AMD Ryzen Mobile APU Benchmarks: 90% Faster Than Previous Gen

Megalith

Megalith

24-bit/48kHz
Staff member
Joined
Aug 20, 2006
Messages
13,000
Early benchmarks for AMD’s upcoming Raven Ridge APUs, which feature the Zen CPU and Vega GPU microarchitectures, have surfaced at Geekbench: a Ryzen 5 2500U APU managed a score of 3,625 points in the single-core portion of the benchmark, and 9,723 points in the multi-core portion. Compared to a current generation A12-9800B, which is AMD's fastest mobile Bristol Ridge APU, the new Raven Ridge is around 90 percent faster in Geekbench's multi-core test (9,723 versus 5,115) and about 56 percent faster in single-core performance (3,625 versus 2,315).

Our standard disclaimer when it comes to these sorts of things applies—this is just a single benchmark, and partially for that reason, we have to be careful not to draw any premature conclusions. That said, the scores on display are in line with what AMD has been promising. AMD previously said Raven Ridge would deliver a 50 percent bump in CPU performance and 40 percent jump in GPU performance, while consuming 50 less power, compared to its 7th generation APU.
 
just out of curiosity. anyone know how that compares to the competition? funny how you see some intel laptops selling with the same processors in them as tablets! ... and people buy that junk!, and for big bucks!
 
d3athf1sh said:
just out of curiosity. anyone know how that compares to the competition? funny how you see some intel laptops selling with the same processors in them as tablets! ... and people buy that junk!, and for big bucks!
Click to expand...

Pardon my ignorance, but from a performance standpoint with the Intel processors, which is losing out? The laptop or the tablet? To me, a laptop should always perform better and faster than a tablet, but if a tablet just so happens to have laptop performance on it, then that would be one hell of a tablet.
 
Cool, been waiting to see what the Ryzen APUs can do. Now let's see some GPU info. I think these things are going to kick ass in budget/htpc and mobile sectors.
 
This is big news. It'll be good for the low end desktop market as well, running APUs. Something to compete with the i3 desktops without discreet graphics.
 
d3athf1sh said:
just out of curiosity. anyone know how that compares to the competition? funny how you see some intel laptops selling with the same processors in them as tablets! ... and people buy that junk!, and for big bucks!
Click to expand...

A 8550U scores 4766ST/15110MT on a Dell laptop.
https://browser.geekbench.com/v4/cpu/3986326

The APU benches isn´t done on a laptop and will by nature be quite lower in real life.
 
40% better iGPU performance with half the power consumption...very, very impressive.

I'm thinking a replacement for my aging Ivy Bridge i5 with iGPU laptop may be on the horizon.
 
Shintai said:
I5 8350U
i5 8250U

Good luck ;)
Click to expand...
Those are $300 chips with very low base clocks (1.6-1.7Ghz?!!?) and a stupid wide turbo range that will almost never be hit in a thin n' light platform (which is what those chips are meant for) due to heat (10-15W chips!!) and power constraints.

AMD's Zen APU's will probably be cheaper, have similar power usage/TDP's, and perform close enough to not matter per core and have the same or more cores with better iGPU's.

They'll be fine.
 
it's obvious that the best product coming out AMD in the last or maybe ever, is Raven ridge, they are in a unique postition to deliver high compute and graphic chip, Ryzen by design is perfect for this, and Vega also, since it doesn't use alot of stream processors, AMD driver team should be able to keep the utilisation very efficient.
now comes the question about memory bandwidth... HBM or not to HBM
really this could be a terrific product, just hope they don't screwit up.
 
I believe this M-mobile version will not be HBM. At least not in any 2017 release. When the desktop APU comes out next year I am hoping that they can get some HBM on it.
 
You will never see HBM in an APU.

repeat that to yourself at night.

You will never see HBM in an APU.
 
Been waiting for AMD's new APU's. Hopefully I can build a HTPC with one.
 
griff30 said:
And we'll never need more than 640k of ram......
Click to expand...

It's a cost thing. APUs = cheap, HBM = expensive.

Remember back when people were going crazy about how ZIP drives were the storage of the future, or that micro-dvds would replace CD music? The truth is that cheaper solutions usually win out over more expensive ones. HBM is an expensive solution to a simple problem that is solved with faster than standard DDR4 speeds.
 
KazeoHin said:
It's a cost thing. APUs = cheap, HBM = expensive.

Remember back when people were going crazy about how ZIP drives were the storage of the future, or that micro-dvds would replace CD music? The truth is that cheaper solutions usually win out over more expensive ones. HBM is an expensive solution to a simple problem that is solved with faster than standard DDR4 speeds.
Click to expand...

