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

[prtskg]

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.

That's HP's choice. Hope for better treatment as reputation of AMD improves.

 

[Madpacket]

I have an Asus Zenbook now with 8GB (non upgradable, soldered in etc.) and it makes running more than one VM at a time painful. Windows itself is snappy enough but I'll be sure to stay away from anything less than 16 going forward. Especially with these Raven Ridge APU's with up to 8 threads, multiple VM's should be a good fit. I really want an ultrabook with 15W TDP that can be used for heavier tasks. Moderate gaming performance is icing on the cake.

 

[Av9114]

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

Why are people talking about HBM? Are they hoping to add separate VRAM? Unless I totally misunderstand HBM, you really don't want to use it INSTEAD of DDR. In any case, I don't see how that makes sense. I don't think memory bandwidth is going to be the major bottleneck for a GPU that has to share a 15-25W TDP. If you're just looking for a design that can handle 1080p 60fps, I think what you really want is just a bigger GPU with like eDRAM if it really does end up being bandwidth limited. But I would also guess you'd need a little more thermal headroom too.

As far as memory is concerned a bigger concern would seem to be lack of support for LPDDR, it seems like there's a very good chance that these parts deliver on the performance side but not battery life. Specifically, it seems plausible that it is very competitive on high utilization battery life but not at lower utilization. That's purely speculative, but 2 of the 3 designs that were announced were larger than I would have expected for a 15/25W TDP and the smallest is limited to single channel memory and could also be running TDP down?

 

[hojnikb]

I really hope someone makes a NUC type mini pc using these chips. Another big thumbs up if they can make it passive (sorta like zotac CI series).
Would buy immediately.

 

[Madpacket]

Why are people talking about HBM? Are they hoping to add separate VRAM? Unless I totally misunderstand HBM, you really don't want to use it INSTEAD of DDR. In any case, I don't see how that makes sense. I don't think memory bandwidth is going to be the major bottleneck for a GPU that has to share a 15-25W TDP. If you're just looking for a design that can handle 1080p 60fps, I think what you really want is just a bigger GPU with like eDRAM if it really does end up being bandwidth limited. But I would also guess you'd need a little more thermal headroom too.

As far as memory is concerned a bigger concern would seem to be lack of support for LPDDR, it seems like there's a very good chance that these parts deliver on the performance side but not battery life. Specifically, it seems plausible that it is very competitive on high utilization battery life but not at lower utilization. That's purely speculative, but 2 of the 3 designs that were announced were larger than I would have expected for a 15/25W TDP and the smallest is limited to single channel memory and could also be running TDP down?

That's a bit worrying about the lack of LPDDR support. Although how much power does this really save? And does LPDDR come in higher speeds (2400) which is kinda necessary to get decent performance out of the GPU. I thought LPDDR's real advantage is during sleep mode, not really during idle use so worst case we may have to go back to hibernation for long periods of inactivity which kind of sucks, although not as bad since SSD's become standard. My Zenbook can easily sleep for a week and still a few hours left of active use, this almost instant sleep/on mode would be missed.

 

[moinmoin]

The highest LPDDR4(X) seems to go is 2133.

 

[NTMBK]

The highest LPDDR4(X) seems to go is 2133.

Samsung claim to have 4266. https://www.anandtech.com/show/10782/samsung-introduces-8-gb-lpddr44266-dram-package

 

[Arzachel]

Why are people talking about HBM? Are they hoping to add separate VRAM? Unless I totally misunderstand HBM, you really don't want to use it INSTEAD of DDR. In any case, I don't see how that makes sense. I don't think memory bandwidth is going to be the major bottleneck for a GPU that has to share a 15-25W TDP. If you're just looking for a design that can handle 1080p 60fps, I think what you really want is just a bigger GPU with like eDRAM if it really does end up being bandwidth limited. But I would also guess you'd need a little more thermal headroom too.

The memory bandwidth is definitely going to be a bottleneck. RX 550 has a similar shader throughput with ~3x the memory bandwidth and the 50W TDP includes 4GB GDDR5 so it's not as apples to oranges as it might seem at first glance.

As for HBM, using it instead of DDR for system RAM would actually be preferable for designs where the RAM chips are soldered anyways, if not for the fact that it costs an arm an a leg right now.

 

[Madpacket]

Samsung claim to have 4266. https://www.anandtech.com/show/10782/samsung-introduces-8-gb-lpddr44266-dram-package

Nice. Looks promising for single channel memory controllers like in smart phones but probably not very practical for RR unless price isn't much of a factor. Also isn't the memory controller of the mobile part limited to 2400Mhz?

 

[prtskg]

Nice. Looks promising for single channel memory controllers like in smart phones but probably not very practical for RR unless price isn't much of a factor. Also isn't the memory controller of the mobile part limited to 2400Mhz?

