AMD Ryzen (Summit Ridge) Benchmarks Thread (use new thread)

[krumme]

We dont know efficiency of zen just a rough estimate of it but surely from b2b to b2c all buying it will only care for total efficiency. In context of efficiency: Who cares if vrm is integrated or not and the same for more of the functionality? Zen is weighted and bought on serverside solely on TCO so it better deliver here.
And as a desktop user i couldnt care less if the cpu was efficient but demanded a system be it memory or mb that was inefficient. The same goes for cost and performance. Thats how i value my car and my wife.

 

[AtenRa]

Well sorta. If the CPU is sucking up more power, than its likely harder to cool. A system where the CPU uses less power is more likely to OC better.

what all this has to do with one system being more efficient than the other ??
I dont care if CPU A uses more power than CPU B if the entire system A draws less energy at the wall than System B.

 

[moonbogg]

Does anyone else feel their brain ripping in half at the thought of having to wait MONTHS for this thing to be reviewed?

 

[coercitiv]

Everyone seems to be missing his point, unless its me who is missing something. I understand that lower power at both idle and load is, of course, better. But we are looking at the CPU, and the CPU's both consume essentially zero watts at idle. The power consumption figures at idle reflect total system power consumption for both systems, of which the CPU's contribute nothing to. There-to-the-fore, the Intel system's power consumption increased less upon loading than the AMD system, so the Intel CPU itself is more power efficient than the AMD CPU.

Idle power usage for desktop CPUs is not near zero, and heavily depends on idle clocks (and temps). Your explanation holds true for mobile CPUs, where a different process and optimized power management will keep a quad core CPU package under 1W @ 800Mhz or bellow 2W at base clocks. (my 4700HQ reports 0.8W when completely idle)

In case you're curious to see how idle power usage for 6600K looks like at different idle clocks, here's one more pic bellow. (package power is sensor based, kill-a-watt power is entire system at wall)

PS: however, the entire debate seems rather pointless to me. Any number of factors, from hardware specifics to software settings can have drastic effect on idle power usage.

 

[krumme]

Does anyone else feel their brain ripping in half at the thought of having to wait MONTHS for this thing to be reviewed?

This. We will hire a certain german to break into the Dresden facilities.

 

[Abwx]

But we are looking at the CPU, and the CPU's both consume essentially zero watts at idle.
Make sense?

21W on the CPU dedicated 12V rail for the 6900K, at idle of course..

 

[krumme]

Does anyone else feel their brain ripping in half at the thought of having to wait MONTHS for this thing to be reviewed?

AMD hurts your feels.
Demand a discount.

 

[realibrad]

what all this has to do with one system being more efficient than the other ??
I dont care if CPU A uses more power than CPU B if the entire system A draws less energy at the wall than System B.

Again, it has to do with OC. If a CPU is more efficient, its generally true that you will be able to OC more*. That is why I said sorta true. Most people do not OC so in that case system usage is what is important. For someone that OCs then more efficient components allow them more thermal headroom.

 

[moonbogg]

Ah, so the CPU doesn't use zero power at idle. Well I'll be doggone darn damned.

 

[itsmydamnation]

I don't know. My 2009 4 core/thread i5 750 at 3.8 Ghz beat someone's 4 GHz 8 core/thread 8350 with the 100 samples test.

That suggests to me that either his CPU is throttling or the test is mostly of FPU performance.

update:

Someone else's posted time for 200 samples on a 4.6 GHz FX is basically identical to my Lynnfield's performance at 3.8. This seems to confirm my belief that this test is not taking advantage of 8 integer cores.

Blender is a renderer, it mainly deals with xyza kind of stuff so really its a SIMD throughput /cache/memory test.

Also They likely don't want to do single thread until they 100% dial down boost clock speed.

 

[zinfamous]

I'm just waiting for Donald Trump to claim that he is responsible for AMD's remarkable comeback.

 

[ElFenix]

Ah, so the CPU doesn't use zero power at idle. Well I'll be doggone darn damned.

well, yeah, it can't. even when the lights are out for many of the execution units there's got to be a doorman waiting to see if anything happens. and even lights out is more of a dim and buzzing state, because there's leakage as well. i assume large parts of the uncore can't be turned off, either.

 

[sirmo]

Your computer never stops executing code either.. all those background processes, checking various statuses and being triggered by interrupts. It just happens in small bursts.

 

[zinfamous]

Your computer never stops executing code either.. all those background processes, checking various statuses and being triggered by interrupts. It just happens in small bursts.

