AMD Raven Ridge 'Zen APU' Thread

[VirtualLarry]

Interesting that the i3 4C/4T (unlocked) is close to the same price (in USD) as the i5-8400 6C/6T (locked) 65W TDP CPU.

So, it looks like either 4C unlocked, or 6C locked, for $180-190. I think I'd pick the locked 6C, but someone who's gaming and needs faster single-threaded speed might go for the 4C unlocked. Plus, though, you'd need a Z370 board.

 

[mohit9206]

I think the leaked sample was at 35W TDP. Ryzen is wonderfully efficient at lower clockspeeds/voltages.. the primary wild card for now is how those Vega cores will perform...

I think once we know how RX Vega shakes out we will be able to much more accurately predict where RR will land performance and efficiency-wise.

Now that we know how Vega shakes out, i would like to know your predictions for Raven Ridge.

 

[swilli89]

Now that we know how Vega shakes out, i would like to know your predictions for Raven Ridge.

I'm keeping the article about the RX Vega 56's undervolting in mind here. Raven Ridge will use Vera's NCU's but will be different from RX Vega as we know it in a few ways.

-will not be using an interposer
-will not be using HBM2
-won't be pressured to clock nearly as high as there is literally no competition to a high end APU from Intel.

Now, given the above considerations we can reasonably surmise that AMD can bin these APU's with Vega cores a bit higher both because cost to manufacture will be lower thus Margins are higher to begin with, as well as the fact that these will be mobile focused, the total opposite of RX Vega in both ways.

With just 1 cluster for 4 Zen cores and 704 Vega cores, binned and down clocked, I think these chips are going to be mobile gaming monsters at 35-45 watts. There won't be anything like them and they should be in an entirely different league than Iris Pro chips.

 

[PeterScott]

With just 1 cluster for 4 Zen cores and 704 Vega cores, binned and down clocked, I think these chips are going to be mobile gaming monsters at 35-45 watts. There won't be anything like them and they should be in an entirely different league than Iris Pro chips.

All the stuff about Vega undervolting and pressure are largely pointless.

Raven Ridge won't be a mobile gaming monster, unless you define mobile to only included IGP/APU devices.

Any kind of modern dGPU will beat it, even low end ones.

 

[jpiniero]

It'll be a lot faster than Iris Pro for sure but something like the MX150 will be loads faster.

 

[krumme]

It'll be a lot faster than Iris Pro for sure but something like the MX150 will be loads faster.

Can we get some guesses for numbers? Aren't we memory bandwith limited?

I think at 45w tdp a bit like 15% above top iris pro for gpu perf with a dual 2667 setup and perhaps 10% for cpu. Plus better drivers.
Iris pro is a 500 usd solution. 14nm and ram bandwith is pretty stiff limitations. Iris cache is there for a reason.

Clearly a mx 150 with a tdp of 25w and gddr 5 is like what 100% faster? Lol have not followed it but nv mobile offering is darn strong and uncontested. But a dgpu is not a similar aolution and neither for cost.

I think this apu will be absolutely stellar 4c mobile cpu. Looking forward to having it.

 

[PeterScott]

I think this apu will be absolutely stellar 4c mobile cpu. Looking forward to having it.

I think it will also be a nice AIO/SFF/HTPC CPU, and I might be interested in it in that application.

It should be a great product for AMD and can't come soon enough. It's too bad it looks like they will miss back to school this year, when they sell a ton of laptops.

I just think some people get carried away with performance expectations, calling it a gaming monsters, earlier I remember some thinking it would perform like a GTX 1050.

While it is hard to predict based on almost nothing, we can be sure that it won't be anywhere near a GTX 1050.

 

[swilli89]

It'll be a lot faster than Iris Pro for sure but something like the MX150 will be loads faster.

Well given the MX150 has 384 Pascal cores and Raven Ridge is rumored to top out at 704 Vega cores I am not so sure about that. Of course you can't compare the two but you can guesstimate. The MX150 has 38GB/sec and the RR APU will have access to a shared 42.6GB/sec. It remains to be seen how they both perform against each other in the real world. However this is really apples and oranges. This is an APU that will be on a single package with a single cooling solution, with lower power consumption that a dGPU.

All the stuff about Vega undervolting and pressure are largely pointless.

