AMD Ryzen 5 the most warmly welcomed CPUs in seven years

[QuorTek]

AMD is a big player out there regardless of whichever... not a fanboy of either intel nvidia or amd though, usually going with best choice for me whenever I have to get some hardware... but thought about building an additional machine with vega + ryzen.

[Xlucine]

It might be a tad hard to put that in your laptop though?

The crazy thing about ryzen is that it's totally feasible. Scaling off the 65W TDP of a 1700, a die with 2 cores fused off should just squeeze into the 45W TDP that the 4 core intel chips in all the top end gaming laptops use (maybe with the clocks turned down a little). So with a single mask they're matching or beating intel in every benchmark, from 8-core prosumer down to laptops

[The Hand]

Well you can get 8 cores(4c/8t) with the Ryzen 5 1400 for roughly £150 and get 12 cores(6c/12t) with the Ryzen 5 1600 for £199.. AMD are on to a winner with the R5 range. Warmly welcomed they should be

[Carbon Juniors]

Nice to see!

[azrael-]

I'm *positive* about Ryzen as such. SMT issues or not.

I'm *negative* about AMD (baffingly) playing into Microsoft's hand by only supporting it on Windows 10. I can't see any true upside for AMD there.

[qasdfdsaq]

The server market doesn't care about halo products. It's all about perf/watt. The server market should be WAY more lucrative for AMD right now than the consumer market, so they're absolutely spot on to target perf/watt at lower clocks if they can only do one silicon run.

So why server first? Intel's current range topping server CPU offers 24C/48T @ 2.4GHz in a 165W thermal envelope. I reckon (based on the chart kompukare shared above) that AMD's naples processor will offer 32C/64T - i.e. 50% more parallel throughput - at the same clock speed and thermal envelope. That Intel chip? Suggested customer price is just under $9000. That's why server first.

AMD don't exist in the server market. If they had to do only one silicon run why exactly is said silicon being sold in the consumer market only with later silicon being planned for server? Server first, with no server chips... I'd love some of what you're smoking, if only to take the edge off AMD stocks tanking right now -_- Apparently, the "experts" on the stock exchange reckon AMD are doing a far worse job than you do, and sadly they're the ones defining how much money I make off AMD and not you.

I just can't understand the dinosaur idiots who keep harping about single-threaded performance. Give it a rest! That is so legacy in terms of coding practice. In this age of cloud computer, AI, VR/AR, multi-threading parallel processing is the way to go and the future of all software.

Because it's marketed as an enthusiast gaming product that is inferior to the competition for gaming.

Sadly, I don't play multiplayer games so I'll leave the discussion of losing because of lack of skill and internet speed up to your clearly superior experience of the above.

Ryzen unfortunately won't be getting any faster until major fabrication improvements are achieved as the chips are already running close to the limits of their current capabilities.

The crazy thing about ryzen is that it's totally feasible. Scaling off the 65W TDP of a 1700, a die with 2 cores fused off should just squeeze into the 45W TDP that the 4 core intel chips in all the top end gaming laptops use (maybe with the clocks turned down a little). So with a single mask they're matching or beating intel in every benchmark, from 8-core prosumer down to laptops

We've already seen tests that show Ryzen's power consumption scaling with different core loadings:

http://www.anandtech.com/show/11244/...eads-vs-four/2

With XFR enabled just two cores will eat over 50w and with it disabled three cores will still consume over 50w. Any configuration with 4+ cores (i.e. >1 active CCX) will eat >60w and even the 4-core 1400X eats close to 60w as it's a 2+2 configuration.

There's a very good reason mobile and low-end APUs are planned last in AMD's release cycle, i.e. 6+ months away, the design's going to need a lot of refinement to be competitive in those power envelopes.

[Xlucine]

We've already seen tests that show Ryzen's power consumption scaling with different core loadings:

http://www.anandtech.com/show/11244/...eads-vs-four/2

With XFR enabled just two cores will eat over 50w and with it disabled three cores will still consume over 50w. Any configuration with 4+ cores (i.e. >1 active CCX) will eat >60w and even the 4-core 1400X eats close to 60w as it's a 2+2 configuration.

There's a very good reason mobile and low-end APUs are planned last in AMD's release cycle, i.e. 6+ months away, the design's going to need a lot of refinement to be competitive in those power envelopes.

Rubbish. Those all boost to ~4 GHz, whereas even intel's top end laptop chips only boost to 3.5-3.8 GHz, with a base of 2.6-2.8 GHz. Running the CPU slower will save energy, especially when you aren't trying to hit 3.5GHz on all cores

[qasdfdsaq]

Rubbish. Those all boost to ~4 GHz, whereas even intel's top end laptop chips only boost to 3.5-3.8 GHz, with a base of 2.6-2.8 GHz. Running the CPU slower will save energy, especially when you aren't trying to hit 3.5GHz on all cores

Clearly you haven't even bothered reading the article - or tests of Intel's laptop chips you refer to...

