AMD's Ryzen is here!


#103

@pauliunas i think i found our miscommunication, i was talking about broadcasting, which i also referred to as streaming, which may misled you.


#104

At the end of the day there’s one thing “enthusiasts” tend to forget:.

You’re the 1%, the general populace doesn’t give a hoot whether the other CPU is 5% faster or slower, they care about whether it’s cheaper or not and Ryzen delivers on that sector currently.


#105

The point is that it’s slower. Cheaper, but slower. You can buy a cheaper Intel processor too, it will also be slower. The real question is which one offers better bang for the buck.


#107

Or better just post something on reddit or some forum and then wait for clickbait sites to pick on it :smiley: Maybe that way it would look more like a real leak so they would fall for it easier. And of course we need to delete these comments before launching our happy hype train :smile:


#108

I think we should leave promotion in the hands of Underdog for the time being, as Eve is specifically paying them to reach the desired audience and they are specialized in this kind of thing.

The absolute worst thing either Eve or the community can do, is to purposely spread misinformation. Such as making anything look better than it really is.


#109

How could we even make it look better :wink:


#110

Now that R5 is out, the CPU scene finally gets interesting. Id say in the $150-$200 segment, now AMD has the edge.


#111

Oh boy - here we go I waited for this post to be made :slight_smile:

And honestly - I 100% agree that the R5 makes for a super interesting buy especially compared in it’s price-range. While the i5 is better in most games due to the higher single-core performance at this time, what I find more interesting is the shift from current gen-games to utilize all CPU cores more and more - which is why I think in some years the i5 might (finally) become a bottleneck for people who still use their (t)rusty old 2500K :slight_smile:

Especially the R5 1600(X) seems like a killer for me personally - 6C/12T and a solid single-core performance. :slight_smile:

However what’s also been pointed out on reddit recently is the fact that the NV driver seems to favor higher single-threaded performance (?) as again pointed out on reddit by some people, so that makes comparing numbers pretty hard interestingly -

Especially in RoTR this gets super interesting in some areas:



Those charts make it pretty hard in my eyes to compare sites from now on as it now matters in a CPU benchmark if they used an AMD/NVidia GPU.


#112

First of all, Id say Rocket League is an outlier. Not only its built on the aging DX9 platform, but the FPS is high enough that the difference doesn’t really matter anymore.

The driver situation is quite weird, on DX11 games, the opposite applies. Until DX12 matures and NVIDIA got some time to update their drivers, Id call this a tie.

Another issue to take into account is that fast RAM is pretty much required for a Ryzen system to be able to perform to its full potential, so the added cost should be taken into account. That’s simply not the case for Intel.

On the other hand, Ryzen gives you ‘free’ overclocking while Intel requires you to buy a more expensive CPU and mainboard to begin with. This is especially handy on the ~$200 price range, since youd have to buy the highest-end i5 to be able to OC, and the lower-end i5 doesn’t have that much advantage against an overclocked Ryzen.


#113

Anyone interested in an enlightening TDP versus performance benchmark that was conducted in the AnandTech forums? You might be very impressed by just how much better Ryzen performs at even lower TDPs compared to Intel.


#114

This AnandTech forum post has an excellent footnote section entitled simply “An easter egg” and nothing more. What it features are impressive charts showing how Ryzen’s Zeppelin cores are in their sweet spot when used at mobile power levels. Note that the best 45 W Kaby Lake Intel CPU gets only around 750 in Cinebench R15. See here: http://www.notebookcheck.net/Intel-Core-i7-7820HK-Laptop-Processor.189821.0.html

On the other hand, AMD’s Ryzen at 30 W gets around 850!

Low power is where Ryzen truly shines!