Yes the 2 launched chip supports ram only up to 2400MHz

 

[moinmoin]

Yes the 2 launched chip supports ram only up to 2400MHz

Be careful about the unit, what you wrote is 4800 MT/s which doesn't exist yet. I wasn't paying attention and what I referred to is actually in MHz. LPDDR4 is 1600 MHz/3200 MT/s and what NTMBK linked is commonly referred to as LPDDR4E with 2133 MHz/4266 MT/s.

On topic, would be nice for a future Ryzen Mobile SoC to support LPDDR4(E) indeed. (LPDDR4X appears to be an even lower power version of LPDDR4, using a voltage of 0.6 instead 1.1.)

 

[prtskg]

Be careful about the unit, what you wrote is 4800 MT/s which doesn't exist yet. I wasn't paying attention and what I referred to is actually in MHz. LPDDR4 is 1600 MHz/3200 MT/s and what NTMBK linked is commonly referred to as LPDDR4E with 2133 MHz/4266 MT/s.

On topic, would be nice for a future Ryzen Mobile SoC to support LPDDR4(E) indeed. (LPDDR4X appears to be an even lower power version of LPDDR4, using a voltage of 0.6 instead 1.1.)

Thanks.

 

[Kenmitch]

Looks impressive!

Makes a person wonder what AMD could do with a R&D budget like Intel or even Nvidia for that matter.

 

[krumme]

Looks impressive!

Makes a person wonder what AMD could do with a R&D budget like Intel or even Nvidia for that matter.

Make a new bulldozer

 

[el etro]

Matters more how much you manage your architectcs teams than how much money do you spend. Apple spent realively little money to develop its king-of-the-hill cores(Big and little).

 

[scannall]

Looks impressive!

Makes a person wonder what AMD could do with a R&D budget like Intel or even Nvidia for that matter.

You rapidly run into diminishing returns. As budgets rise, focus is often lost, and more administration layers are added.

 

[krumme]

A mere 300 persons made the core of ryzen. The imc was made outside.
Had this processor had a good imc and a high fmax process like intel it would probably have walked all over it for performance core vs core outside of more fixed function types like avx2. Some of the most advanced tech like prefetcher and smt is working extremely good. The core is really good.
Now its takes lot of man hrs to adapt to different dies and IF takes up a lot of space to handle some of the scalability in the product portfolio but for the basics design you aparently dont need more people.

 

[raghu78]

A mere 300 persons made the core of ryzen. The imc was made outside.
Had this processor had a good imc and a high fmax process like intel it would probably have walked all over it for performance core vs core outside of more fixed function types like avx2. Some of the most advanced tech like prefetcher and smt is working extremely good. The core is really good.
Now its takes lot of man hrs to adapt to different dies and IF takes up a lot of space to handle some of the scalability in the product portfolio but for the basics design you aparently dont need more people.

Yeah you don't needs thousands of engineers. You need a highly motivated, focussed team of skilled engineers with a strong lead architect. AMD have just got started with Zen. Process wise they will get a slightly better process in 2018 with 12LP. But in 2019 AMD will have a process to go toe to toe with Intel's best. Thats when it will get really interesting. Zen 2 vs Icelake is the contest of this decade.

 

[moinmoin]

I thought unlike the licensed IMC in previous chips Zeppelin actually has an MC designed in-house?

 

[Mockingbird]

When does the NDA on Ryzen Mobile expires?

Come to think of it, is there even such a date?

Did AMD sent any reviewers laptops with Ryzen processors?

 

[CatMerc]

TDP restrictions is obviously a problem for the graphics.

MX150 draws more power for the graphics alone, and you want the less efficient Vega architecture to perform while also sharing power with the CPU?

 

[mahoshojo]

TDP restrictions is obviously a problem for the graphics.

MX150 draws more power for the graphics alone, and you want the less efficient Vega architecture to perform while also sharing power with the CPU?

It's cheaper, maybe?

 

[lefty2]

Does anyone think Raven Ridge with 2GB of HBM + dual DDR4 controllers would be a good idea? The Vega iGPU would use HBCC to control caching between main memory and faster HBM.

 

[krumme]

Does anyone think Raven Ridge with 2GB of HBM + dual DDR4 controllers would be a good idea? The Vega iGPU would use HBCC to control caching between main memory and faster HBM.

Excellent. But its a 7nm product at the earliest imo. Hbcc have to be working in both driver and engines. Raja says its the most interesting feature but say its 5 years out.
Yeaa making the hardware was the easy and cheap part
For now they have to get the ngg path working. Hopefully this december.

 

[lefty2]

Excellent. But its a 7nm product at the earliest imo. Hbcc have to be working in both driver and engines. Raja says its the most interesting feature but say its 5 years out.
Yeaa making the hardware was the easy and cheap part
For now they have to get the ngg path working. Hopefully this december.

I think HBCC is already working in Vega, but you don't see a benefit unless the workset is bigger than VRAM.