Raven Ridge won't be a mobile gaming monster, unless you define mobile to only included IGP/APU devices.

Any kind of modern dGPU will beat it, even low end ones.

Actually if it can handle most games with decent settings at 1080p, then yeah I personally think that is a mobile gaming monster. I'd prefer to not play semantics and debate the interpretation of "monster" so I will just say that it will be a truly new level of performance from an APU and all the thin devices that will be created around it. And yes when I say "mobile" I mean systems that can game on mobile battery, not just be portable.

 

[swilli89]

I think it will also be a nice AIO/SFF/HTPC CPU, and I might be interested in it in that application.

It should be a great product for AMD and can't come soon enough. It's too bad it looks like they will miss back to school this year, when they sell a ton of laptops.

I just think some people get carried away with performance expectations, calling it a gaming monsters, earlier I remember some thinking it would perform like a GTX 1050.

While it is hard to predict based on almost nothing, we can be sure that it won't be anywhere near a GTX 1050.

Well that is ridiculous to even consider. 1050 can draw +60watts which is more than Raven Ridge will be rated for both CPU and GPU all together by a large margin!

A system with an equivalent Intel quadcore and 1050 would last maybe 1 hour on a battery where the RR system, given the same battery, could feasibly last over twice or three times as long. If you're on an airplane, bus, or anywhere away from home and you are playing some games the #1 priority is battery life.

 

[Arachnotronic]

Well that is ridiculous to even consider. 1050 can draw +60watts which is more than Raven Ridge will be rated for both CPU and GPU all together by a large margin!

A system with an equivalent Intel quadcore and 1050 would last maybe 1 hour on a battery where the RR system, given the same battery, could feasibly last over twice or three times as long. If you're on an airplane, bus, or anywhere away from home and you are playing some games the #1 priority is battery life.

When I'm on an airplane or a bus, I usually use my iPhone. Laptop is not the ideal form factor for those uses

 

[NewFatMike]

Hopefully we'll be getting some Freesync/2 monitors on these mobile products as well, which will dramatically improve the experience. 1080p30 on medium with Freesync on a laptop would be perfect for most folks, I think.

Sent from my Pixel using Tapatalk

 

[raghu78]

It'll be a lot faster than Iris Pro for sure but something like the MX150 will be loads faster.

Probably faster. But I doubt MX 150 is going to be 2x the perf of RR APU. Because thats what you need to have a significantly better gaming experience than on a RR APU. I think we will see that RR APU provides roughly the same gaming experience as MX 150. Raven Ridge APU brings 50% higher CPU and 40% higher GPU performance at 50% of the power of Bristol Ridge.

https://www.pcper.com/news/Processo...bile-APUs-Zen-CPU-Cores-and-Die-Vega-Graphics

http://www.anandtech.com/show/10705...e-and-am4-analysis-a12-9800-b350-a320-chipset
https://us.hardware.info/reviews/72...am4-review-benchmarks-igpu-3dmark-fire-strike

A12-9800 Bristol Ridge (65w) scores roughly 2000 in 3DMark Firestrike. We are looking at 2800 points with RR notebook APU at 35w. Geforce MX 150 scores 3200 in 3DMark Firestrike.

https://www.notebookcheck.net/NVIDI...pecs-of-the-GT-1030-for-Laptops.223530.0.html

The way I see it for the vast majority of consumers who want to use a notebook for productivity and light gaming a RR APU will be the best and most power efficient option. If someone is serious about gaming on a notebook they need to move up the stack to a 128 bit GDDR5 based GTX 1050 which scores 5500 on 3DMark Firestrike. But then GTX 1050 alone draws more than 40w while RR APU will draw 35w for CPU+GPU.

https://www.notebookcheck.net/NVIDIA-GeForce-GTX-1050-Notebook.178614.0.html

64 bit GDDR5 cards are just too badly crippled to make a difference against a RR APU.

 

[ao_ika_red]

I think it will also be a nice AIO/SFF/HTPC CPU, and I might be interested in it in that application.

It should be a great product for AMD and can't come soon enough. It's too bad it looks like they will miss back to school this year, when they sell a ton of laptops.

I just think some people get carried away with performance expectations, calling it a gaming monsters, earlier I remember some thinking it would perform like a GTX 1050.

While it is hard to predict based on almost nothing, we can be sure that it won't be anywhere near a GTX 1050.

