Official AMD Ryzen Benchmarks, Reviews, Prices, and Discussion

[Tup3x]

CMK16GX4M2B3000C15 is
https://www.newegg.com/Product/Product.aspx?Item=N82E16820233852

It doesn't meant that it's running at that speed though.

 

[GroundZero7]

45ns is latency on Intel CPUs. Having 25ns on top of that? Please.

You don't know what you are talking about. My Phenom 2 gets 45ns. Bulldozer/??? gets the same.

 

[french toast]

Stuff i work with is generally unfriendly to any cache structure and consists of a load of random accesses.

So, for my workloads it probably does. I just hope AMD was aware of it ahead of time, so that stuff is fixed proper by now (in Zen 2, that should soon tape out, shouldn't it?). After all, that's about the only glaring flaw with Ryzen right now. Rest are either imperfections of deliberate choices.


45ns is latency on Intel CPUs. Having 25ns on top of that? Please.


Yep, cache should carry it for now.

Yea possibly for corner case workloads, but just remember your getting 50-100% more cores or half the price for that, i know we are looking at latency here and not value for money but it has to be taken into account, if you were choosing between 6900k or 1800x for those corner case workloads you speak of, 10-30% latency perf penalty BUT half the price, still looks a bargain. That is in corner case workloads that i think is not going to matter for most.
Besides its not horrendous, Ryzen has a solid cache subsystem as you allude to so any impact will be mitigated, some teething issues are to be expected on a brand new platform.

 

[GroundZero7]

It doesn't meant that it's running at that speed though.

It says the speed it's running at at the very bottom. 2133 should be running at CL13 or less.

 

[Tup3x]

It says the speed it's running at at the very bottom. 2133 should be running at CL13 or less.

Why do you assume CL13? Often if you do not use XMP profile the RAM defaults to really low speed and high latencies.

 

[lightmanek]

Where do you get 3000MHz memory speed from if SiSoft is reporting 2133MHz CL16?

What I see is:
AMD FCH SATA Controller [AHCI mode]; 16GB Corsair CMK16GX4M2B3000C15 DIMM (2.13GHz) PC24000 (16-16-17-39 4-55-18-6)

and

Aggregated Results

Component Average Score Average Capacity Average Speed State Platform Operating System Popularity

AMD FCH; 16GB Corsair CMK16GX4M2B3000C15 DIMM PC24000 69.5ns 16.00GB 2133MHz Normal Desktop Windows x64 10.0.6 2

I've been using SiSoft Sandra since K6-2 and Pentium III days, and it almost always displayed correct memory clocks. I see this test was ran at 2133MHz CL16 16 17 in DC mode.

 

[GroundZero7]

Why do you assume CL13? Often if you do not use XMP profile the RAM defaults to really low speed and high latencies.

Because almost all newer 2133 can hit CL13. You adjust it in the bios. 13-14-13-41 1t.

 

[lolfail9001]

You don't know what you are talking about. My Phenom 2 gets 45ns. Bulldozer/??? gets the same.

I know, what i am talking about. AIDA64 reports latency for the whole package: memory and IMC. So does Sandra, Passmark and literally every single memory latency test you can conduct, because that's just how it is.

Because almost all newer 2133 can hit CL13. You adjust it in the bios. 13-14-13-41 1t.

My 2133 works at 10ish timings (screw Intel), it does not mean the memory in this test does.

 

[GroundZero7]

I know, what i am talking about. AIDA64 reports latency for the whole package: memory and IMC. So does Sandra, Passmark and literally every single memory latency test you can conduct, because that's just how it is.

My 2133 works at 10ish timings (screw Intel), it does not mean the memory in this test does.

And What does that have to do with what I said? Intel and AMD both have similar latency with tuned ram. Pop in XFR, or AMP and the results are about the same.

I was being conservative. How the H do I know what Ram you have? Why would it matter? I hit 7-7-7-21 on first gen patriot DDR3. There isn't a huge amount of deviation in performance ram. Lower bins are relegated to the bottom of the barrel.

My point is either that ram is complete garbage, that bios still needs tweaking, or it was deliberately underclocked. We know it isn't Ryzen because there are motherboards that work just fine with DDR4 3600

 

[Glo.]

Are you saying we'll have 15W 4c8t 3GHz APUs with 512SPs at 1.2GHz? At full load? Not power throttling? Don't be daft. At 45W that might be possible, but the GPU will throttle no matter what. Remember, the 896SP 460 is 75W alone.

There is already floating around 4C/8T APU engineering sample with 11 CUs in Mobile package that has 35W TDP, and 3.0/3.3 GHz. 11 CU design is cut down from 12 CPU.

