AMD/Ryzen_5_2400G_Vega_11/20.html

This actually puts it a bit behind clock for clock as the 2200G is clocked at 1300X speeds, but only matches a 1200 in dGPU gaming.

The 2400G exhibits the same trend compared to the 1500X, 1% ahead at 720P but 4% behind at 1080P.

In general, since TPU rarely reruns their old results and just adds new ones, it's hard to compare precisely with previous results.

 

[3DVagabond]

On the youtube reviews subject.
True, these are obscure reviewers, but I won't discredit them because of that.

IF they can provide very detailed breakdown of settings, software, patches and hardware used, then the numbers should be reproducible by the rest of us. I DO mean very detailed, not just "2200G at 4.0GHz"
Pictures of UEFI settings, details of the board and BIOS version, drivers version used for GPU and chipset, etc.

The established sites, with all respect due, many of them are just looking for hits. Many of them, as they have to work with several platforms, deploy the test beds from a single master image, that in most of the cases was created on intel. Many of them keep testing AMD with the generic Microsoft AHCI driver, when the AMD SATA driver is faster, but their mind is stuck in 2010 when AMD SATA was buggy...
My rant is that compared to most sites, my own AMD numbers are always higher, in many cases, not a trivial amount. But then, my master image was created on a promontory chipset, with all the latest AMD chipset drivers, SATA forced to AMD driver instead on MS AHCI generic, etc...
I wouldn't be surprised if many of these big sites that always show underwhelming AMD numbers have their AMD test beds fully patched for meltdown... (*cough* PClab.pl *cough*)

So, if the small reviewers show very promising numbers, let's find out why.
We don't know what these obscure guys are doing, maybe their fine tweaking is much better.
How much better? That is when common sense kicks in.
2400g faster than i5-8400? Possible. 50% faster? Can we see the settings for everything please?

I haven't really kept up with reviews lately like I used to. But I can recall often thinking, "Man, they seem to get the slowest AMD cards in the world.", quite often reading reviews.

 

[frozentundra123456]

I see the opposite. Actual detailed reviews, in depth analysis from the established reviewers, vs the benchmark only youtube channels that do nothing but needlessly record a benchmark in video form for hits. No detail, no analysis, no speaking even. Just quick and dirty, undetailed benchmarks with Muzak. Minimum work to create a video. I see no reason to value this over actual reviews/analysis.

Yea, and obviously, youtubers arent looking for hits or other compensation either, right?