This, couldn't agree more. RR absolutely will be more power efficient than BR while having more performance, and that's the reason it will land on my living room next year as HTPC. And also, because Vega supposed to have better memory compression than Fury, it will run nicely on limited bandwidth space (dual channel DDR4 for RR and single channel DDR4 for Banded Kestrel). I will keep tracking on 2C/4T-3CU laptop because it should be the new king of ultra low cost notebook and I will say goodbye to my old Bobcat.

 

[mohit9206]

I'm keeping the article about the RX Vega 56's undervolting in mind here. Raven Ridge will use Vera's NCU's but will be different from RX Vega as we know it in a few ways.

-will not be using an interposer
-will not be using HBM2
-won't be pressured to clock nearly as high as there is literally no competition to a high end APU from Intel.

Now, given the above considerations we can reasonably surmise that AMD can bin these APU's with Vega cores a bit higher both because cost to manufacture will be lower thus Margins are higher to begin with, as well as the fact that these will be mobile focused, the total opposite of RX Vega in both ways.

With just 1 cluster for 4 Zen cores and 704 Vega cores, binned and down clocked, I think these chips are going to be mobile gaming monsters at 35-45 watts. There won't be anything like them and they should be in an entirely different league than Iris Pro chips.

What about the desktop Raven Ridge? The top model which i guess would be around $180-190 better beat the 1030 for that price.
While hopefully the mid range model should compare to the current flagship A12-9800 for around $130.
I currently use a gddr5 730 so if I'm looking to upgrade in future i could see my option being either i3-8100+gt 1030 or the top model of Raven Ridge. I'm really looking forward to this comparison.

All the stuff about Vega undervolting and pressure are largely pointless.

Raven Ridge won't be a mobile gaming monster, unless you define mobile to only included IGP/APU devices.

Any kind of modern dGPU will beat it, even low end ones.

It'll be a lot faster than Iris Pro for sure but something like the MX150 will be loads faster.

Since MX150 is a desktop 1030, i find it doubtful even the top Raven Ridge laptop can beat it. Maybe come close within 20% or so but definitely not beat it.

 

[PeterScott]

What about the desktop Raven Ridge? The top model which i guess would be around $180-190 better beat the 1030 for that price.

Since MX150 is a desktop 1030, i find it doubtful even the top Raven Ridge laptop can beat it. Maybe come close within 20% or so but definitely not beat it.

So it better beat it, but you think it definitely won't?

It may be close to a 1030. It' hard to say, but a separate device, not having to share a memory bus will still likely come out ahead, even if they have similar bandwidth. The APU has to share it. A dGPU doesn't.

Great if it does get close, but really I think if you are build a desktop in a big case with card slots, and you care at all about GPU performance, get a GPU card for one of those slots.

An APU will excel in smaller form factors. There are many cases without room for card slots, or cool cases with passive cooling that works for the APU, but not for a GPU card.
http://www.streacom.com/products/db4-fanless-chassis/

There are so many uses for a good APU, and Intel nearly owned the market. Something like 60% of PCs sold were Intel IGP machines. This is where AMD can make some big headway, with a competitive CPU and superior GPU.

I am waffling on my next PC. Another big box with slots (definitely with a GPU card), or something really neat, smaller and probably passively cooled (definitely an APU)...

 

[LightningZ71]

Considering how both memory bound and processor core IPC bound the previous APUs were, I expect RR to be a significant improvement in the overall experience for several reasons. The cores are more efficient, both with power and performance in single threading (vs. previous cores). The memory bandwidth should improve significantly with DDR4 as opposed to older LPDDR3 (this is for much of the previous generation of delivered devices, I'm aware of the last APU having DDR4). The effective memory bandwidth should improve (Ryzen CCXs have L3 cache, most previous APUs didn't, reduces CPU memory bandwidth demands, VEGA modules have better memory bandwidth usage efficiency as compared to previous GCN units). The actual power efficiency of the total package should improve (newer, more power efficient process should enable more actual throughput per watt when processor is TDP limited). Thermal efficiency should improve (newer process has better thermal characteristics, enabling more performance per unit of thermal load). I don't think that it's beyond the realm of possibility for a properly designed laptop to easily more than double the performance of the previous best APUs when connected to wall power and provide better performance for a longer runtime for a given battery capacity.

