AMD launches Ryzen Mobile 7 2700U & 5 2500U with Vega Graphics

[Yakk]

Even at the 600 point score for Cinebench in the Acer laptop for the Ryzen 5 2500U,that is almost FX8350 level in an ultrabook CPU!

Seriously! My Dell Precision Workstation laptop suddenly lost A LOT of appeal to me.

Gonna be time to hand it down at work and I smell upgrade time coming, and shedding a few pounds in the process!

 

[majord]

For reference. A dirty old 1700 @ 3.2Ghz (stock All core TB) uses bang on 7w / core during cinebench run. - That's 28w total for just cores.

So one could take that as a worst case scenario for RR.. since the 700+ first-run score would require a 3.2ghz frequency throughout.

In reality though a top-binned RR SKU has a lot more in it's favour, so I'm thinking around 25w short-term 3.2Ghz burst is probably what we're seeing, the only assumption being RR has a lot lower 'Uncore'/SoC power consumption than Zeppelin

The sustained scores suggest settling around 2.5Ghz. pretty much the same as an 8550U

 

[Glo.]

IF Raven Ridge is able to be fit in 15W TDP, with very decent performanc, Im wondering what will the rumored low power desktop Chips be.

 

[Yakk]

IF Raven Ridge is able to be fit in 15W TDP, with very decent performanc, Im wondering what will the rumored low power desktop Chips be.

The GF/Samsung process is more geared towards low power from what I've read, more so than high performance so I'd expect similar performance to what we have for Ryzen already.

 

[krumme]

IF Raven Ridge is able to be fit in 15W TDP, with very decent performanc, Im wondering what will the rumored low power desktop Chips be.

Also remember these 4c apu can be configured down to 9w tdp.

 

[Yotsugi]

Also remember these 4c apu can be configured down to 9w tdp.

Actually 12-25W.

 

[cfenton]

I'm still a bit disappointed in the graphics performance. Yes, it's faster than Intel's junk, but it's still not fast enough for 1080p/60 even in the graphically simple esports games. I'm really interested in what they could do with a 45W chip in something like a 15" Macbook Pro. Doesn't have to be a Macbook, but just something that size and weight.

 

[Flash831]

I'm still a bit disappointed in the graphics performance. Yes, it's faster than Intel's junk, but it's still not fast enough for 1080p/60 even in the graphically simple esports games. I'm really interested in what they could do with a 45W chip in something like a 15" Macbook Pro. Doesn't have to be a Macbook, but just something that size and weight.

I believe that is a memory problem, not a problem with RR itself. DDR4 simply does not provide all necessary bandwidth. But that means the gpu part can run at lower speeds without losing performance.

 

[Glo.]

The GF/Samsung process is more geared towards low power from what I've read, more so than high performance so I'd expect similar performance to what we have for Ryzen already.

Well, that would be actually mind blowing if 4C8T lets say 3.3/3.6 GHz CPU would fit in 35W TDP, and hover around this power consumption under load. That would be amazing.

 

[flash-gordon]

I'm still a bit disappointed in the graphics performance. Yes, it's faster than Intel's junk, but it's still not fast enough for 1080p/60 even in the graphically simple esports games. I'm really interested in what they could do with a 45W chip in something like a 15" Macbook Pro. Doesn't have to be a Macbook, but just something that size and weight.

Just a full die with 2GB GDDR5 on HBC would put it much closer to this spec.

They have all the cards on the table now, I really can't understand why this isn't happening.

 

[Qwertilot]

Well, that would be actually mind blowing if 4C8T lets say 3.3/3.6 GHz CPU would fit in 35W TDP, and hover around this power consumption under load. That would be amazing.

Not really I think. Good of course But the i7700T is 35W/ 2.9 base and 3.8 boost, and I think they die pick their mobile stuff to get a little more efficiency. The very top CPUs from both people are clocked well past their optimum efficiency level.

 

[LightningZ71]

The biggest issue that the desktop RR SKUs will face is potential disabling of memory overclocking. We just saw that with BR, and I wouldn't be surprised to see it happen on RR desktop. If they don't lock the memory bus, then, the next big issue will be package total power limits. It shouldn't be too difficult to provide sufficient cooling for the processor when there are solutions for the 1800X that can hold it at 4.0Ghz all around. So, provided that there is a way in the BIOS to allow the APU to run both the GPU and the CPU cores at max turbo for extended durations, and provided that the memory can be overclocked to work with with the premium low CL high Mhz memory modules, we could see desktop parts with easily 40-40% better graphics performance for enthusiasts.

Now, the flip side is, does any of that make sense? If you could afford to spend $100+ dollars MORE on high end memory AND another $50+ on a high end processor cooler, why wouldn't you just use a mid range Ryzen 5 and get a Radeon 560 4GB card? NewEgg has a 560 4GB card for $109 right now. No matter what you do to that APU, you aren't going to get that video performance up enough to make a huge difference in those benchmark scores, and especially without spending much more than the cost of that 560. The only place that any of this makes sense is in a mini-itx case for use in an entertainment center or in a very small desk situation. Someplace where you can't use even a low profile video card. Then, you'll run into the problem of how do you dissipate all the heat that a max turboed RR will pump out? That case will be tiny. The RR will easily be pushing 80+ watts when pushed to it's limits, if not more. I'm just not seeing it as a better solution than just getting a case that can hold a 560 or low profile card and making that work with a desktop Ryzen.