Source:


#115

Very interesting. I’ve been waiting a long time for AMD to get their game together regarding fast and light mobile processors. Although far from the efficiency we would need for a V-like device, this seems a step in the right direction for AMD. Inevitably, as ARM gets more powerful and x86 gets more power-efficient, we’ll find that sweet spot in the middle where either platform could take the cake for mobile devices like tablets and laptops, especially in light of the development of Windows 10 for ARM.


#116

It should be noted that that is an eight-core Ryzen CPU in that metric. Due to the minimal power requirements (physical size) to sustain that large a number of cores, that is why there is steeper and steeper tapering or fall-off in performance towards the far-left end of the plot. For reasons outside of the scope of this discussion and for the sake of brevity, if the core count is reduced, that performance fall-off will start to occur at much lower wattages. For example, with a quad-core configuration, the tapering or fall-off would instead begin to occur more down around 15W, not 30W. Additionally, with a dual-core configuration, the tapering or fall-off would instead begin to occur more down around 7-8W, not 30W. In other words, you decrease the minimal power requirements for optimal performance/watt as you decrease the number of cores.


#117

While the multi-core performance may look to be good, that might not the case with per-core performance, as even the highest-end desktop Ryzen 1800X cant keep up with the per-core performance of the 7Y75 that has only 5% the TDP.

As your second graph shows, the sweet spot for Ryzen seems to be 3.3 GHz or 3.5 GHz. Considering the gap in IPC, thats equivalent to 2.8 - 3.0 GHz on the Kaby Lake platform, which means it will perform around 20-30% slower in single-threaded performance compared to the 7Y75, which arguably still is the most important workload.

Id say its too early to call.

The reason is the mobile version will have quite a lot of changes to accomodate the iGPU. Not only the iGPU will need some power (even just to idle), but it also needs shared RAM bandwidth, which eats performance. Especially as you know, Ryzen is sensitive to RAM speed. RAM speed itself doesnt have that much freedom on mobile as we need to consider the RAM power consumption, unlike in desktop platform.

In addition to that, the slope gets very steep once you get into 30W, making the uncertainty too large to predict anything mobile.


#118

See my prior comment. With less and less cores, this slope tapering or fall-off would not occur until lower and lower wattage levels. This is due to the fact that the physical size of the silicon comes with a minimal power level required just to sustain the sheer number of transistors and this mostly irregardless of clock speed. Generally speaking, that fall-off corner is proportional to the number of cores, meaning that the fall-off corner does occur until a half of the power/wattage for a half of the cores, until a quarter of the power/wattage for a quarter of the cores, etc.

Sorry, but that is incorrect. Ryzen’s single-core/single-thread performance is very close to that of a current generation i5 desktop processor. It is at just about a 5% performance deficit in single-core/single-thread for the same clock speed, making it about ever so slightly ahead of prior generation Broadwell desktop processors. (Note: Skylake and Kaby Lake have identical IPC performance.) Mind you, this is just desktop. Ever since Bristol Ridge, AMD processors generally scale much better into mobile. This means Ryzen should either be neck-and-neck and perhaps even ahead in mobile performance.

Fanless, passive Core M/Y-class Core is indeed impressive, and it even comes close to both current Intel Core i5 and Ryzen 5 desktop processors. It’s good, but it’s not that good–at least, not yet. The Core i7-7Y75 you cite currently sits at a single-core/single-thread score of around 130 for Cinebench R15:

The second graph is a frequency versus voltage chart and does not give any insight into performance but does give us some insight into scaling and power draw. For this chart, we want as little voltage as possible which equates to as flat a line as possible, not a steep line. The first critical point refers to the first point where the frequency versus voltage slope changes to a higher, steeper sloop. This is most definitely not a sweet spot. The second critical point is even worse and is absolutely not what we want. Actually, anywhere below and before these two critical points (from 2000 MHz to roughly 3300 MHz) is the sweet spot.