I'm hoping that AMD sees fit to release a version of the RR that has, in a similar fashion to Iris Pro and the jaguar core in the XBOX One that they already produce, a local ESRAM cache of at least 16MB. Or, at least keep the second half of the Ryzen core L3 cache of 4MB, and dedicate that to the GPU side of the core. You and I and everyone reading this knows how badly the notebook manufacturers have been configuring most of the APU laptops when many of the processors in them are capable of dual channel operation, but they cheap out on the board and only implement a single channel. That sort of cache could help keep some semblance of decent performance for those laptops. Heck, even keep that second half of the L3 just for the CPU cores and give the CCX 8MB and give the GPU memory channel priority.

Has anyone seen anything about AMD producing a 15w RR product? I'd imagine that it would be a difficult target to hit, even with just two cores enabled, the L3 halved, and the GPU seriously gimped.

 

[krumme]

This product will be highend. It fits 14llp and single ccx zen perfect.
So it will have a price to match it. The alternative is iris pro or some expensive 4c i7 nv solition that probably uses more power. Everyone and his brother will want this apu in their laptop. Ofc. Why not? It will not come cheap but the ultrabook market is changed forever. I have had my part of expensive 2c ultrabooks with slow graphic. Not anymore. It cant come soon enough.

Eg my current skl lenovo thinkpad t460 have a chassis for 35w tdp. And i have this two core dud. Man the thought of 4 core and some decent graphics is really a huge huge uplift. Its like this market have stood still like forever.

 

[cbn]

Well that is ridiculous to even consider. 1050 can draw +60watts which is more than Raven Ridge will be rated for both CPU and GPU all together by a large margin!

A system with an equivalent Intel quadcore and 1050 would last maybe 1 hour on a battery where the RR system, given the same battery, could feasibly last over twice or three times as long. If you're on an airplane, bus, or anywhere away from home and you are playing some games the #1 priority is battery life.

One thing a person could do with the GTX 1050 laptop is use MSI Afterburner to lower max power draw and core clocks on the GPU.....think something like a DIY "Max Q" version of the GPU (though the CPU I see in laptops with GTX 1050 is still 45W processor).

Also perhaps this is something that should be considered for anyone testing with Blender? (See my previous post about Blender and video editing)

P.S. GTX 1050 2GB VRAM* laptops (with Core i5) are currently priced at $650 (After $100 rebate) and up. Normal price appears to $800 (around the same price as the MSI Core i7 MX150 laptop).

*Some of the GTX 1050 laptops (like this one for $800) have 4GB VRAM (unlike the desktop GTX 1050 which is always 2GB)

 

[Bouowmx]

Now these rebated entry-level NVIDIA laptops are interesting. Are the rebates consistent for each generation? Or just a fluke for Pascal?

 

[cbn]

Now these rebated entry-level NVIDIA laptops are interesting. Are the rebates consistent for each generation? Or just a fluke for Pascal?

Usually the rebates and discounts come out after products have been out for a while.

And GTX 1050 has been around for almost a year now. (This, in contrast, to the Nvidia Pacal MX150 laptops which are very recent and I haven't seen discounted yet)

 

[majord]

Well AMD's claims are pretty clear, albiet too simplistic.. 40% more graphics performance @ 15w vs 35w 7th series. They show this comparison because it looks better of course, (than simply comparing 15w vs 15w APU.) since BR was optimised for 15w, hence 35w didn't have much performance uplift.

Knowing 35w is somewhere between 20-30% faster, one can conclude, the claims are somewhere between 1.6 to 1.7x of 15w Bristol ridge.

I'm sure the reality in games will be a bit lower, though there's many many things potentially in RR's favour:

some key things We're yet to see, afaik?:

- DDR4-2400 support @ 15w (this combined with current gen DCC would increase effective bw by 1.4x over 15w Bristol

- L3 cache - will it be available to the GPU ? or , in a more helpful way than that of carrizo's access to the shared L2

- 16 ROPs, yay or nay? , i think if clockspeeds are going to be kept low, (compensated by CU increase) this would be advantageous, even though BW is still quite low.

Everything else - the bread and butter stuff, is pretty much a given..

