AMD Raven Ridge 'Zen APU' Thread

[CatMerc]

I expect it will be slower than an RX550. The RX550 has 128 bit GDDR5 with 112 GB/S of bandwidth, almost double dual channel DDR4-3800. Also, RR is a laptop chip. It will be running DDR4 2400 at best.

If the RX550 could have got by with half the memory bandwidth, AMD would have given it a 64 bit bus, and built a less expensive card.

I expect RR will slot in under the RX550/GT 1030.

RR is Vega based though. DSBR and larger L2 cache are two benefits it should have over Polaris era GPU's - both help save bandwidth.
There are other cache increases which I forgot, but the architecture should also be less reliant on memory bandwidth for geometry processing.

Also consider that RX 460/560 had the same amount of bandwidth as that 550 and yet had far higher performance with double the shading units being fed by it. Polaris 12 had an overprovisioned bus size, for whatever reason. Possibly for memory capacity and costs reasons.

All in all, Raven Ridge has very strong potential for matching or potentially even beating RX 550 even when using 2666 memory, as long as it's in dual channel and given the proper power headroom. In any case, it's miles ahead of Intel's highest end Iris offering, without the need for expensive eDRAM. And I'm fairly certain 200mm^2 is still better than Intel's highest end Iris and Quad Core.

Here's a picture of i7-6785R. The long die is a quad core Skylake with 72 EU's, Intel's absolute top end.
RX 550 performs about twice as fast as that.

 

[AtenRa]

RR will be an OEM chip well into next year.

newegg.com

https://www.newegg.com/Product/Prod...IsNodeId=1&name=Desktop Computers&isdeptsrh=1

There are 518 Desktop Computers with HD630
There are 520 Desktop Computers with HD620
There are ZERO Desktop Computers with GT1030

RR even with 2400MHz memory will do fine against the competition on the OEM desktops/Laptops

 

[DrMrLordX]

without the need for expensive eDRAM.

I am beginning to wonder about the value of eDRAM on Intel machines anyway. On some bog-standard OEM box running DDR4-2133, sure, it will have benefits. But push the DDR4 speeds high enough (or go to DDR5) and eDRAM seems to lose its benefits, at least on Skylake/Kabylake/Coffeelake.

 

[PeterScott]

newegg.com

https://www.newegg.com/Product/ProductList.aspx?Submit=ENE&N=100790619 601286690 601286691 601286692&IsNodeId=1&name=Desktop Computers&isdeptsrh=1

There are 518 Desktop Computers with HD630
There are 520 Desktop Computers with HD620
There are ZERO Desktop Computers with GT1030

RR even with 2400MHz memory will do fine against the competition on the OEM desktops/Laptops

Since this is primarily about laptops, why are you only searching desktops? Also GT 1030 is quite new. Look how many laptops have 940mx, or GTX 1050.

No one anywhere is arguing that this won't beat Intel iGPUs, but that is a pretty weak battle.

 

[CatMerc]

I am beginning to wonder about the value of eDRAM on Intel machines anyway. On some bog-standard OEM box running DDR4-2133, sure, it will have benefits. But push the DDR4 speeds high enough (or go to DDR5) and eDRAM seems to lose its benefits, at least on Skylake/Kabylake/Coffeelake.

It looses value with dual channel 3200MHz memory. Before that however it does give a good bandwidth boost.

Since this is primarily about laptops, why are you only searching desktops? Also GT 1030 is quite new. Look how many laptops have 940mx, or GTX 1050.

No one anywhere is arguing that this won't beat Intel iGPUs, but that is a pretty weak battle.

The current #1 best seller under traditional laptops at Amazon is an HD 620 graphics equipped laptop. There is a huge market that Raven Ridge can completely change there.

 

[PeterScott]

The current #1 best seller under traditional laptops at Amazon is an HD 620 graphics equipped laptop. There is a huge market that Raven Ridge can completely change there.

I agree and have said so before. I have said RR is more important than Desktop Ryzen.

I just think people shouldn't get wrapped up in dGPU beating performance that is unlikely.

 

[CatMerc]

I agree and have said so before. I have said RR is more important than Desktop Ryzen.

I just think people shouldn't get wrapped up in dGPU beating performance that is unlikely.

There is potential for matching the RX 550, which is a good match to the GT 1030. And having everything integrated under one package makes it cheaper for laptop manufacturers.

 

[maddie]

Regarding the HBM2 equipped RR variant. If this rumor proves to be accurate, namely Vega 28 & 32 replacing P10, it might not be so farfetched. A premium APU mainly for the laptop market will be pretty much an uncontested product.

http://www.guru3d.com/news-story/amd-might-replace-rx-500-cards-with-rx-vega-32-and-28.html

Either AMD is screwing up hugely in the GPU market or a lot of more affordable HBM2 capacity is close to coming online.

