AMD Lists Ryzen Raven Ridge Desktop APU Details

  • Thread starter Thread starter Guest
  • Start date Start date
Toggle sidebar Toggle sidebar
Status
Not open for further replies.


That would be nice. My brother needs an upgrade for his overclocked coal burning L5420/HD6770 ASAP.
Will wait for benches with crossed fingers.
 

The 2400G would be $200+ if it had any amount of HBM on it. Also, the 2700U on the mobile side of things is already burying all past IGPs using system DDR4.

I'm not convinced it would be physically possible or practical to cram an HBM chip within the AM4 footprint as that would put the CPU die grossly off-center. If AMD wants to do multi-chip packages for mainstream, it'll likely need a more rectangular socket footprint, which means CPUs with HBM as L4 cache or IGP memory probably won't happen until AM5.
 
^ agreed - in all fairness the i3 8100 is fine & dandy for htpc use too & a good pricepoint - but the cheapest matx board at oresent is $120 or so.

You can drop that ryzen on a $40 a320 board & you're good to go for a lot less money.
 


Probably, yeah. Sure, it's a cut-down Vega, but it seems to have slightly better computing power than the 6770, plus it's compatible with newer versions of OpenCL (v2.2. instead of v1.2), OpenGL (v4.5 instead v4.4), Shader Model (v6.0 instead of 5.0), DirectX (Direct3D v12.0 instead of v11.0), plus support for Vulkan & SPIR-V.





I highly doubt that. The prior desktop APUs, even the ones using the Socket AM4 platform, were 2C/4T Excavator-based CPUs (i.e. slightly improved Steamroller chips akin to the old FX line) using R7 onboard graphics (i.e. R7 250/350). So you're getting a much better base CPU (the 2400G being roughly comparable to the R5 1500X/1500 Pro, the 2200G essentially is the R3 1300X/1300 Pro), with access to faster RAM, & a better graphics package. Barring something being severely gimped on the driver side or chipset, you're talking about having an APU that on the graphics side is probably at least comparable to the GT 1030 (& could potentially maybe even match up to the GTX 1050/1050TI or RX 560), with a very decent Ryzen CPU.
 
I want to see some benchmarks comparing these APU's to i3's and i5's that are paired with GTX 750's and 1050"s. Also would be nice to see these APU's paired with GTX 750's and 1050's during the benchmarks.
 
^ Im sure you'll see those comparitive benches on release.

There really isnt a need though imo , they're to all intents a ryxen 1300x & a 1500x with an igp

We pretty much know the results already.

They're not going to match Intel paired with a half decent gpu at all using just the igp - at the very best with very fast ram thats rx550 performance youre looking at.

 
Agreed...but from a budget ITX build standpoint, they would provide an ideal iGPU-based build.
-- i3-8100 with ASRock Z370M ITX board: $255 USD (https://pcpartpicker.com/list/yDD2vV)
-- i5-8400 with same motherboard: $327 USD (https://pcpartpicker.com/list/sQTBXH)
-- ASRock AB350 ITX board: $80 USD (https://pcpartpicker.com/list/mzgLTH), so would cost $179 USD with the R3 2200G or $249 with the R5 2400G

Until Intel can release the H & Q chipset boards for their Coffee Lake line, you're looking at a Ryzen APU-based system costing about $75 less than the comparable Intel Coffee Lake ITX build, but with the potential for overclocking still in there. From a gaming perspective, any edge the Intel system might have on the CPU side I'm betting will be offset by Vega's performance on the graphics side. And given the current spread between H270 & Z270 boards is only maybe $40 USD, I don't know if the non-Z sets are going to help offset that.
 

Aren't the CPU cores on these new APUs all on a single CCX? All previous desktop Ryzen CPUs were harvested 8 Cores, with the 1300x and 1500x both utilising two cores from each CCX and relying on the Infinity Fabric for communication. These APUs also have precision boost 2 and AMD have confirmed they have the minor tweaks that will appear later in the year in Ryzen + (sometimes called Ryzen 2). In all likelihood we're looking at very modest gains here, but the single CCX design will be interesting as Infinity Fabric latency is blamed for some of Ryzen's lower IPC, particularly under certain workloads. I'm not overly excited about performance leaps here, but they're not identical and if nothing else, it should give us some insight into just how significant a factor the Infinity Fabric latency is for Zen' s performance.
 
I'd love to pick one up for my kids small gaming rig, but ram still makes it cost prohibitive. I will have to wait...
 


Uh... comparable to 1030, sure, but matching a GTX 1050? When even on AMD own slide showing that you'd need 2400G + OCed integrated graphics AND 3600Mhz DDR4 to achieve ~4k Firestrike points?