Maybe our CPUs could be more efficient if we placed our towers in designated safespaces?

 

[BeepBeep2]

Hello everyone, first post. Some of you may know me from XtremeSystems, or HWBOT.
I think the bickering about power delta and performance per watt is happening a little early, and you all seem very heated on the subject.

Typical CPU package power is most certainly not 0w: 2-15w are common depending on P-State, slight load variance (a background process could cause the CPU to switch in and out of its lowest P-State), etc. Your laptop can most certainly run all day when the CPU is only sipping 3w, but when it is eating 30w, not so much.

As far as the talk about idle-load deltas and platform power, the platform is what matters most. Not to say the individual CPU power consumption isn't important, but total platform power is what matters, not only in HEDT / desktop, but more importantly laptop / mobile. The goal of reducing total platform power at both idle and load is the reason we see more and more functions that were previously handled by northbridge / southbridge being integrated into the CPU die in the first place.

I think the most important thing to remember here is that AMD was so far behind, even with XV, that even getting close to Bdw-E is quite an achievement. To those speculating 4.2+ turbo speeds, I don't think AMD is quite there yet. I'm guessing 3.7 to 4.0 turbo, personally.

So TL;DR to remember relating to power consumption as we head toward launch:
Idle CPUs more than zero watts, but this number can still be low, in the 2-5w range. Mobile CPUs usually less due to more aggressive clock / voltage reduction
TDP is thermal design power or thermal design point - NOT indicative of power consumption, though we often relate it to power consumption
Total platform power is key for end users, though it is neat to see CPU only power consumption
Ryzen is likely to be priced cheaper than Intel's HEDT solutions
GloFo's 14nm process is likely not as good as Intel's 14nm or 14nm+, I suspect gate pitch, fin pitch, and metal pitch all to be 10-15% higher than intel's. For intel's 14nm+ there are many news sites that claimed the gate pitch was actually wider, but no source was listed

 

[iBoMbY]

The official 2.78a build AMD used in the demo still seem to lack the new SIMD optimizations. Pulled the newest version from Git and built it with MSVC 2015. 44.4% higher performance than on the official build, on Haswell-EP HCC.

In case anyone wants to try it out: https://1drv.ms/u/s!Ag6oE4SOsCmDhFAm03vWlB3s_qeD (40.5MB)
Password: "ryzen" (without the quotes)

EDIT: According to user reports, > 63% performance improvement over the official 2.78a build on Skylake.

i7-3770k @ 4.5 GHz|2x 8GB DDR3-2133 (KHX2133C11D3/8GX)|Win10 1511 x64

blender-2.78a-windows64:

100 samples: 45.08s
200 samples: 90.37s

Blender_278_1412GIT_AVX2_MSVC:

200 samples: 67.51s

 

[CHADBOGA]

Does anyone else feel their brain ripping in half at the thought of having to wait MONTHS for this thing to be reviewed?

Yep, it would be great to have a series of full reviews over the Christmas/New Year period to read whilst chilling out.

 

[DrMrLordX]

Yes, we'll slow time inside the package.

Hot damn! That'll really improve race-to-idle.

So AMD probably attached a 50W light bulb to the Broadwell system to get similar idle powers so they wouldn't get humiliated?

Of course, that's the only logical conclusion here . . .

What IQ do you have or what is your education?

2 and uh . . . hard to focus . . . zombie intelligence too low . . . urrr braaiiiinnss

I'ma leave you with this quote from someone with more intelligence than this whole forum combined.

So did you just insult the entire forum population or what?

His original point was that the delta told a story of a more efficient chip for Intel, and that is not inherently true. AMD could have a very efficient chip, and a horribly inefficient chipset. It could also be true that AMD could be using more power on their CPU and the chipset is super efficient. Until we get more information, then we will just be guessing.

Also VRM efficiency might be a factor.

Does anyone else feel their brain ripping in half at the thought of having to wait MONTHS for this thing to be reviewed?

My brain is probably already ripped in half, so . . . no! But aren't the top-end Summit Ridge chips launching in January? That's when I expect the first reviews. Gonna be around 1 month.

 

[qookap]

What IQ do you have or what is your education?

The power that is measured is not the Broadwell-E chip and it is not the Zen chip. It is the whole system power consumption. So concluding one chip is more efficienct than the other is a non sequitur. (If you look at my post, I did not claim one was superior than the other.)

It's like if you were to put two different SSDs in each system, say a Samsung one and a Micron one, and concluding that one is more power efficient than the other based on those 4 power numbers.