 

[LightningZ71]

Where do AMD APUs currently sell the most volume? In the $300-$500 end of the laptop market. Their biggest competition there is dual core Celeron and quad core Pentium as well as a few stray outdated i3 chips. If AMD continues that trend, they will have a FAR better product than anything that competes there. Continue the exercise, what screens do those laptops typically come with? Most will have 720p displays with the near $500 ones nudging into 1080p territory. The A10 and A12 APUs drive the 720p displays plenty well enough for playable games with decent quality settings. They even give passable frame rates at 1080p low quality. AMD is advertising the RR chips to have at least 40% better performance than the A10-A12 chips, which translates to solid 720p performance and decent 1080p performance in that same price range. Heck, even my son's old puma cored a8 APU $350 HP laptop can play most popular mmo and esports titles plenty well enough at 720p, and that's with single channel DDR3. What's the R5 Ryzen mobile chip going to be able to do?

This is going to put casual gaming into the hands of a lot of laptop buyers that wouldn't have been able to touch most of it before. More than anything, AND needs to push for either faster ddr4 SODIMMs, or, push for a revision of RR on an MCM with HBM.

 

[PeterScott]

Regarding the HBM2 equipped RR variant.

I don't even think that was a Rumor. It was more like wishful thinking.

If you are going to build an HBM based APU, it's going to be an expensive big die (to justify HBM costs) niche product.

AMD likes to get a LOT of use out of a tapeout, so they probably can't justify the cost of taping out such a product.

 

[Arachnotronic]

I don't even think that was a Rumor. It was more like wishful thinking.

If you are going to build an HBM based APU, it's going to be an expensive big die (to justify HBM costs) niche product.

AMD likes to get a LOT of use out of a tapeout, so they probably can't justify the cost of taping out such a product.

Even if they taped it out, who would buy it?

 

[Seronei]

I don't even think that was a Rumor. It was more like wishful thinking.

If you are going to build an HBM based APU, it's going to be an expensive big die (to justify HBM costs) niche product.

AMD likes to get a LOT of use out of a tapeout, so they probably can't justify the cost of taping out such a product.

I don't think many people expect them to make another die for HBM based APUs, if it's happening it's already part of Raven Ridge and you just need to connect the die to the HBM and then sell it as a premium Iris Pro competitor that will absolutely crush it GPU performance wise..

Even if they taped it out, who would buy it?

Same market as Iris Pro: low power, small form factor premium laptops. Potentially small PCs like the intel NUCs as well.

 

[DrMrLordX]

It looses value with dual channel 3200MHz memory. Before that however it does give a good bandwidth boost.

Yeah, that's what I was thinking. Unless eDRAM gets an upgrade, I see it as being rather pointless once DDR5 rolls out and/or once OEMs decide to start making DDR4-3200 a standard.

Which they should, because there are a lot of DIMMs hitting those speeds on Intel platforms now at relatively low cost. It wouldn't take much for AMD to make that happen on their end save the cloying DDR shortages.

If AMD continues that trend, they will have a FAR better product than anything that competes there.

AMD still has their work cut out for them with respect to OEMs. They have to figure out how to get their product WITHOUT a dGPU into those $300-$500 laptops with proper dual-channel memory configurations. And preferably no 5400 rpm spinners.

 

[maddie]

I don't even think that was a Rumor. It was more like wishful thinking.

If you are going to build an HBM based APU, it's going to be an expensive big die (to justify HBM costs) niche product.

AMD likes to get a LOT of use out of a tapeout, so they probably can't justify the cost of taping out such a product.

I see this from you a lot in your arguments.

What exactly is the cost of HBM2? No BS now, a straight up value.

 

[Seronei]

I see this from you a lot in your arguments.

What exactly is the cost of HBM2? No BS now, a straight up value.

The HBM2 on the Vega gpus supposedly cost $175 for interposer and packaging. 4gb HBM should be around $75($100 with interposer and packaging) currently if this is accurate. This is only going to get cheaper with more adoption and competition as SK Hynix should start being a factor.

 

[Thunder 57]

I don't get all this discussion about RR matching low end dGPUs. Anyone who wants that kind of performance in a laptop will get one with a dGPU. Joe consumer will instead get a 4C8T RR that should be very competitive with the most popular Intel CPU's, along with a much better iGPU.

I really don't think AMD is trying to go after the small market of laptop purchasers that opt or a dGPU. They are going after Intel, and should be quite successful as long the the OEMs don't cripple them.

 

[cbn]

I don't even think that was a Rumor. It was more like wishful thinking.

If you are going to build an HBM based APU, it's going to be an expensive big die (to justify HBM costs) niche product.

AMD likes to get a LOT of use out of a tapeout, so they probably can't justify the cost of taping out such a product.

Even if they taped it out, who would buy it?

HPC and Workstations.....this if it has 1/2 rate DP floating point (like Bristol Ridge).

P.S. Autodesk CFD uses a dual precision solver, but I don't know if it uses the GPU for this? If it doesn't maybe having more GPU hardware with 1/2 rate DP FP would make give Autodesk incentive to make the Solver also work on the GPU?

 

[PeterScott]

I don't think many people expect them to make another die for HBM based APUs, if it's happening it's already part of Raven Ridge and you just need to connect the die to the HBM and then sell it as a premium Iris Pro competitor that will absolutely crush it GPU performance wise..