There is a cheaper version of HBM coming and also as HBM matures there is no need to use the latest stuff for a APU. But who knows if they will ever try it, but speaking in absolutes rarely turns out well.
 
Apple A11 clocked in at 4100+/9800+ on this benchmark...

That leads me to a couple of possible conclusions:

Either

Holy crap the A11 is smoking AMD here and has pulled up near Intel, with a chip that's in a phone. (Granted, those phones cost more than the devices your likely to see and AMD APU installed in)

And/Or

Geekbench is worth its weight in manure.
 
Brian_B said:
Apple A11 clocked in at 4100+/9800+ on this benchmark...

That leads me to a couple of possible conclusions:

Either

Holy crap the A11 is smoking AMD here and has pulled up near Intel, with a chip that's in a phone. (Granted, those phones cost more than the devices your likely to see and AMD APU installed in)

And/Or

Geekbench is worth its weight in manure.
Click to expand...

Yeah, but just look at how fast that Samsung SM-N750 is in geekbench 3!

Yup... the program looks like it can be exploited to obscure results.
 
Just remember, everyone:

When geekbench shows AMD-negative results: it's an interesting talking point that should fuel rumours and discussion.

When geekbench shows AMD-positive results: it's a broken, irrelevant metric that does not reliably demonstrate real world usage.

There. Now the status quo can be maintained, no matter how these results stack up.
 
KazeoHin said:
It's a cost thing. APUs = cheap, HBM = expensive.

Remember back when people were going crazy about how ZIP drives were the storage of the future, or that micro-dvds would replace CD music? The truth is that cheaper solutions usually win out over more expensive ones. HBM is an expensive solution to a simple problem that is solved with faster than standard DDR4 speeds.
Click to expand...
HBM is expensive NOW, soon it will be more cost efficient to use it.
AMD surely sees the benefits to laptops: less power, saves die space. Once the cost is competitive with DDR4, it will happen. It's inevitable.
 
griff30 said:
HBM is expensive NOW, soon it will be more cost efficient to use it.
AMD surely sees the benefits to laptops: less power, saves die space. Once the cost is competitive with DDR4, it will happen. It's inevitable.
Click to expand...

HBM uses more power than DDR, a lot more. Also its capacity limited. And even against GDDR it got zero power and speed benefits.

And cheaper HBM, that story have been on for a couple of years now. HBM will always be more expensive because its penalized by its inherit bad cost structure.

The real question is if there is a future for HBM after it failed due to GDDR not going away. And HMC is killing it in the top.

HBM cost competitive with DDR? Yet its not even remotely close to GDDR or LPDDR? :D
 
Shintai said:
HBM uses more power than DDR, a lot more. Also its capacity limited. And even against GDDR it got zero power and speed benefits.

And cheaper HBM, that story have been on for a couple of years now. HBM will always be more expensive because its penalized by its inherit bad cost structure.

The real question is if there is a future for HBM after it failed due to GDDR not going away. And HMC is killing it in the top.

HBM cost competitive with DDR? Yet its not even remotely close to GDDR or LPDDR? :D
Click to expand...

Yet thats why Nvidia is using it on Volta? Your so full of it. HBM uses less power then GDDR by quite a bit and is well documented and well covered in the Vega thread. There is a cheaper version of HBM coming and at the rate DDR memory costs are going your going to wish you had HBM, GDDR is increasing in costs and is expected to keep climbing.
 
Gideon said:
Yet thats why Nvidia is using it on Volta? Your so full of it. HBM uses less power then GDDR by quite a bit and is well documented and well covered in the Vega thread. There is a cheaper version of HBM coming and at the rate DDR memory costs are going your going to wish you had HBM, GDDR is increasing in costs and is expected to keep climbing.
Click to expand...

You can keep champion HBM year after year it gets you nowhere.

HBM2 doesn't use any less power than GDDR5X/GDDR6. Even Hynix the developer of HBM states so with their GDDR6 slides.

HBM is increasing in cost too, its not excluded.

The P100 and V100 uses HBM2 due to size for the mez. slots and ECC. GP102 and GV102 for example uses GDDR5X and GDDR6. And dont be surprised if GE100 will use HMC instead of HBM.

The cheaper version of HBM is still nothing but an uncommitted slide project. And it also comes with sacrifices, including performance.

SK-Hynix-GDDR6-4.jpg

ePTvwDS.jpg


Looks familiar doesn´t it. The success of HBM is entirely dependent on the collapse of GDDR. Not on its own merits.
 
Shintai said:
You can keep champion HBM year after year it gets you nowhere.

HBM2 doesn't use any less power than GDDR5X/GDDR6. Even Hynix the developer of HBM states so with their GDDR6 slides.

