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

[itsmydamnation]

Low-cost HBM is nothing but concept now.

I would call it more then concept, its taking a bunch of pre-exsisting technologies and methods and packaging them together. Really its just a trade off between power and cost, if the market demand is there i would expect it to be productized quite quickly.

 

[Yotsugi]

if the market demand is there i would expect it to be productized quite quickly.

It's been over a year after Samsung's HotChips2016 presentation and yet we hear nothing about it.

 

[itsmydamnation]

It's been over a year after Samsung's HotChips2016 presentation and yet we hear nothing about it.

Yes and the SOC's things like this would use take ~2-4 years to develop............

 

[neblogai]

Here are some R7 2700U clock calculations done by dividing CB score: if IPC is about the same as Ryzen Desktop (which it should be)- then R7 2700U can do continuous ~2500MHz (CB-553) all thread load at 25W. At 15W- it should be able to do continuous ~2050Mhz (CB-450). Boosted 719CB score must come from about 3250MHz average clock, while single thread boost of 144 points- from one core at 3.5GHz.
Acer Swift 3 (2500U) score of 589 could be explained by the APU going higher on a cool system than expected sustained load.

But from reading the information AMD has provided in the slides, I wonder what devices these APUs would be best for. Here is what I gather:
1) We have no info on how efficient it will be at light tasks. The only efficiency data AMD gave was CPU running Cinebench on RR being 58% more efficient than Bristol Ridge- which is not useful here. I'd think Intel should have better battery life in ultra-portable because of LPDDR use.
2) The info regarding video playback: as IntelUser2000 has already mentioned- RR efficiency in it seems to be only marginally better than BR. Intel, on the other hand- has updated their U8x50 chips and they run video at almost no power. So Ultrabooks built just for browsing and playing video might have way lower battery life (~6hours) than Intel's latest generation (~10 hours). Number of hours in AMD's slide No39 (9-12h) do not make sense, at it shows Bristol Ridge being able to play H264 for 10.6 hours on a 50Wh battery. Real life tests, however, have Bristol Ridge running video only ~5 hours with same battery size. Maybe AMD provided video playback numbers with screen switched of? Anyway- as long as AMD does not have video playback as efficient as Intel- they will have hard time competing in Ultra-portable market.
3) RR CPU at high loads seems to be as efficient and fast as Intel. Which is great- it should be a good fit for almost all gaming (CPU + dGPU market). Intel may have an advantage at the most powerful parts- but at 15-35W TDP CPU AMD should be just as good.
4) I would not expect RR at 15W to be able to do much gaming at 15W cooling. Gaming results in AMD slides are from HP x360, which is probably running 25W or close to it (power brick is now spec'ed at 65W, not 45W). The performance is probably comparable with Kaveri desktop parts (95W TDP), or A8 9600- which can play basically every game at 30fps 720p, and e-sports at 1080p. These could be fine entry level gaming laptops, and better choice than Intel's iGPU because of better drivers, no graphic glitches, tearing removed by freesync, Chill saving the battery and keeping APU from throtling.
5) There should be 35-45W RR mobile parts later- which with DDR4 should be comparable to maybe A12 9800/nVidia 940mx results in gaming. All 11 active CUs with 2 GB of HBM2 could turn it into RX550/x150 or a bit faster at ~same TDP, but not sure if AMD has made it possible.

 

[Yotsugi]

All 11 active CUs with 2 GB of HBM2 could turn it into RX550/x150 or a bit faster at ~same TDP, but not sure if AMD has made it possible.

RR lacks HBM PHY.

 

[mikk]

HBM is a running gag, year for year all the HBM talk and hope and in the end nothing.

 

[Spartak]

Do we have any info on the all core turbo boost frequencies yet and how it works?

I assume this uses the available TDP headroom available when the GPU is not under load? Also curious how the different TDP profiles behave.

My guess would be with 12W at full CPU load there isnt much room for the GPU and there isn't all core turbo boost. With 15W there is more headroom for the GPU under full CPU load or all core boost and finally with 25W both GPU and CPU work under full load. The all core boost could also be higher compared to the 15W profile when the GPU is idling.

I really doubt there is all core turbo boost available on the 12W profile. Can't imagine the 2700U to run at 4*2,3+GHz @12W?

 

[Spartak]

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.

Yeah I was hoping for at least 3.0 base/3.6 turbo but seeing the 2700U I wouldnt be surprised if the desktop APU's at 35W would top out at 3.0 base/3.8 turbo. My money is on 3.0 base under GPU load but all core turbo around 3.3 when the GPU is idling.

 

[Glo.]

5) There should be 35-45W RR mobile parts later- which with DDR4 should be comparable to maybe A12 9800/nVidia 940mx results in gaming. All 11 active CUs with 2 GB of HBM2 could turn it into RX550/x150 or a bit faster at ~same TDP, but not sure if AMD has made it possible.

704 GCN cores, with 1.3 GHz core clock, and 160 GB/s bandwidth would be comparable to GTX 1050 - 1050 Ti desktop.

 

[Yotsugi]

704 GCN cores, with 1.3 GHz core clock, and 160 GB/s bandwidth would be comparable to GTX 1050 - 1050 Ti desktop.

And these are Vega SPs (I wish Vega actually worked).