I think you need another die to support HBM. The memory controllers are completely different for HBM and DDR AFAIK.

Which is why there was never a R9 Fury with GDDR.

Simply adding expensive memory to the small GPU isn't likely to boost performance that much, It has significantly less GCN cores than RX560. So if you are going for HBM you may as well size up to take advantage of the extra BW.

 

[maddie]

I don't get all this discussion about RR matching low end dGPUs. Anyone who wants that kind of performance in a laptop will get one with a dGPU. Joe consumer will instead get a 4C8T RR that should be very competitive with the most popular Intel CPU's, along with a much better iGPU.

I really don't think AMD is trying to go after the small market of laptop purchasers that opt or a dGPU. They are going after Intel, and should be quite successful as long the the OEMs don't cripple them.

I don't understand the need to be stuck in a given niche. It's not just being able to match a dGPU model. To me, it's clear that an HBM2 outfitted RR having equal performance to a dGPU model, will have either much longer battery life, or be much smaller & lighter. How can this not be a positive? There are some prestige laptop makers that would welcome this.

 

[PeterScott]

I don't understand the need to be stuck in a given niche. It's not just being able to match a dGPU model. To me, it's clear that an HBM2 outfitted RR having equal performance to a dGPU model, will have either much longer battery life, or be much smaller & lighter. How can this not be a positive? There are some prestige laptop makers that would welcome this.

It's a negative when the memory is expensive ( see Seronei's post), when you need a new die (further driving up price) and then after driving up the cost a huge amount, you still have less than RX560 performance.

 

[cbn]

Simply adding expensive memory to the small GPU isn't likely to boost performance that much, It has significantly less GCN cores than RX560. So if you are going for HBM you may as well size up to take advantage of the extra BW.

Yep, for regular consumer workloads I would agree.....but 1/2 rate DP floating point (assuming Raven Ridge has this) needs a lot more bandwidth. Think the type of engineering that would use a 235W 3584sp Nvidia Quadro GP100 with its 1/2 rate DP and 4096 bit HBM2 rather than a 250W 3840sp Quadro P6000 with its 1/32 rate DP and 384 bit GDDR5.

https://www.anandtech.com/show/11102/nvidia-announces-quadro-gp100

Meanwhile the second market for the Quadro GP100 is the traditional high-end CAD/CAE market. For those more specialized users who need a workstation card with fast FP64 performance and ECC memory for maximum accuracy and reliability, the Quadro GP100 is the first Quadro card since the K6000 to offer that functionality. Arguably this is a bit of a niche, since most CAD users don’t need that kind of reliability, but for those who do for complex engineering simulations and the like, it’s critical

 

[maddie]

I think you need another die to support HBM. The memory controllers are completely different for HBM and DDR AFAIK.

Which is why there was never a R9 Fury with GDDR.

Simply adding expensive memory to the small GPU isn't likely to boost performance that much, It has significantly less GCN cores than RX560. So if you are going for HBM you may as well size up to take advantage of the extra BW.

I think he means that the HBM PHY is also there in addition to the DDR4 memory controllers. HBM memory controllers are a small fraction the size of DDR 4 or 5 ones with most of the memory interface logic being on the base HBM die. There is also an incoming cheaper HBM spec with 512bit bus, further reducing the size.

We have already seen AMD incorporate structures in Zen on the die that are not needed for a given market. Ryzen vs Epyc as an example.

 

[cbn]

There is also an incoming cheaper HBM spec with 512bit bus, further reducing the size.

Yep, the low cost HBM2:

I wonder if a HBM2 1024 bit controller will be compatible with both the regular 1024 bit HBM2 and the upcoming low cost (512 bit) HBM2? If not maybe future APUs will have more than one type of HBM controller on them?

P.S. Too bad the low cost HBM2 doesn't have ECC (ECC needed for workstation and HPC usage of FP64)....it would obviously be good for regular consumer though.

 

[maddie]

It's a negative when the memory is expensive ( see Seronei's post), when you need a new die (further driving up price) and then after driving up the cost a huge amount, you still have less than RX560 performance.

You keep making these assumptions that ignore any possible change or innovation.

Is it more expensive to produce an HBM2 equipped APU? Obviously yes.

Is there a cheaper HBM variant to reduce costs? yes
Is there a market for such a premium product given it's power benefit to dGPU competitor? I suggest yes.
Can a new niche be created of a lighter, slimmer product of equal performance? Again I suggest yes
Can we make 1 die to span both worlds? Ryzen vs Epyc points to a yes.

Is the additional cost excessive? Ask AMD & customers, but my guess is no.

Why do some think the possible market segments are locked for eternity.

 

[Thunder 57]

I don't understand the need to be stuck in a given niche. It's not just being able to match a dGPU model. To me, it's clear that an HBM2 outfitted RR having equal performance to a dGPU model, will have either much longer battery life, or be much smaller & lighter. How can this not be a positive? There are some prestige laptop makers that would welcome this.

Just because it would make a good product for some people doesn't mean that it's worth the cost of development to make it. There's no denying there's a market for something like this, but is it big enough to make money to produce at this time? I doubt it.