AMD RyZen CPU Architecture for 2017

So I decided to sell my 1700 and pick up a 2600X instead. Managed to get the 2600X from a local Microcenter for $190 so ended up being around $60 difference. I lose 2 cores but I gain 1Ghz on average and IMO 6c/12t should be fine for 95% of games. I had a lot of trouble keeping my 1700 stable at any decent clocks and its default barely sustains 3.2 Ghz consistently. This 2600X easily manages 4.2Ghz out-of-the-box in games and 3.9 Ghz in Handbrake without any tweaking. Since 90% of my use is in gaming, I'm a lot happier with the new CPU.

With the new Nv GPUs coming out, I might actually replace my 970 soon and have a decent gaming rig for maxing out modern games.
 
Last edited:
AMD just (silently) released what seem to be their "Raven Ridge 2018", in the form of the 45W "H"-series:

Ryzen 7 2800H
4-core/8-thread - 3.3GHz base, 3.8GHz turbo
Vega 11 "up to" 1.3GHz
Official support for DDR4 3200MHz

Ryzen 5 2600H
4-core/8-thread - 3.2GHz base, 3.6GHz turbo
Vega 10 "up to" 1100MHz
Official support for DDR4 3200MHz


These seem to be aimed directly at the Core 8x50U + MX150 combos in many affordable laptops with 13.3" and more.
With significantly higher TDP leeway, these can probably compete very well with those notebooks in performance.

Though nothing will come of it unless AMD enforces strict rules on RAM (obligatory dual-channel + 3000/3200MHz memory) and thermal performance next to laptop makers.
Which they didn't do for the Ryzen U-series, resulting in yet again a bunch of terrible performing laptops that stain AMD's brand.
Like this piece of work that throttles the GPU down to 300MHz in a game, or this one that has soldered single-channel RAM with no upgrade slot. (Both from Lenovo.. damn they must love AMD..)


My hopes for AMD turning this around comes from them explicitly saying Raven Ridge 2018's improvements coming from higher binnings but also from system design, at HotChips.
 
I don't think AMD is in a position to enforce anything, unfortunately. And big question will be idle power, in my opinion.
 
Though nothing will come of it unless AMD enforces strict rules on RAM (obligatory dual-channel + 3000/3200MHz memory) and thermal performance next to laptop makers.
I think high-speed ddr4 might still be somewhat problematic. About 99.9% of all ddr4 modules over ddr4-2666 (or rather ddr4-2800) are factory overvolted (1.35V instead of 1.2V) (even if they often list 1.2V, which they'll do on standard frequency, not their advertized speed). On the desktop, nobody ever cares, but for notebooks I'd suspect this is quite a problem.
That said, ddr4-3200 modules at 1.2V seem to exist now (they are all quite recent additions as far as I can tell). Kingston for instance has hyperx fury (dimm) and hyperx impact (so-dimm) which are explicitly saying they can do ddr4-3200 at 1.2V, if their website is to be believed... But even if it can do it, such expensive memory might really not be what a OEM might want to include...
So I suspect the fastest you can realistically expect is ddr4-2666 (hopefully with dual channel or at least an empty memory slot...).

edit: I even forgot, IIRC using memory speeds of ~3000Mhz or so already requires some quite increased "SoC voltage" (because the IF runs at half the mem clock), driving up power consumption of the chip itself. Again, not a problem for a AiO or SFF system, but for notebooks higher memory speeds may not be desirable even outside of cost considerations...
 
Last edited:
I think high-speed ddr4 might still be somewhat problematic. About 99.9% of all ddr4 modules over ddr4-2666 (or rather ddr4-2800) are factory overvolted (1.35V instead of 1.2V) (even if they often list 1.2V, which they'll do on standard frequency, not their advertized speed). On the desktop, nobody ever cares, but for notebooks I'd suspect this is quite a problem.
That said, ddr4-3200 modules at 1.2V seem to exist now (they are all quite recent additions as far as I can tell). Kingston for instance has hyperx fury (dimm) and hyperx impact (so-dimm) which are explicitly saying they can do ddr4-3200 at 1.2V, if their website is to be believed... But even if it can do it, such expensive memory might really not be what a OEM might want to include...
So I suspect the fastest you can realistically expect is ddr4-2666 (hopefully with dual channel or at least an empty memory slot...).