For business usage, a RR pro will be a big deal. It'll do a lot of things that businesses want, and I can see it being acceptable for even some higher end creative use that doesn't need bleeding edge video cards, but still needs acceleration to perform well enough.

 

[firewolfsm]

If AMD added HBM to this processor it could change laptops dramatically. Motherboards would get smaller without sodimm slots or even soldered on memory, and bandwidth to the gpu and processor would increase. Also power consumption would be reduced without shuttling data over the bus.

 

[Schmide]

I was thinking about the 11cu s. It doesn't seem likely that 1 is disabled due to production defects, so I was wondering if they are actually cycling an idle cu to keep the power down and allow for lower temperature operation.

 

[VirtualLarry]

so I was wondering if they are actually cycling an idle cu to keep the power down and allow for lower temperature operation.

That's actually a really interesting thought. I wonder what effect that would have on overall performance, though? Switching threads through various CPU cores on an SMT machine does have a performance cost to it.

 

[epsilon84]

Does anyone have any idea how real world gaming performance of the IGP compares to say, a GeForce MX150? I saw one of the slides claimed 3DMark scores close to the MX150 but I'm guessing actual gaming perfomance wouldn't be quite as strong as the 3DMark scores suggest.

I'm looking at an ultrabook for work when I'm travelling and wouldn't mind a bit of light gaming on the side if possible.

 

[Shmee]

I must say that these APUs seem pretty impressive. Still, I would like to see higher end graphics with faster memory options for use in more high end laptops.

 

[cbn]

I wonder how cheaply they could offer a fanless NUC / mini-ITX version of this? They could use chips that don't meet the 15 watt binning and have them run at 20-30 watt TDP.

Akasa Euler covers up to 35W. The original version of this thin mini-itx case is about $110.....but the motherboard would need LGA 115x mounting holes* for it to work.

*The CPU mounting surface of the AMD BGA processor also needs to be at the same level as an Intel LGA 115x processor headspreader.

P.S. I suspect any chips that don't make 15W bin would probably become some kind of low bin 35W.

 

[Trovaricon]

FM2+ Thin ITX from ECS (A78F2-TI) uses intel mounting - precisely because of "intel-standardized" CPU placement in passive cases and cases with integrated cooling (e.g. DYI AIOs)

edit: I hope we will see multiple motherboard vendors offering thin ITX for AM4

 

[cbn]

FM2+ Thin ITX from ECS (A78F2-TI) uses intel mounting - precisely because of "intel-standardized" CPU placement in passive cases and cases with integrated cooling (e.g. DYI AIOs)

edit: I hope we will see multiple motherboard vendors offering thin ITX for AM4

Thank you. I updated this thread with the info from post (gave you credit as well).

 

[Jan Olšan]

The biggest issue that the desktop RR SKUs will face is potential disabling of memory overclocking. We just saw that with BR, and I wouldn't be surprised to see it happen on RR desktop. If they don't lock the memory bus, then, the next big issue will be package total power limits. (...)

Bristol Ridge has no locks on memory overclocking. The licensed memory controller IP just doesn't have support for raising the internal multipliers any higher than what is used for 2400 MHz. So you can get higher by raising BCLK, but then your system becomes unstable. It is the limit of the design, not arbitrary lock on OC.

 

[ksec]

1. These are still 14nm, and they could only get better with 12nm and 7nm down the road.

2. HBM, it used to be HBM was expensive, but now due to demand DDR Ram is up as well. I wonder if the price / performance would sway to HBM soon.

3. I am waiting for AMD to adopt Thunderbolt.

Finally i dont have to choose between CPU performance and GPU in Laptop.

 

[Yotsugi]

2. HBM, it used to be HBM was expensive, but now due to demand DDR Ram is up as well. I wonder if the price / performance would sway to HBM soon.

There's more to HBM than the memory itself.
2.5D integration is not the fastest process right now.

 

[itsmydamnation]

There's more to HBM than the memory itself.
2.5D integration is not the fastest process right now.

The Samsung cheap HBM was said to use organic interposer, if they did they could just have the HBM interface on the APU and optionally use the interposer+HBM on selected SKU's. I have always wondered if you could do that with silicon interposer i assume you can but i have never been able to find a resource saying yes.

 

[Yotsugi]

The Samsung cheap HBM was said to use organic interposer, if they did they could just have the HBM interface on the APU and optionally use the interposer+HBM on selected SKU's. I have always wondered if you could do that with silicon interposer i assume you can but i have never been able to find a resource saying yes.

Low-cost HBM is nothing but concept now.