Hot on the heels of the HP leak that showed the first AMD Raven Ridge based notebook that may be hitting store shelves later this year, another leak of potential Raven Ridge APU performance is cycling through. The AMD Ryzen 7 2700U with integrated Vega-based graphics architecture, and also rumored to have a ~35-watt TDP, is showing 3DMark11 graphics scores near that of the discrete NVIDIA GeForce MX150.
With a graphics score of 4072, the integrated graphics on the upcoming AMD APU is slightly behind the score of 4570 from the MX150, a difference of 11.5%. Interestingly, the Physics score on the Raven Ridge APU of 6419 is solid as well, and puts an interesting light on the 8th gen KBL-R processors. As you can see in the graph below, from two systems we already have in-house with quad-core parts, CPU performance is going to vary dramatically from one machine to the next depending on the thermal headroom of the physical implementation.
The HP Spectre x360 with the Core i7-8550U and the MX150 GPU is able to generate a Physics score of 8278, well above the leaked result of the Raven Ridge APU. However, when we ran the 3DMark11 on the ASUS Zenbook 3 UX490UA with the same Core i7-8550U, the Physics score was 6627, a 19% drop! Clearly there are configurability shifts that will adjust the performance of the 8th gen Intel parts. We are diving more into this effect in a couple of upcoming reviews.
Though the true power consumption of these Ryzen 7 2700U systems is still up in the air, AMD has claimed for some time that it would have the ability to compete with Intel for the first time in several generations. If these solutions turn out to be in the 35-watt range, which would be at or lower than the typical 15-watt Intel CPU and 25-watt NVIDIA discrete GPU combined, AMD may have a winning combination for mobile performance users to entertain.
“The AMD Ryzen 7 2700U with
“The AMD Ryzen 7 2700U with integrated Vega-based graphics architecture, and also rumored to have a ~35-watt TDP”
Yes but will the OEM’s mostly relegate the APU to 15 watt thin and light form factor laptop designs. I’d like to know if these APU benchnmarks were done at 35 watts using dual channel DDR4 memory.
That’s still a very good initial showing for Raven Ridge against Intel’s Intel 620 graphics and Intel’s Iris Plus Graphics 650 benchmarks about 51% – 59%(depending on SKU tested) higher in 3DMark 11(At 1280×720) on notebookcheck’s comparsion(1) than Intel’s UHD Graphics 620 that the i7-8550U uses.
There still is not enough infromations to make a resonable comparsion what with all the unknowns about the system configurations used for the testing but I do hope that HP will be getting some 35 watt ProBook Raven Ridge options with dual channel DDR4 memory of at least 16GB.
“Intel UHD Graphics 620 vs Intel Iris Plus Graphics 650”
I haven’t seen so many leaks
I haven’t seen so many leaks so recently since I found out about the Iraqi navy.
I don’t know about the 3d
I don’t know about the 3d Mark benchmark itself, but does the physics test run while any graphics are being processed? If so, then the difference in performance is pretty clear. In the IGP case you have GPU+CPU in one thermal limit. In the DGPU case you have them in separate thermal zones.
If the physics test runs while no graphics are going, then I don’t know. Maybe they can set the power limit of the processors and they just set them differently due to different thermal solutions in the two machines?
It has low level graphics
It has low level graphics running along with high compute physics
Don’t forget, this laptop is
Don’t forget, this laptop is stock specced with a SINGLE stick of DDR4 and therefore is running single channel, choking the APU for memory bandwidth. In a proper dual channel config, I expect Ryzen Mobile to quite exceed an MX150.
We don’t know that, and
We don’t know that, and tbh. I kinda doubt that, these are already fantastic results if we assume it has dual channel memory available vs the GDDR5 available on the MX150. Would be cool if I’m proven wrong.
According to the
According to the leaked specifications on the HP Envy in question, it is spec’d in single channel. 1x8GB DDR4-2400.
We still don’t know. The SKU
We still don’t know. The SKU you referenced had a 2500U.
Channels usually don’t change
Channels usually don’t change with the OEM laptop platform and its quite (depressingly) normal for vendors to stick AMD’s APU’s with a single channel of memory to save costs.
They also usually leave them heat limited too for some weird ass reason. Given Intel’s history of paying OEM’s to shaft AMD its not too hard to believe something like that is still going on with the consistent 2nd or 3rd rate laptop platforms AMD’s CPU’s and APU’s keep getting shoved into.