If you want to get nitty gritty about how real power measurement is -- should be -- done, you might want to read up on it before you waste any more keystrokes on nonsense like that post.

http://www.anandtech.com/show/6529/busting-the-x86-power-myth-indepth-clover-trail-power-analysis





As one can see, platform power here is more than an order of magnitude more than CPU power.


Questioning someones IQ is an insult, and thats not allowed here.
Markfw
Anandtech Moderator

For engineering you are right but here is a problem: precise words

I'll say this

Zen platform idle: 93W
Intel platform idle: 106W

Zen platform load: 187W
Intel platform load: 191W

Zen delta: 94W (RyZen)
Intel delta: 85W (Broaderwell-EP?)

maybe you think this is common sense..but remember this, there is not everyone service on EE.
so IQ is the point..I'm glad mine good. if everyone had high IQ. we will lose the job. : )

as customer with a view angle from marking, both product had so close performance.
the final decision should be price. if RyZen 8C16T can make a huge performance improvement or nice price. it's will success.

8 core product too expensive especially Intel desktop.

 

[Maverick177]

For engineering you are right but here is a problem: precise words

I'll say this

Zen platform idle: 93W
Intel platform idle: 106W

Zen platform load: 187W
Intel platform load: 191W

Zen delta: 94W (RyZen)
Intel delta: 85W (Broaderwell-EP?)

maybe you think this is common sense..but remember this, there is not everyone service on EE.
so IQ is the point..I'm glad mine good. if everyone had high IQ. we will lose the job. : )

as customer with a view angle from marking, both product had so close performance.
the final decision should be price. if RyZen 8C16T can make a huge performance improvement or nice price. it's will success.

8 core product too expensive especially Intel desktop.

I don't know the ball park of my IQ but I'm pretty sure Zen came out on top in terms of power consumption here.

 

[PhonakV30]

https://www.reddit.com/r/Amd/commen...amd_robert_how_can_we_do_the_blender/db7q21q/

Set render samples to 150. A new file with the sample level set correctly will be uploaded shortly. Apologies for the confusion.

Blender 2.78a x64 is what we used, binary download from blender.org.

 

[bjt2]

Hello everyone, first post. Some of you may know me from XtremeSystems, or HWBOT.
I think the bickering about power delta and performance per watt is happening a little early, and you all seem very heated on the subject.

Typical CPU package power is most certainly not 0w: 2-15w are common depending on P-State, slight load variance (a background process could cause the CPU to switch in and out of its lowest P-State), etc. Your laptop can most certainly run all day when the CPU is only sipping 3w, but when it is eating 30w, not so much.

As far as the talk about idle-load deltas and platform power, the platform is what matters most. Not to say the individual CPU power consumption isn't important, but total platform power is what matters, not only in HEDT / desktop, but more importantly laptop / mobile. The goal of reducing total platform power at both idle and load is the reason we see more and more functions that were previously handled by northbridge / southbridge being integrated into the CPU die in the first place.

I think the most important thing to remember here is that AMD was so far behind, even with XV, that even getting close to Bdw-E is quite an achievement. To those speculating 4.2+ turbo speeds, I don't think AMD is quite there yet. I'm guessing 3.7 to 4.0 turbo, personally.

So TL;DR to remember relating to power consumption as we head toward launch:
Idle CPUs more than zero watts, but this number can still be low, in the 2-5w range. Mobile CPUs usually less due to more aggressive clock / voltage reduction
TDP is thermal design power or thermal design point - NOT indicative of power consumption, though we often relate it to power consumption
Total platform power is key for end users, though it is neat to see CPU only power consumption
Ryzen is likely to be priced cheaper than Intel's HEDT solutions
GloFo's 14nm process is likely not as good as Intel's 14nm or 14nm+, I suspect gate pitch, fin pitch, and metal pitch all to be 10-15% higher than intel's. For intel's 14nm+ there are many news sites that claimed the gate pitch was actually wider, but no source was listed

Even if the base clock will be 3.4GHz, how can you say turbo clocks is only 3.7-4GHz?
Turbo core max is when only one core is active.
On an 8 core CPU, that is about 1/6-1/7 of the total power.
With all this power given to just one core, only 300-600Mhz of turbo?
FX 8370E (95W) managed 3.3/4.3, 1GHz of turbo.
If the turbo is so low, then this would mean that we are near the top of the process, area of diminished return.
How come a so big chip draw only 95W if it is near the top power limit?
How AMD think to further raise base clock if they are already near the limit?
How they think to have high yield if they are near the top limit?
How they think to not go in thermal runaway, and so have thermal instability if they are near the top of the process? If you are at the limit, any tiny perturbation might crash the CPU. Think of highly overclocked CPU.

