Official AMD Ryzen Benchmarks, Reviews, Prices, and Discussion

[Magic Hate Ball]

GPU takes care of video decoding. Unless you have some video with losslessly compressed RGB or YUV with no chroma subsampling.

Mumble audio decoding and networking measages are hardly intensive.

There are many fine justifications for more cores but Google Chrome and Discord...

Open Youtube, watch a 1080p video fullscreen, watch your CPU usage. Is it more than 10-20% on a single core at least? My work laptop hits 25-50% (in spikes to 50%, usually hovering at 25%) with its dual core i5.

 

[sirmo]

"There are many fine justifications for more cores but Google Chrome"

As someone who can have 100 tabs open at one time.. you'd be surprised how sluggish google chrome can become on weak systems. Many sites can be javascript heavy and a bunch of intensive tabs open can eat up a good amount of resources.

 

[Ketchup]

I think a lot of us are sitting and waiting/watching this thread and others. Just too much information!...

I know I am also curiously waiting to see what happens. On the one hand, the more I hear, the more I like it. But on the other hand, it has been several years since I saw a CPU from AMD that got me excited. So for a company that loses money year after year to suddenly have a very competitive and very well-priced CPU, I am fine with the wait-and-see approach.

 

[Atari2600]

Because Google Chrome is a massive CPU hog? ok.

The point is that in the real world, the user experience will be one where a CPU having more cores* is at least as competitive as the benchmarks show, and usually will be more competitive than one having less cores.


*For reasonable core counts. So don't reply saying since a 22core xeon isn't the gaming champ is proof of your position!

 

[Bouowmx]

Woops, I guess I stand corrected on those hardcore browser users.

Intel HD Graphics 620 on H.264 1920x1080 30 fps (I assume 4:2:0)

 

[Atari2600]

GPU takes care of video decoding.

Is that the same GPU that is flat out trying to run your game?

Or is it another GPU you have in the system?

 

[bjt2]

During CB benchmark the CPU clocks goes above 3.6GHz, they dont run the entire benchmark at 3.6GHz.

So the clock needed to make that score is even higher... I posted that calculation as a minimum...

 

[lolfail9001]

Is that the same GPU that is flat out trying to run your game?

Decoding is done by ASIC, so flat out or not, it barely cares.

 

[Bouowmx]

Is that the same GPU that is flat out trying to run your game?

Or is it another GPU you have in the system?

Within a single GPU, shaders (for games) and the video block are separate. Though GPU-Z only picks up this observation on Nvidia GPUs, which show separate shader and video load.

 

[Dresdenboy]

Woops, I guess I stand corrected on those hardcore browser users.

Intel HD Graphics 620 on H.264 1920x1080 30 fps (I assume 4:2:0)

Browsers with several script and ad-laden pages (even behind ad blockers, sites' inefficient codes remain) is my main non -gaming/developing use case.

 

[bjt2]

Wasn't there some consensus that Ryzen's SMT implementation is better than HyperThreading? If you account for that, the clock speeds don't have to be quite as high.

I was talking of 1800x vs 1800x at different clocks. The stock 1800x scores 1601-1615. The overclocked 1800X scores 2449. 16 threads vs 16 threads. If you make the proportion (assuming perfect scaling) you have 5.5GHz needed to score 2449. If turbo is involved, then the needed clock is even higher. If you also consider that the scaling with clock is not 100% linear, even higher clock is needed...

 

[Magic Hate Ball]

Woops, I guess I stand corrected on those hardcore browser users.

Intel HD Graphics 620 on H.264 1920x1080 30 fps (I assume 4:2:0)

That is interesting. I see something different on my Haswell processors. I guess they must be missing the hardware decode from that?

 

[Phynaz]

Streaming even just 1080p video eats more CPU time than you'd think...

I can stream Youtube with about 1% CPU usage. GPU decode for the win

 

[bjt2]

Nah, not just that.

And not that either. And yes, it probably does run entire bench on 3.6 or 3.7Ghz.

Better SMT throughput was established by earlier leaks. Damn, you have just made the whole thing more confusing. Gave some hope it actually OCs better than Broadwell too, though, so i'll give that possibility a shot.

I can't still explain why at idle is sitting at 5.1-5.2GHz. If you look at the screens, the multiplier is under 40x and the FSB is about 140.

Furthermore there is a more confusing problem with Ryzen.