- Much higher CPU performance will mean more TDP can be channeled to the GPU without causing a partial cpu bottleneck (and hopefully smarter GeAPM implementation ) ,

- L3 cache: Regardless of any chance to how cache is shared, the large L3 will significantly lower burden on DRAM bandwidth from the CPU during gaming workloads, further increasing the effective bandwidth from the GPU's perspective

 

[cbn]

Well AMD's claims are pretty clear, albiet too simplistic.. 40% more graphics performance @ 15w vs 35w 7th series. They show this comparison because it looks better of course, (than simply comparing 15w vs 15w APU.) since BR was optimised for 15w, hence 35w didn't have much performance uplift.

Are you referring to the iGPU by itself? Because I know at 35W the combined CPU and iGPU performance of Carrizo was much better than it was at 15W . (see link below)

http://www.portvapes.co.uk/?id=Latest-exam-1Z0-876-Dumps&exid=thread...or-compared-to-carrizo.2459807/#post-37946141

(Bristol Ridge uses the same die as Carrizo, so I expect the difference at 35W and 15W to be similar)

 

[AtenRa]

I have no idea if it is legit or not

https://gfxbench.com/compare.jsp?be...D+A12-9800E+RADEON+R7,+12+COMPUTE+CORES+4C+8G

 

[cbn]

- DDR4-2400 support @ 15w (this combined with current gen DCC would increase effective bw by 1.4x over 15w Bristol

- L3 cache - will it be available to the GPU ? or , in a more helpful way than that of carrizo's access to the shared L2

(snip)

- L3 cache: Regardless of any chance to how cache is shared, the large L3 will significantly lower burden on DRAM bandwidth from the CPU during gaming workloads, further increasing the effective bandwidth from the GPU's perspective

I believe the effective bandwidth would much greater than 1.4 since Bristol Ridge at 15W only supported DDR4 1866. (1.4 effective bandwidth would imply only a 9% increase in Delta Color Compression going from GCN 1.2 to GCN 1.3 and this document from AMD (on page 10) is showing a ~17% increase going from R9 Fury X (GCN 1.2) to RX 480 (GCN 1.3). Furthermore, according to the article below Vega implemented DSBR (rather than another increase in Delta color compression) to further reduce memory bandwidth requirements)

http://techreport.com/review/31224/the-curtain-comes-up-on-amd-vega-architecture/3

The DSBR can schedule work in what AMD describes as a "cache-aware" fashion, so it'll try to do as much work as possible for a given "bundle" of objects in a scene that relate to the data in a cache before the chip proceeds to flush the cache and fetch more data. The company says that a given pixel in a scene with many overlapping objects might be visited many times during the shading process, and that cache-aware approach makes doing that work more efficient. The DSBR also lets the GPU discover pixels in complex overlapping geometry that don't need to be shaded, and it can do that discovery no matter what order that overlapping geometry arrives in. By avoiding shading pixels that won't be visible in the final scene, Vega's pixel engine further improves efficiency.

To cure this headache, Vega's render back-ends now enjoy access to the chip's L2 cache in the same way that earlier stages in the pipeline do. This change allows more data to remain in the chip's L2 cache instead of being flushed out and brought back from main memory when it's needed again, and it's another improvement that can help deferred-rendering techniques.

The draw-stream binning rasterizer won't always be the rasterization approach that a Vega GPU will use. Instead, it's meant to complement the existing approaches possible on today's Radeons. AMD says that the DSBR is "highly dynamic and state-based," and that the feature is just another path through the hardware that can be used to improve rendering performance. By using data in a cache-aware fashion and only moving data when it has to, though, AMD thinks that this rasterizer will help performance in situations where the graphics memory (or high-bandwidth cache) becomes a bottleneck, and it'll also save power even when the path to memory isn't saturated.

By minimizing data movement in these ways, AMD says the DSBR is its next thrust at reducing memory bandwidth requirements. It's the latest in a series of solutions to the problem of memory-bandwidth efficiency that AMD has been working on across many generations of its products. In the past, the company has implemented better delta color compression algorithms, fast Z clear, and hierarchical-Z occlusion detection to reduce pressure on memory bandwidth.

 

[formulav8]

It's too bad it looks like they will miss back to school this year, when they sell a ton of laptops.

The best I've heard them mention was holiday season time-frame for laptops. AM4 is a 1/4 later at least. So XMAS area for lappies and up to March for AM4 sockets.