The 45W intel HQ CPU's suddenly look very datedOriginally Posted by CAT-THE-FIFTH
The mobile-targeted Ryzen APUs should be pretty interesting. While they might struggle to hit the same clock speeds as Intel on GloFo's 14nm, it does look to be incredibly efficient at the lower clocks used on the likes of the 1700 (non-X), which bodes well for mobile.
Anyway to watch that video in the UK?
I'm really interested to see if they use a single CCX paired with a GPU. You would think it'd be obvious, but... this is AMD.
A 4c8t CPU with an AMD onboard GPU would be a real beast!
Which one?
Bits and Chips are at the rumours again. So far they have been pretty good with Ryzen leaks:
Source: https://twitter.com/BitsAndChipsEng/...73386391891968
Intel already announced the price of their 16C/32T Core i9-7960X @ $1,699, this is a lot less.
Max clocks on Threadripper is likely to be the same as for Ryzen, but Intel's are unknown. Computerbase had listed the TDP as 165W for the 14-18 core parts, while 10 cores has 140W and 3.3GHz (4.3GHz turbo), so it looks like 16 core doesn't have a big enough TDP to be running at those kind of speeds. The fastest current Broadwell Xeon seems to be the E5-2697A @ 2.6GHz base although that has a 145W TDP (and lists at $2,891).
You could add $200 and it would still seem good value in comparison to Intel.
Purposefully aimed at half the price?
Probably, marketing department wet dream.
Particularly given that the GPU has 11 CUs (704 shaders). Even at lower mobile clocks that should be ~ RX 550 performance. We're probably looking at base/boost clocks in the 2.5GHz/3.?GHz range. Their claimed performance gains (+50% CPU, + 40% GPU) line up almost exactly with the IPC gain and shader increase, so I'm pretty sure we're looking at the same clocks as current mobile parts (possibly even lower on the GPU).
Another consideration is that they're talking about doing that at half the power. AMD's mainstream Bristol Ridge APUs are ~ 35W, so you're looking at +50% CPU and + 40% GPU at 17.5W. What are the 45W top-end chips going to be like?
Well, the recent price cut to Ryzen 7 frees up a bit more pricing space. The 1700X is now at $349, and given Threadripper is basically two of those glued together, $849 could be argued to be on the high sideThat said, the 1800X is still $469, so if they're potentially looking at the top end TR being $999, $849 for one step down kinf of makes sense. They've then got a rnage of price points between $499 and $799 to slot other processors - however many cores they have - into....
Perhaps they'll use some of the increased power budget to squeeze on a stack of HBM2 (2gb). Iirc it's been confirmed that the CUs are Vega rather than Polaris, which means they could have HBCC too.Particularly given that the GPU has 11 CUs (704 shaders). Even at lower mobile clocks that should be ~ RX 550 performance... What are the 45W top-end chips going to be like?
I can dream right?
Of course you can - just don't bet your mortgage on it
I don't think we'll see HBM on this generation of APUs, given how much trouble they're having getting supplies for the full Vega cards! I'm sure that APUs with HBM caches will happen eventually - I mean, a single stack of HBM2 could give you 8GB of RAM, which is actually enough for the entire system...But I think that'll be a couple of generations away.
Sure, if you take TDP figures at face value. They actually use closer to 30w at full load.
I'd totally jump at a 14" laptop with a quad-core CPU and decent next-gen graphics. Been looking for a decent machine for years but had to go up to 15" to get anything sensible in it since at the time the best you could get in a <=14" was a 28nm 940M, which consumed >30w while being outperformed by a 15w Intel iGPU on 14nm.
Vega HBM would be a total game changer in this segment and shoving it on-package with a Ryzen CPU would be huge. Thermal matters aside, we'd be looking at a 75% or greater reduction in board area required which is the biggest impediment to powerful ultrabook and convertible systems right now.
Holy carp next year will be an exciting one for portable gaming performance.
The sooner the better in any case. The biggest advantages of HBM outside of raw performance are power and space requirements both of which are far more important on mobile platforms than on desktops. It's got the potential to be transformative in the portable space while being merely good-to-have-but-not-game-changing in desktops.Of course you can - just don't bet your mortgage on it
I don't think we'll see HBM on this generation of APUs, given how much trouble they're having getting supplies for the full Vega cards! I'm sure that APUs with HBM caches will happen eventually - I mean, a single stack of HBM2 could give you 8GB of RAM, which is actually enough for the entire system...
Completely agree with the principle of this, but I think it's worth saying something about thermals: desktop Ryzen and Polaris are both run for performance above efficicency, and require plenty of voltage to be stable there. Dial the clocks back at all and you can back off the voltage dramatically. Ryzen scales exceptionally right down to ~ 25w for an 8C/16T part, and Polaris has a steep voltage drop off as you back down to ~ 900MHz. Vega is meant to be more efficient, particularly in terms of IGP, compared to Polaris, and Ryzen APUs are apparently getting Vega-based IGPs, so they can really pull the clocks back on those and still get impressive performance. So even the first generation Ryzen APUs should be looking at excellent performance within very tight thermal envelopes.
I think the interesting thing will be how AMD approach integrating HBM onto APUs. Vega refers to its HBM as a high-bandwidth cache, with a controller that can directly access off-card volatile and non-volatile memory with a huge address space (which is how they're doing the professional cards with large NVMe storage on-card_. A similar approach but using that cache for the entire APU could easily make system RAM redundant - a 4GB or 8GB cache + fast SSD should make an overall fast system.
OTOH, that would be quite expensive, and I'd guess that as all the data has to traverse the data fabric on the chip there's currently no advantage to having a high bandwidth cache on-package as the latency to it would be determined by the data fabric, not the cache itself (much like stepping from the L3 cache of one CCX to the L3 cache of the other take a big latency hit). If they can speed the data fabric up and add a HBM cache that has lower latency than a trip to main memory, OTOH, I can start to see massive possibilities; it could operate as the primary system RAM for mobile devices and a high-speed L4 cache on desktop (where you'd probably have at least 16GB of system RAM anyway).
The Biostar X370GTN mini-ITX motherboard is only £100:
http://www.ebuyer.com/792153-biostar...tm_term=249371
How are Biostar's generally? I was going to buy a Gigabyte B350 Gaming 3 but that Biostar one seems like a good deal
There are currently 3 users browsing this thread. (0 members and 3 guests)
Posting Permissions
LinkBack URL
About LinkBacks
Reply With Quote