Personally I suspect with proper dual channels of DDR4 2666 or even 3000 these new AMD APU’s will be able to punch a bit above their expected weight class but laptops capable of allowing that aren’t going to be common or affordable or even necessarily available at launch from big name (ie. HP) OEM’s. I’d expect someone like ASUS or MSI to have one out that allows that by early 2018.
OK we need to keep
OK we need to keep expectations in check, but this looks awesome. I’m really curious how soon they launch and what we should expect in terms of configurations – will the HPs of the world just release configs with TN/1366*/HDD/1-channel memory or give these CPUs a decent shot?
Dang, 75% CPU performance at
Dang, 75% CPU performance at 300% power.
Intel have been advancing since Sandy Bridge, just not aiming that advancement at top-end performance.
Reading comprehension error:
Reading comprehension error: it’s 75% of the CPU performance with 5W lower total TDP. Pretty impressive tbh.
If you don’t care about the graphics it doesn’t help you, but then if you don’t care about the graphics it’s not really the APU for you.
The ‘physics’ score is the
The ‘physics’ score is the CPU score, and the i7-8550u is 15W TDP. But even if you assume the MX150 is also running at full 25W tilt during the physics test, that’s still 75% performance at 87.5% the power.
The GPU shouldn’t be running
The GPU shouldn’t be running during the physics test at all; it’s supposed to be entirely CPU-dependent.
You didn’t notice the test
You didn’t notice the test using the SoC alone? The one where even with pitiful iGPU performance the CPU couldn’t get enough power to soundly beat the 2700U’s CPU. And of course the 8550U & MX150 Combo has a higher CPU score, there’s no iGPU to share power with, you’d likely see the exact same happen if you added a dedicated gpu to Ryzen.
So it’s more like 98% CPU and 230% GPU performance at similar power use.
AMD has also advanced, in mobile, they managed to cram bulldozer & GCN into 15W laptop APU’s, that, to me shows that they’ve come a long way.
Even if you look at SoC
Even if you look at SoC alone, that’s 103% the performance at 43% the power. More than twice as efficient.
I don’t see any numbers for
I don’t see any numbers for power consumption. So, your whole claim is pointless. Since the launch of Zen AMD’s efficiency is on the same level. And better. That won’t change with RR. 😉
You’d have to be pretty
You’d have to be pretty ignorant to believe that, on Desktop Ryzen is showing to be pretty good as far as effeciency goes, and that’s with 14nm LPP running above what it’s meant for, I have no doubt it will be competitive like this with the 8550U at 15W.
This is a 15W APU, not 35W.
This is a 15W APU, not 35W. (U suffix = ultra low power)
I expected nothing less than at least double the efficiency compared to a two chip solution from Nvidia and Intel.
Zen is substantially more efficient than Intel’s current core.
AMD APU will be the #1 choice for thin form factor graphics.
That single channel to DDR4
That single channel to DDR4 would not be a problem for AMD’s Raven Ridge APUs if AMD would just create an APU with a single stack of HBM2. Even if the HBM2 only ammounted to only 1GB/2GB that would allow for Vega’s HBCC to make use of the HBM2 as HBC and then the rest of the VRAM could be virtual VRAM paged out to DIMM based DDR4 over a single or dual memory channel, or even an SSD/Paged Memory swap file.
Under Vega’s HBCC the HBM2 effectively becomes an L4 cache/HBC for any GPU/APU that has HBM2, or even eDRAM, on die/package. It really should be AMD’s highest priority to get at least one High End portable(laptop) based APU SKU that has some HBM2 and a larger number of nCUs and folks will gladly pay more for the possibility of better graphics performance non dependent on any main system memory bandwidth constraints.
For the portable workstation market it would really be nice for AMD to maybe produce a Radeon Pro “WX” branded Workstation Grade APU on an interposer with an HBM2 Feature and AMD needs to take it to the next logical level and get that market share. Intel lacks the GPU IP to really compete currently and AMD can really beat Intel on the APU graphics front with HBM2 there to supply the necessary bandwidth for any integrated graphics that could run without and main system memory bandwidth constraints such as what a single channel to system DRAM causes currently on AMD’s APUs.
AMD’s APU graphics have always been memory bandwidth constrained and that’s not as bad for non FPS dependent non gaming graphics workloads. But for gaming workloads where FPS can mean the difference between playable and non playable then AMD really needs what HBM2 can provide for the Vega graphics to really be able to provide higher FPS on any APU integrated graphics based SKUs. A single stack of HBM2 with a 1024 bit wide interface is more than 3 times as wide as even 2 channels to system DRAM and can provide more effective bandwidth using clock rates that are much lower and more power efficient than regular DIMM based DRAM.