From the Firestrike score in the slide, the best case scenario seems to suggest the 11CU on the 2400G will put it just behind a 1030 unless you spend a good deal of money on 3000+ DDR4, and somewhat behind a 750 Ti which generally seems to do ~4800 points.



The actual difference is not that great when you take into account that the integrated graphics solution on those system will eat into the DDR4 of the system memory making the 8GB-DDR4-only or 8GB-DDR4-for-now option a bit more difficult to work with on some titles.
 
2 FD2RAPTOR: check the Wolfenstein2 benchmark here- it has anecdotal evidence of 2500U (basically an underpowered 2200G) being faster than the GTX1050: https://www.computerbase.de/2018-01/amd-ryzen-5-2500u-test-raven-ridge/2/
 
I think those comparing the 2400G to a 1500X didn't notice that these things appear to have only 4 MB of L3 cache, which ends up being 4x less than a 1500X (1/2 that of a 1400 and either Ryzen 3)

No idea what kind of hit they'll take from it, I doubt it'll hurt too much though.

Still will be absolutely killer for HTPC builds. Basically a no brainier there.
 


The article shows 6MB of L3 cache.

What i'm really worried is HOW turboboost (or wathever they call it) will handle continuous loads. Remember that old APU's had 95-100w of TDP. Yes, they had 32nm/28nm fab tech, but when the IGPU was taxed, turboboost would limit the CPU. (some 4GHZ+ models locked at 3.0GHZ).

When the system is cool things do just fine. Burst loads can be handled. But how about CONTINUOUS LOADS?

Don't take me wrong. I'm asking that because i Had a Llano and a trinity system and cooling was a MAJOR consideration for sustained performance and even then, there were hard limits that could'nt be pushed because of TDP limits. Some lesser models performed really similar to higher end models in gaming because of that.
 
FYI, I recently confirmed on reasonable authority, that RR has 24 pcie3 lanes usable including chipset. (I still dont get why ryzen isnt 32 lanes, in keeping with 64 & 128 on TR & Epyc?)

I had heard that each core yields 4 lanes, so since it's a single 4 core ccx, there was a danger it would be limited to 16 lanes.
 


Wolfenstein 2 is a heavily AMD-leaning title though:


And even then a below 30 fps result at 1600x900 lowest quality settings AND 50% rendering resolution i.e. actual resolution of 800x450 for an FPS isn't what one would consider a good experience, I believe.

These APU will be an option worth considering for those with a tight budget to be sure, but saying that this $100-170 APU package will be equal to a dedicated ~$100 GPU is hyping it a bit much.
 


Lol yeah... no..... 704 Vega cores connected to 4x sufficiently power x86 cores for once, and over Infinity Fabric, & using DDR4 (aka ≈40-60GB/s memory bandwidth) will take the 512 GCN3 (Tonga/Fiiji) cores connected w/ 2x molasses-esque Steamroller modules via old school Hyper-Transport, & using DDR3 (≈20-30GB/s) that Carrizo packed (last widely available APU; AMD technically launched Bristol Ridge w/ Excavator & DDR4 to succeed it, but they were barely ever for sale) to the absolute freaking cleaners, no questions asked (and Carrizo was already stomping all non Iris Pro Intel parts).

Sure, dedicated VRAM would help dramatically; iGPU's are chronically bandwidth starved, but not enough that the doubling from 25ish to 50ish GB/s from the DDR3 to 4 jump won't be enough to facilitate major gains with RR, and adding some HBM2 would greatly increase the price, size, socket, etc...
 
'Onion and Garlic' fusion has come quite a-ways ... and leaps ahead of the ol' 785G with 'Side Port' memory (not to mention 'GMA' 810/15s or even the Geforce 6150).

I think some folks are whiffing on the APU graphics side if things. My understanding is those Radeon cores first and foremost are 'SIMD Engines' -- gaming on them is simply gravy on the biscuit. Until we see cost-effective HBM, fancy IGPs will remain outside the mainstream, and cannot really compete with a basic CPU pushing low-cost discreet 'entry-level' graphics.

The GTX 1030 though, best be sweating bullets.

I also suspect with the Ryzen APU there is mo' low-hanging gravy to be had in optimizing the dual channel DDR4 controllers -- not to mention whatever voodoo Dr Su and friends can pull off with 'memory unification' and whatever evolves from HSA; i.e., L2 sniffing, 'Radeon Control Links' etc.
 
Status
Not open for further replies.

TRENDING THREADS

Latest posts

Moderators online

Share this page

COMPANY
RESOURCES
FOLLOW
Top Bottom