HBM is increasing in cost too, its not excluded.

The P100 and V100 uses HBM2 due to size for the mez. slots and ECC. GP102 and GV102 for example uses GDDR5X and GDDR6. And dont be surprised if GE100 will use HMC instead of HBM.

The cheaper version of HBM is still nothing but an uncommitted slide project. And it also comes with sacrifices, including performance.

SK-Hynix-GDDR6-4.jpg

ePTvwDS.jpg


Looks familiar doesn´t it. The success of HBM is entirely dependent on the collapse of GDDR. Not on its own merits.
Click to expand...

That is GDDR6 which no one has, almost everyone is running GDDR5 and a tiny amount are running GDDR5X. Also never did I say I championed HBM but your FUD needs to be corrected. Both have their pros and cons to using them and you like to forget that Nvidia is using HBM as well.
 
Shintai said:
You can keep champion HBM year after year it gets you nowhere.

HBM2 doesn't use any less power than GDDR5X/GDDR6. Even Hynix the developer of HBM states so with their GDDR6 slides.

HBM is increasing in cost too, its not excluded.

The P100 and V100 uses HBM2 due to size for the mez. slots and ECC. GP102 and GV102 for example uses GDDR5X and GDDR6. And dont be surprised if GE100 will use HMC instead of HBM.

The cheaper version of HBM is still nothing but an uncommitted slide project. And it also comes with sacrifices, including performance.

SK-Hynix-GDDR6-4.jpg

ePTvwDS.jpg


Looks familiar doesn´t it. The success of HBM is entirely dependent on the collapse of GDDR. Not on its own merits.
Click to expand...
I don't really care about HBM/GDDR but the first slide doesn't say anything really, and the second says HBM uses less power for more bandwith.
 
Gideon said:
That is GDDR6 which no one has, almost everyone is running GDDR5 and a tiny amount are running GDDR5X. Also never did I say I championed HBM but your FUD needs to be corrected. Both have their pros and cons to using them and you like to forget that Nvidia is using HBM as well.
Click to expand...

You talk about HBM in 100-200$ products. Nvidia is using HBM in 10000$ products, yet not in 1000-5000$ products. Yet you cant see the difference.

Nvidia has of any told you how bad the cost structure on HBM is.
 
Last edited:
Uvaman2 said:
I don't really care about HBM/GDDR but the first slide doesn't say anything really, and the second says HBM uses less power for more bandwith.
Click to expand...

The 2 slides says the same vs GDDR5. In short HBM2=GDDR6 in performance metrics.
 
MrDeaf said:
Yeah, but just look at how fast that Samsung SM-N750 is in geekbench 3!

Yup... the program looks like it can be exploited to obscure results.
Click to expand...

I believe Apple's chips are wider (can execute more instructions simultaneously) than most other architectures. That really shows up on some benchmarks but often doesn't in real-world tasks. There's no reason to think the A10/A11 etc is inferior to Ryzen. AMD's last architecture was poor and Ryzen arch is like 50% faster. AMD has openly said that Ryzen 2 should be much faster as well because there is low hanging fruit that they didn't have time to get to for Ryzen. Apple releases new chips every year or two (way faster than AMD) and has been consistently going bigger and faster each time. It would be great if Apple made a desktop chip soon so that we could compare running real software. If Apple and AMD had competitive cpu architectures it could really make for some good tech news :)
 
Whoops. The FUD squad doesn't want you to see this one.

RR bench.jpg
 
So about 25% better on the graphics side compared to Iris 640. Not by a whole lot, but at least it's something.

Interested to see the Power Consumption and CPU performance comparisons.
 
That is probably with DDR4-2133 though.

The iGPU in the APU will likely be fairly bandwidth constrained so going to DDR4-2400 or DDR4-2666 will give you nice gains and those aren't impossible or particularly expensive to get in SODIMM's or regular DIMM formats either.
 
Unfortunately, not a single OEM will ship a product with anything other than bare basic generic 2133.
 
OEM's? Yeah probably.

DIY'ers looking to build a cheap but good enough Mom-box or something for their kid like those that post at [H]? They'll buy the faster stuff by default.
 
mesyn191 said:
OEM's? Yeah probably.

DIY'ers looking to build a cheap but good enough Mom-box or something for their kid like those that post at [H]? They'll buy the faster stuff by default.
Click to expand...

True. That said, the price premium of higher speed RAM can negate the APU advantage. Let's hope that's not the case. I want to see the ultimate gaming NUC with a tiny board, a CPU, RAM, SSD ant that's it. With active cooling and a sub-30w chip, there's no reason it can't be done in a system that can fit in the palm of your hand, but play modern games at 1080p 60
 
You must log in or register to reply here.
Back
Top