Conclusion: i don't think we are near the top limit and with 6-7x the power for one core, it will clock at least at the 4.3GHz of FX8370E... Probabily at 3.5-3.6 base and 4.5-4.6 turbo

 

[BeepBeep2]

Even if the base clock will be 3.4GHz, how can you say turbo clocks is only 3.7-4GHz?
Turbo core max is when only one core is active.
On an 8 core CPU, that is about 1/6-1/7 of the total power.
With all this power given to just one core, only 300-600Mhz of turbo?
FX 8370E (95W) managed 3.3/4.3, 1GHz of turbo.
If the turbo is so low, then this would mean that we are near the top of the process, area of diminished return.
How come a so big chip draw only 95W if it is near the top power limit?
How AMD think to further raise base clock if they are already near the limit?
How they think to have high yield if they are near the top limit?
How they think to not go in thermal runaway, and so have thermal instability if they are near the top of the process? If you are at the limit, any tiny perturbation might crash the CPU. Think of highly overclocked CPU.

Conclusion: i don't think we are near the top limit and with 6-7x the power for one core, it will clock at least at the 4.3GHz of FX8370E... Probabily at 3.5-3.6 base and 4.5-4.6 turbo

Easy answers:
Zen is a completely different architecture compared to Piledriver, and I don't think it is capable of as high clocks. Comparing these two in terms of achievable clock speed is not rational.
8370E is a special case, low leakage bin with architecture designed for high clocks, 4.3 is not radical for that architecture and they could afford to do that in a 95w envelope on one core.
TDP ≠ Power Consumption.
High yield near limit? I dunno, take a look at K8 @ 90nm, 6000+ and 6400+ were already near the limit, and only F3 stepping could get that high reliably. Sold many.
K10 @ 45nm ("K10.5") - 1100T had a (useless, never working) turbo up to 3.7 GHz and that was the plausible limit considering many were not stable past 4 GHz without exotic cooling.
Look at Kaby Lake, 7700K only has 300 MHz turbo, and needs large voltage increase above 4.8 GHz.

I don't think it's safe to assume that Ryzen CPUs will have above a 4 GHz boost clock.

 

[Abwx]

Turbo core max is when only one core is active.
On an 8 core CPU, that is about 1/6-1/7 of the total power.
With all this power given to just one core, only 300-600Mhz of turbo?
FX 8370E (95W) managed 3.3/4.3, 1GHz of turbo.

In ST the power budget is in the 30-50% of the MT power budget, chips like the 8370E are close to the upper
percentage because they have low base frequency and quite high turbo, besides it consume only 65W in MT, so it cant be used as comparison with Zen, that being said this doesnt invalidate your point about Zen ST frequency, if it does 3.4 in MT there s no way that it would be limited to 3.7 in ST.

 

[bjt2]

Easy answers:
Zen is a completely different architecture compared to Piledriver, and I don't think it is capable of as high clocks. Comparing these two in terms of achievable clock speed is not rational.
8370E is a special case, low leakage bin with architecture designed for high clocks, 4.3 is not radical for that architecture and they could afford to do that in a 95w envelope on one core.
TDP ≠ Power Consumption.
High yield near limit? I dunno, take a look at K8 @ 90nm, 6000+ and 6400+ were already near the limit, and only F3 stepping could get that high reliably. Sold many.
K10 @ 45nm ("K10.5") - 1100T had a (useless, never working) turbo up to 3.7 GHz and that was the plausible limit considering many were not stable past 4 GHz without exotic cooling.
Look at Kaby Lake, 7700K only has 300 MHz turbo, and needs large voltage increase above 4.8 GHz.

I don't think it's safe to assume that Ryzen CPUs will have above a 4 GHz boost clock.

8370E was on 32nm. Three nodes away from Zen. And it seems that zen will have FO4 similar to BD. Anyway lower than core, because zen have 19 stage integer pipeline, versus 14-17 (not disclosed, these are the estimations) of core architecture. And Excavator with 28nm bulk HDL managed to have 4.3GHz turbo too. With two node and similar FO4 they did not managed to increase max clock turbo? With GPUs that gained 20% max clock? Surely I don't expect 20% gains at so high clocks, but certainly i don't expect a regression from 4.3GHz...