 

[lobz]

You should think full well before re-posting some kiddy fanatics conspiracy fairy tale on here. That post and poster are plain sad.

Anyone who works in the IT field as an Architect/CTO/Director knows fully well that story is complete and utter BS, every letter of it. AKA smear campaign.

These threads have for the past week become full of random trash, little substance. No one in the decision making IT world even follows Ryzen right now.

In a mass virtualized/cloud orientated IT world, they will discuss Ryzen only once Naples is out.

Even then it will be very few, as that department is in every major company, outscourced.

There is no Evil and Good here. it is just business vs business.

And these practices are NEVER MFG<->Business. It is MFG<->OEM/ODMs.

Yes,you're probably right. I personally know for a fact, that nobody in IT decision making has ever planned ahead but rather well after market changes happen, and noone in decision making were ever following the technical and practical advancements of theirbrespective market fields. Also none of them ever wondered if they could make their purchases and departments financially more effective with simple actions such as purchasing products with better price/performance ratio then before in large quantities.
I'm absolutely sure about your statement being fully correct. Nobody, I repat, NOBODY in decisiong making even discusses Ryzen until Naples is out... Why would they? Who cares? They certainly won't lose their jobs if other companies win a lot of money by lowering their costs, except then... The argument 'Naples wasn't even out' will sound totally convincing in front of the board of investors.

 

[R0H1T]

What frequency are those 1024SP@50/75W? because i was talking of 512SP at a base frequency of 900-1000MHz on the 14nmFF (versus 512SP@800MHz on the 28nm BULK)

Boost clock of ~1.17GHz for 50W TDP & 75W would be same as RX 460 btw RR will have Polaris cores or Vega?

 

[Glo.]

RR will have Polaris cores or Vega?

Vega.

 

[utmode]

This last bit may explain why Intel PR sent out a last-minute “call us before you write” email to most of the press,
^^ from semiaccurate?
My question what Intel mean it by writing Call us before you write.
link: http://semiaccurate.com/2017/02/22/amds-ryzen-7-1800x-beats-intels-i7-6900k-half-price/

 

[Valantar]

And the full P11 is 1024 SP @75W TDP, we also have the highly binned P11 for pros with a 50 W TDP ~
http://www.anandtech.com/show/10821/now-shipping-amd-radeon-pro-wx-series

I'm not saying that a 15W APU will throttle (or not) but you're basing your assumptions on a highly leaky, locked GPU.

Sure. And the GPU itself is far from the whole thermal load of a dGPU. Let's assume that the GPU itself represents 50W (66%) of the 75W TDP. That means a 512SP version at full tilt, no RAM or anything included, is 25W. Alone.

I calculated that many weeks ago and maybe i remember bad, but 15W excavator has 2.7GHz base clock for 4 core (2m) and 1100Mhz of (max?) frequency for the 512SPs. Given that at low power/Vcore the process allow 65% power saving or +80% clock, I calculated that if the SPs will remain 512 a +100MHz will give anyway some power to the CPU. 2.7GHz Zen 4C8T will draw just a little less than 2.7GHz XV 2M4C due to power scaling.
Giving the increased power budget on the CPU, we can reach 3GHz.
Or alternatively, still 2.7GHz Zen cores, but with 1024SPs at 900MHz-1GHz max, but with higher area consumption...

EDIT: to be clear, on the CPU the clock is base clock. I don't know/remember turbo clocks on the BR APU. For the GPU part it's the MAXIMUM clock. Base clock, if i remember well, is 800MHz... On the 28nm BULK.

Base clock numbers go out the window once the iGPU gets any sort of load, remember that. With Intel as an example, their 15W >2.5GHz base clock chips often dip well below 1.5GHz when the iGPU kicks in and that's with a far smaller GPU. My desktop A8-7600 (65W) power throttles when the GPU (384SP) is under load, although not by that much (from 3.1GHz down to ~2.4-2.6). APUs are optimized to utilize their thermal window to the fullest extent possible. As such, it would be silly for the CPU not to have a higher base clock when the GPU is idle, as they have cooling to spare. I'd love to see actual clock speeds for a Carrizo chip with the GPU under load. They'd definitely not be 2.7GHz.

I'm not saying Ryzen/Raven Ridge won't give us some awesome APUs. I'm simply saying this: Don't expect miracles. 15W will not be the sweet spot for full-APU load, not by a long shot. It will no doubt be more than enough for a decent CPU. But with a GPU, even 25W will be a massive improvement - either throttling the CPU less, or allowing it to run full-tilt with 10W to spare for the GPU alone. 35W or more would allow this to really shine. GPUs are power hungry. That's just the way it is.