 

[wahdangun]

So If there will be thread ripper version with quad channel ram, will the igp have sufficient bandwidth to perform ?

 

[Yotsugi]

So If there will be thread ripper version with quad channel ram, will the igp have sufficient bandwidth to perform ?

No, RR die lacks GMI links.
Besides, why?

 

[Madpacket]

So am I the only one worried about the "up to 8GB" statements made so far? Hoping this is just marketing failure as 8GB shared (6GB for OS, 2GB for GPU) could be pretty limiting.

 

[wahdangun]

No, RR die lacks GMI links.
Besides, why?

Hmm where do you find that RR will not have GMI link ?

I think it will be great for VDI server, so it can support many VM with remoteFX,

 

[wahdangun]

So am I the only one worried about the "up to 8GB" statements made so far? Hoping this is just marketing failure as 8GB shared (6GB for OS, 2GB for GPU) could be pretty limiting.

No it will support up to 64 GB of ram

 

[neblogai]

704 GCN cores, with 1.3 GHz core clock, and 160 GB/s bandwidth would be comparable to GTX 1050 - 1050 Ti desktop.

In structure- it looks similar to Polaris12- only a single compute engine with 11 NCUs- and 8 or 16 ROPs. Floating point performance would be up to 704x1300x2= 1.8TFLOPs if given a lot of power- which is between 1.2TFLOPs of mobile RX550, and ~2.1TFLOPs of RX460. So- slower than GTX1050/RX560 even if it had a lot of bandwidth.

 

[scannall]

So am I the only one worried about the "up to 8GB" statements made so far? Hoping this is just marketing failure as 8GB shared (6GB for OS, 2GB for GPU) could be pretty limiting.

No, just means the OEM's won't sell it to you with more than 8. It should support 64 gig if you install it yourself.

 

[PeterScott]

No, just means the OEM's won't sell it to you with more than 8. It should support 64 gig if you install it yourself.

64GB? That seems like an awful large assumption for low-mid end laptops.

I would make damn sure about ram slots and supported configurations, before leaping in assuming 64GB support.

 

[scannall]

64GB? That seems like an awful large assumption for low-mid end laptops.

I would make damn sure about ram slots and supported configurations, before leaping in assuming 64GB support.

Finding 32 gig dimms would be the hard (Impossible) part. The CPU would handle it is all. But 16 gig dimms should work, and are available.

 

[PeterScott]

Finding 32 gig dimms would be the hard (Impossible) part. The CPU would handle it is all. But 16 gig dimms should work, and are available.

There is more to it than that. What a CPU supports and what a laptop supports are two different things. Here is what the manual for the Bristol Ridge Envy says (likely what the first Raven Ridge design is based on):
http://h10032.www1.hp.com/ctg/Manual/c05527166

Memory For computer models equipped with an AMD FX-9800P quad core processor or AMD A12-9720P quad core processor:

Two SODIMM memory module slots, non-customer-accessible/non-upgradable DDR4-1866 dual channel support (DDR4-2400 downgrade to DDR4-1866) Supports up to 16-GB of system memory Supports the following conƭgurations ● 16384-MB (8192-MB × 2) ● 12288-MB (8192-MB + 4096-MB) ● 8192-MB (8192-MB × 1 or 4096-MB × 2)

 

[Glo.]

In structure- it looks similar to Polaris12- only a single compute engine with 11 NCUs- and 8 or 16 ROPs. Floating point performance would be up to 704x1300x2= 1.8TFLOPs if given a lot of power- which is between 1.2TFLOPs of mobile RX550, and ~2.1TFLOPs of RX460. So- slower than GTX1050/RX560 even if it had a lot of bandwidth.

So explain to me, why when you clock GTX 1050 Ti, and RX 460 EQUALLY - lets say at 1.2 GHz, both GPUs have the same level of performance in games?

If you will use Vega architecture, which is up to 15% faster per clock, than Polaris architecture, clock it at 1.3 GHz, and give it enough high bandwidth you will get just that - desktop GTX 1050 Ti performance level, because it has 768 CUDA cores, with 1.392 MHz boost clock. The difference between both GPUs: 704 vs 768 Cores; 1.3 vs 1.392 GHz is minimal, hence and higher IPC of Vega GPUs might result in exactly the same level between both graphics.

Unfortunately and fortunately for Intel and Nvidia - there is no HBM PHY/Integrated memory controller.

 

[wahdangun]

There is more to it than that. What a CPU supports and what a laptop supports are two different things. Here is what the manual for the Bristol Ridge Envy says (likely what the first Raven Ridge design is based on):
http://h10032.www1.hp.com/ctg/Manual/c05527166

It's really doesn't matter what OEM think, because RR is SOC.

 

[PeterScott]

It's really doesn't matter what OEM think, because RR is SOC.

It matters if the RAM is soldered in, or sockets are inaccessible, or if the bios doesn't support it.

 

[Yotsugi]

If you will use Vega architecture, which is up to 15% faster per clock

But it's not; at least not right now.

 

[scannall]

It matters if the RAM is soldered in, or sockets are inaccessible, or if the bios doesn't support it.

Yes, if the ram is soldered in, you're out of luck upgrading it. And the ultrathins pretty much have to do that. No room for dimms in them. When shopping for laptops, of course make sure it meets your needs. Including whether or not you need replaceable and accessible ram.