APUs are more directly wired up CPU cores to Integrated Graphics cores with no extra latency inducing hops out over PCIe required for the APU’s CPU/GPU communication. So at least the Raven Ridge APUs have that advantage and with the Infinity Fabric maybe the Raven Ridge CPU cores and the Vega nCUs can share some direct CPU to nCU Cache Coherency traffic via the Infinity Fabric on the APU.
Once AMD finally officially introduces its Zen/Vega Raven Ridge APUs it would be nice if AMD would produce some more detailed whitepapers on any on APU Infinity Fabric Cache Coherency traffic between Zen cores and Vega nCUs and AMD is really lacking in the amount of technical hardware whitepapers relative to the number of whitepapers that Nvidia provides on its related hardware. AMD is more open with its software/driver, and middleware, ecosystem than Nvidia but AMD really needs to get more highly technical hardware related whitepapers published.
Adding HBM2 is an option
Adding HBM2 is an option later, when the prices come down. But at the moment AMD needs an efficient and fast apu, that can cover 90% of the laptop market. More graphic performance can always be added with a dedicated gpu.
You could probably add a
You could probably add a dedicated GPU for cheaper than HBM at this point. I wouldn’t mind a laptop with HBM, but it may have to be a dedicated GPU with HBM for a much higher price. HBM should have better power consumption than a GDDR5 so a mobile Vega would be interesting if the power consumption at lower clocks isn’t out of control. I have thought that it would be a good idea if they would just make a laptop with GDDR connected directly to an APU (this is what the Xbox One and the PS4 are essentially). Maximum memory capacity would be limited, but a lot of laptops have soldered on memory anyway. That solves the memory bandwidth issues at a price more than an APU with DDR4, but less than a dedicated GPU + separate CPU.
HBM/2 would indeed solve any
HBM/2 would indeed solve any bandwidth problems but its too expensive and difficult to do in 2017 or even 2018. APU’s are supposed to be cheap after all.
Maybe in 2019. —Maybe—
No not all APUs have to be
No not all APUs have to be Cheap, and people pay a lot for Intel’s SOCs. So AMD can price a single stack of HBM2 with 2GB or 4GB and can charge a higher price and I’d gladly pay extra for any Laptop APU based SKU with at least one stack of HBM2 as that’s plenty for the small numbers of nCU that come on Raven Ridge. That’s potentially 256GB/s of effective bandwidth for a single stack and Vega’s HBCC treating that 2GB or 4GB of HBM2(HBC) like a last level cache and manage in the background some extra virtual VRAM paged out to regular system DIMM Based DRAM while keeping the Vega Graphics working from the HBM2/Cache at high bandwidth.
AMD can very much afford to create some high end Raven Ridge/Pinnacle Ridge APUs with a single stack of HBM2.
Sk Hynix will probably have its HBM2 supply issues sorted out by the end of this year and production of smaller HBM2 capacities is eaiser than the 8-HI HBM2 variants. So if a single stack of 2GB/4GB of HBM2 and Interposer/Packaging adds $100-$150 to the price to an APU I’ll gladly pay it. It would not be hard for AMD to take the monolithic Reven Ridge Zen cores/Vega graphics die and repackage it on an interposer with a single stack of HBM2, and then any single channel to regular DIMM based DRAM would not affect Vega’s integrated graphics needed bandwidth as the Vega nCUs would be working from HBM2.
The most affrodable HBM2 dies would be the 2Hi at only 2GB but that 2GB used by Vega’s HBCC as HBC could very likely be managed by the HBCC for a Virtual VRAM pool size of 6GB-8GB paged out to regular DIMM based DRAM. And Vega’s HBCC can manage that with ease as backgroud paging transfers over a single or dual channel connection to/from the DIMM/s to HBM2/Cache. Look at Intel’s eDRAM size on its Iris Pro graphics with its size of only 128GB max and that little bit helps a lot. So 2GB of HBM2/HBC on a Raven Ridge high end variant will definitely be managable and able to leverage a much larger pool of virtual VRAM paged to and from regular system DIMM/DRAM.
Bottom line here is that
Bottom line here is that every 15W Kabylake-R is fitted with UHD 620. Which is going to get completely spanked, even by a 15W R5 2500U. Basically, when it comes to graphics, Intel are bringing a re-branded penknife to a gunfight. So they’re going to ‘get rekt’.