What frequency are those 1024SP@50/75W? because i was talking of 512SP at a base frequency of 900-1000MHz on the 14nmFF (versus 512SP@800MHz on the 28nm BULK)

No higher than ~1200MHz, at least. And sure, cutting back a few hundred MHz will save you power. But there's also data suggesting that Polaris power draw flattens out below ~900MHz. So there's not much to save. I hope Raven Ridge has Vega-based iGPUs, but I'm not betting on it. And even so, you'd need 15-20W for a 512SP GPU alone for it to get up to speed. Fitting it inside a 15W TDP APU and not expecting throttling is a pipe dream.

 

[Valantar]

There is already floating around 4C/8T APU engineering sample with 11 CUs in Mobile package that has 35W TDP, and 3.0/3.3 GHz. 11 CU design is cut down from 12 CPU.

That doesn't answer the question of CPU clocks under GPU load. Again: any GPU load means CPU base clocks go out the window. Any at all. If it's a 35W 4c8t chip with a 3GHz base and 11CUs (704SP), I'd expect it to allocate anywhere between 50 and 75% of its power to the GPU under heavy loads - which would mean significant CPU throttling. I'd be very impressed if it maintained 2GHz on all cores.

 

[itsmydamnation]

Which also shouldn't be giving exponential scaling, unless they went and loosened the timings to go along with the lower clockspeeds.

Well this is what i get on ivb with DDR3 in Dragon Age inc . looks like you can hit latency points where things blow out

 

[raghu78]

https://www.scan.co.uk/products/3xs...370-crosshair-16gb-corsair-ddr4-corsair-h100i

If retailers are selling 1800X overclocked to 4.2 Ghz I think there might be another 200-300 Mhz headroom left. I doubt anybody would sell a OC CPU at its highest limits. You don't want to sell CPU running at its silicon limits for the purposes of longevity and warranty.

4 more days to go. Hurray !!!

imo if Raven Ridge comes with anything more than 512 sp (8 Vega NCU) AMD better equip it with 2 GB High bandwidth cache. A 2 Hi stack should be easy to manufacture at good yields and low cost. More importantly a RR APU at 95w with 3.4/3.8 Ghz 4C/8T and 1024 sp can sell for atleast USD 300 if AMD's Vega GPU is allowed to perform to its full potential. I think AMD will get very close to Nvidia in terms of perf/sp wrt perf/cc with the new NCU and Vega architecture. I would not be surprised to see a 1024 sp Vega GPU perform very close to a Rx 470 or even better due to a combination of higher clocks, higher IPC and significant architectural improvements.

 

[inf64]

Yep, we are discussing the scan link in the Ryzen OC thread. Looks really good so far.

 

[lolfail9001]

I doubt anybody would sell a OC CPU at its highest limits.

They sell Intel CPUs at their highest pre-silicon lottery limit. Consider 4.2Ghz precisely that for Ryzen.

 

[raghu78]

We will see. XFR clocking the 1800x to 4.1 Ghz and retailers selling 1800x pre overclocked to 4.2 Ghz gives me some hope that there might be another 200-300 Mhz headroom. It shall be known in 4 days what OC headroom does SR have.

 

[lolfail9001]

It shall be known in 4 days what OC headroom does SR have.

And in 2 months what degradation-free OC headroom does SR have.

 

[A///]

And in 2 months what degradation-free OC headroom does SR have.

What does that mean? Loss of possible OC room due to pushing a processor at near max?

 

[raghu78]

And in 2 months what degradation-free OC headroom does SR have.

dude you are telling as if electromigration is a problem which only AMD CPUs face. Maybe possible in your biased mind. But the fact is its universal and Intel CPUs face the same too. But by the time your clocks are affected due to such effects its a long time and these enthusiasts who purchase such high end CPUs have moved on to the next latest shiny CPU / rig.

 

[bjt2]

No higher than ~1200MHz, at least. And sure, cutting back a few hundred MHz will save you power. But there's also data suggesting that Polaris power draw flattens out below ~900MHz. So there's not much to save. I hope Raven Ridge has Vega-based iGPUs, but I'm not betting on it. And even so, you'd need 15-20W for a 512SP GPU alone for it to get up to speed. Fitting it inside a 15W TDP APU and not expecting throttling is a pipe dream.

FX 9800P on 28nm BULK, with 15W TDP/12W CDP, has: 4 XV cores with 2.7 base 3.6 turbo (probabily you are right on CPU throttling under base on high GPU load) plus 512 SP at 758MHz+
On the 14nm up to +80% clock is expected around 0.9V at same TDP. So 512SP at same power should gain at least 300MHz. Let's say that we use 900MHz 512SP. We can expect that some W can be allocated to the CPU, so to have 2.7-3GHz base 4c8t Zen...