At minimum base speed with no boost, cores use 8-9w each but the CCX "uncore" uses 10-15w of overhead and the package another 10-15w on top of that. Good luck fitting six cores into a 45w TDP when you have nearly 30w eaten up with overhead already just to keep the chip running.

"Fusing off" or disabling two cores on an 1800x drops power consumption by 5w or ~6% from 92w to 87w so how you get the impression fusing off two cores on lower clocked a 1700 will drop power consumption by over 20w or 30% is beyond me.

[Xlucine]

Clearly you haven't even bothered reading the article - or tests of Intel's laptop chips you refer to...

At minimum base speed with no boost, cores use 8-9w each but the CCX "uncore" uses 10-15w of overhead and the package another 10-15w on top of that. Good luck fitting six cores into a 45w TDP when you have nearly 30w eaten up with overhead already just to keep the chip running.

"Fusing off" or disabling two cores on an 1800x drops power consumption by 5w or ~6% from 92w to 87w so how you get the impression fusing off two cores on lower clocked a 1700 will drop power consumption by over 20w or 30% is beyond me.

30W of overhead, where?

[kompukare]

Clearly you haven't even bothered reading the article - or tests of Intel's laptop chips you refer to...

At minimum base speed with no boost, cores use 8-9w each but the CCX "uncore" uses 10-15w of overhead and the package another 10-15w on top of that. Good luck fitting six cores into a 45w TDP when you have nearly 30w eaten up with overhead already just to keep the chip running.

"Fusing off" or disabling two cores on an 1800x drops power consumption by 5w or ~6% from 92w to 87w so how you get the impression fusing off two cores on lower clocked a 1700 will drop power consumption by over 20w or 30% is beyond me.

That is not how power scaling works. Not even close. The relationship between clock and voltages are well understood. Have a look at The Stilt's Strictly Technical thread over at AT:
https://forums.anandtech.com/threads...nical.2500572/

Basically, if you look at the Vmin chart, you'll see that 3.3GHz is already the first Critical (point past which the process losses it perf/watt sweet point and the overhead you talk about will also scale with clocks. Your overhead figures are also very high. By package do you mean the power lost by mobo's VRMs? Ryzen is a SOC but 20W-30W for the built-in chipset sounds crazy and I haven't seen anything like that. System idle has generally been about 40-45W which is about the same as Kabylake. Mobo choice can make a big difference there.

But if they target a more reasonable clock (2.5GHz to 2.8GHz with a reasonable single core turbo on top of that), it should be easily possible to match the mobile quad core i7 (the 45W parts). The cTPD chart from that thread is very telling.


With a cTDP of 45W, it still scores just under 1200 in Cinebench15 versus just over 1400 it scored at its full TPD. And that's for the 8C/16T chip. Mobile is likely to be all 4C/8T at least initially, so a 45W chip should get pretty close to the Ryzen 5 desktop performance. And unlike Ryzen 5, mobile will get its own mask. Obviously, graphics get added back to those wattage figures, but then that will probably mean throttling when both are fully loaded. But that's what everyone does.

No, the reason Ryzen is mobile last is because all the required parts (Zen, Vega, BIOS memory compatibility, and perfecting cTDP) were not ready and desktop non-APU is easier.

[preter_s]

I dunno. Being able to hit 4.5Ghz+ would have given them the crown of the fastest CPU bar none, whereas now they still lose to Intel for single-threaded perf.

The choice you claim they made might give them more efficient high-volume mainstream parts but leaves them without a market leading halo product.

The server market doesn't care about halo products. It's all about perf/watt. The server market should be WAY more lucrative for AMD right now than the consumer market, so they're absolutely spot on to target perf/watt at lower clocks if they can only do one silicon run. Having a consumer part @ 4.5Ghz stock at the cost of only being able to do ~ 2GHz @ 65W would not be worth it.

So why server first? Intel's current range topping server CPU offers 24C/48T @ 2.4GHz in a 165W thermal envelope. I reckon (based on the chart kompukare shared above) that AMD's naples processor will offer 32C/64T - i.e. 50% more parallel throughput - at the same clock speed and thermal envelope. That Intel chip? Suggested customer price is just under $9000. That's why server first.

Exactly, those same kids who want to brag about the highest over-clocking number on their rig built with their parents' money... no doubt wouldn't be able to comprehend what the power bill means, unlike the real commercial world.

Look at these idiot kids are still talking about "single-threaded performance" for "bragging rights". Sheesh - the whole world is moving towards multi-threaded software and hardware you moron! Ryzen will give you years of future proofing. Your Ryzen build will get faster and faster over the next 3-5 years you fool!