The hyperX impact even claims 3200MHz @ 1.2V without XMP, at standard SPD.
There's also the GSkill Ripjaws with 3000MHz @ 1.2V, which are neither rare nor very expensive. Sure it's only rated for those values using XMP, though if the memory supports it then the laptop manufacturers should have no problem forcing those clocks through the BIOS.
Besides, all the RAM makers are making DDR4 3200MHz chips that operate at 1.2V: Samsung, Micron and SK Hynix, and only the last isn't on mass production yet. Some OEMs could choose to purchase these chips and solder them directly into the motherboard.


Like I mentioned, my hope for this comes from the fact that AMD stated that the coming of the H-series Raven Ridge would be focused on system optimization. Meaning AMD may be working closely with laptop makers to optimize the systems where their SoCs are put..


edit: I even forgot, IIRC using memory speeds of ~3000Mhz or so already requires some quite increased "SoC voltage" (because the IF runs at half the mem clock), driving up power consumption of the chip itself. Again, not a problem for a AiO or SFF system, but for notebooks higher memory speeds may not be desirable even outside of cost considerations...
It's irrelevant that prior SoCs needed a manual tweak to get the memory controller to work at 3000MHz.
If AMD states the H-series can reach 3200MHz DDR4 at the rated TDP, then they have to provide that. Worst case scenario is they only support 3200MHz with the cTDP at 54W, though I find that really hard to believe.
If the H-series were released almost a full year after the U-series, it's because AMD did at least some optimization or tweaks to the Raven Ridge design.
 
Besides, all the RAM makers are making DDR4 3200MHz chips that operate at 1.2V: Samsung, Micron and SK Hynix, and only the last isn't on mass production yet. Some OEMs could choose to purchase these chips and solder them directly into the motherboard.
Alright maybe it's not that rare anymore (but still most ddr4-3200 memory is using higher voltage).

It's irrelevant that prior SoCs needed a manual tweak to get the memory controller to work at 3000MHz.
I'm not talking about manual tweaks, I believe the bios do this automatically (even for officially supported speeds like 2933).
If AMD states the H-series can reach 3200MHz DDR4 at the rated TDP, then they have to provide that. Worst case scenario is they only support 3200MHz with the cTDP at 54W, though I find that really hard to believe.
I'm not really worried about TDP, more the effect this will have on low-load / idle power draw.
 
Here's a very rare comparison, that shows GloFo's 14nm vs 12nm differences really well:


Ryzen 5 2500U vs. 3500U
Exact same platform (Acer Aspire A315-41), using the same 2*4GB 2400MHz memory and the same driver.
The iGPU is consistently clocking about 20% higher in the new SoC, meaning it's much more often bottlenecked by bandwidth. Even so, it's getting around 10% higher performance.

AMD's mobile APUs are in desperate need of higher bandwidth. Here's hoping (again) they will adopt LPDDR4/5 or something similar for the 7nm APUs, otherwise there's just no point to releasing higher performance iGPUs.
 
Faster ddr4 support would be a good start already. It's quite a pity the Ryzen 3x00U chips don't improve on that. ddr4 so-dimms up to ddr4-3200 (with non-overvolted 1.2V) are nowadays easily available. Although they are stuff you'd never find in notebooks pre-installed, the budget so-dimms go up to ddr4-2666 too, that would easily get you another couple percents in graphic workloads. But as I mentioned, higher memory speeds require higher Ryzen SoC voltage at some point, and maybe 2400 is the limit for whatever voltage the 15W chips use.
IMHO there's no way for lpddr5 support for the Ryzen 7nm APUs (which will appear only next year, but that's still too early), we'll first see this memory in smartphones.
lpddr4 would be very neat too, since it goes up to 4266 (with just 1.1V default voltage). But since those 7nm APUs (like the rest of the 7nm Ryzens) are still supposed to be all the same platform as the older ones, I doubt we'll see new memory supported with them - I guess technically it would be possible, and lpddr4 is all soldered always anyway so there's no real "motherboard incompatibility" there. But well maybe at least faster ddr4 support...
FWIW Icelake-U chips are rumoured to support up to lpddr4-3733, which would give them a rather massive advantage right there for graphics over Ryzen APUs (they are also supposed to have previously unheard of amounts of L3 cache just for the gpu to further help bandwidth deficits).
 
For APUs I can see a future with a Chiplet with 4C+IGP + HBM on the other linked by the OI Chip or maybe 8C+IGP with HBM on the die itself?
 
I feel like we should be seeing some performance leaks/more details from AMD by now?
7nm Ryzen is supposed to be 1h right?
 
Back
Top