A softer, flatter slope for the frequency versus voltage is always the desireable in processor design. Why? Processor power/wattage draw is generally an exponential cubic function of the voltage. In other words, with a minuscule increase in voltage comes an enormous increase in power. That is why we always want to keep voltage demands down at a minimum. Hence, the flatter the slope, the better and lower the power draw as we scale the design up and down. This second chart is just another reason why Ryzen is a very mobile-ready processor design.


#119

I agree, we are therefore going to be looking at steep slope on either quad-core ~15W. or dual-core ~8W, all are basically the relevant ones for mobile use. Especially where the V is relevant, we are looking at either quad-core config at ~15W or dual-core config of ~8W. Especially at dual-core, I don’t see any benefit for going Ryzen over Intel (for now).

In quad-core, I see that it may be able to make a dent on the market, but then again, you still need to include an iGPU into the equation, and as far as I know, iGPU takes quite a significant portion of power.

So we are looking at two massive uncertainties, that are the steep slope and iGPU.

The reason is because the sole source of that score is from one device, that is XPS 13 2-in-1. As you may already know, that device prioritizes minimizing heat generation over performance.

In case you didn’t know, we managed to get nearly the same score (125) on the V using just the Core i5-7Y54, so we are looking at even higher scores with the 7Y75.

so basically we are looking at even more deficit in frequency and per-core performance?


#120

No. Again, that second graph has nothing to do with performance. Think of the slope as voltage per MHz because it is all it is. Again, a higher slope means more and more power draw which is bad. Most tablets have base frequencies well below 3.3 GHz. Boost frequencies are a trickier matter since there is silicon trickery going on to achieve them (hundreds of thousands of quadrants of transistors being dynamically switched on and off to save power on the fly). So the flatter slope from 2 GHz to 3.3 GHz means less power draw.

Yes, the 7Y75 will be faster, but only barely. The Core i7-7300U gets 144 in Cinebench R15. Taking into account that it is fanless, the 7Y75 will probably be around 130 to 140. This is extremely close to even an i5 desktop processor. But it is still just that little bit of a ways off.

What some keen eyes have noticed is this performance curve is much better than the current AMD Bristol Ridge-based quad cores, which include 15W models. These 15W Bristol Bridge models in particular can actually manage a single-core score of around 2000 in Geekbench 4.0 (https://www.notebookcheck.net/AMD-A10-9600P-Bristol-Ridge-Notebook-APU.173188.0.html). Taking into account the confirmed and benchmark verified 52% IPC improvement of Ryzen, for a Ryzen 15W quad-core mobile processor, we should see a single-core score of at least 3000 and possibly more in Geekbench 4.0. Take away two cores and the power requirements already drop to 7W territory.


#121

@Hifihedgehog and @Patrick_Hermawan The mobile chips you are looking for from amd are most likely based on raven ridge which is a apu. ( zen cpu paired with the technology from the upcoming vega). How well it performs we will have to see.

Edit: Found this bench, said to be reliable, free to your interpretation of the article.


#122

The benchmark image seems to be some Chinese CPU manufacturer (huh?) claiming that their CPU perform as good as American counterparts.

I take it with a pinch of salt.


#123

I don’t really understand this, because I see it from another point of view. So does that mean that, running a Ryzen at 2 GHz doesn’t save much power as 3 GHz?

In that case, I think it wouldn’t make a very efficient chip, isn’t it?

Based on that link, the CPU can manage 76 on Cinebench single-core.

So if we add the IPC improvement of Ryzen

76 * 152% = 115

That is ~10% behind i5, and ~20% behind i7.

This is where I fail to understand why we would need these Ryzen chips. The advantage of Ryzen has been the core count, and if you ended up with only dual-core config, you just removed the Ryzen’s advantage, leaving it with its single-core deficit.

However, Id say a 4C/4T or 4C/8T config has quite a strong value, as Intel’s offering only starts at 45W if you want quad core.

Looking at their hypetrain history (ahem, RX 480 on AOTS), I wouldn’t trust the benchmarks until we get more variety of independent benchmarks