Normal benchmarks, even CB, measures the time it took to perform some calculation.
How does CB measure the time? With the real time clock (RTC) into the South bridge.
Until now, the south bridge was external and probabily not invested by the LN2 and so it was at sane temperatures.
But now the SB is integrated into the CPU. Where is the quartz? What temperature it have during heavy LN2 session? Is the RTC clock reliable at -100C? Tons of questions... One should use a bench that gives time elapsed, like handbrake, and compare the time elapsed with one taken with an external clock...

 

[bjt2]

That seems pretty unlikely as Inte'sl 14 nm is far more mature. We'll probably get a better idea once the 4 core Ryzen CPUs drop, but from the 7700k LN2 runs, we know that the process is quite mature in terms of what those chips can hit. Also is the 14 nm LPE/LPP process that GF is using even designed for high clock speeds?

Any idea on how a CPU can do with 5.14GHz a score that should have with 5.5+GHz?

 

[Torn Mind]

"There are many fine justifications for more cores but Google Chrome"

As someone who can have 100 tabs open at one time.. you'd be surprised how sluggish google chrome can become on weak systems. Many sites can be javascript heavy and a bunch of intensive tabs open can eat up a good amount of resources.

Chrome wants RAM first and foremost. It is actually justifiable to get a high end desktop just because Chrome's memory usage easily goes over 32GBs.

Four cores will be enough for Chrome until the RAM bottleneck is cleared.

 

[tamz_msc]

Woops, I guess I stand corrected on those hardcore browser users.

Intel HD Graphics 620 on H.264 1920x1080 30 fps (I assume 4:2:0)

Run a Steam backup of a game like GTA V while you watch that video and see where the CPU utilization goes.

 

[Atari2600]

Decoding is done by ASIC, so flat out or not, it barely cares.

What would the thermal load be?

Would it be insignificant enough to avoid thermal throttling of the GPU?

 

[bjt2]

Open Youtube, watch a 1080p video fullscreen, watch your CPU usage. Is it more than 10-20% on a single core at least? My work laptop hits 25-50% (in spikes to 50%, usually hovering at 25%) with its dual core i5.

I have a notebook with celeron T3100 (dual core 1.9GHz) and series 4 chipset. I can watch 1080p movies with less than 6% CPU usage with VLC. Nowadays the DXVA is awesome...

 

[PotatoWithEarsOnSide]

Another user made a calculation, but with the 1700X.
I report here with 1800X as base
At 3.6GHz+ the 1800x scores at least 1601 at CB 15 nT.
The CB record is said to be 2449.
The minimum frequency needed to score 2449, given 1601 at 3.6GHz is 3.6*2449/1601=5.5GHz. If turbo was engaged (almost certainly) at stock, that number is even higher. If scaling is not 100% (almost certainly) this number is even higher.

I think that that 5.14-5.2 were idle clocks. During CB run the clock must be at least 5.5GHz to score that score. Otherwise it is unexplainable how a CPU that has similar IPC that BDWE scores more with 1GHz disadvantage (previous world record on 6.1GHz of previous intel architecture, HSW)

New member here. Been reading the various threads for a week or so. The above post drew my attention, especially given the responses that followed didn't really address it adequately from my POV.

CB15 scores posted are multi-thread, right? With results being linear, right? I assume so, given some of the calculations that I've seen in these threads.

If new record is 2449 @ 5.14GHz, even if R7 1800X hit 1601 @ 3.6GHz, the numbers don't add up.

2449 / 1601 ~ 1.53
5.14 / 3.6 ~ 1.42

So, for 1.42x base clock there was a result of 1.53x the base result. The implication being that the 5.14Ghz cannot have been the actual clock during the record run, unless AMD's implementation of SMT sees improved SMT as clocks increase. That seems to lack logic to me.

For me, that LN2 run was clearly run well above 5.14GHz.

Thoughts?

Disclaimer: I make no claims to having any knowledge about this kind of stuff.

 

[pj-]

That is interesting. I see something different on my Haswell processors. I guess they must be missing the hardware decode from that?

KBL had some good video decode optimizations

 

[GroundZero7]

So, for 1.42x base clock there was a result of 1.53x the base result. The implication being that the 5.14Ghz cannot have been the actual clock during the record run, unless AMD's implementation of SMT sees improved SMT as clocks increase. That seems to lack logic to me.

AMD's SMT is quite a lot better, we have seen it on countless benchies these last weeks.

 

[Magic Hate Ball]

I have a notebook with celeron T3100 (dual core 1.9GHz) and series 4 chipset. I can watch 1080p movies with less than 6% CPU usage with VLC. Nowadays the DXVA is awesome...

Yeah I see that now lol.

Just shows how much being a few years behind can do to you!

 

[bjt2]

New member here. Been reading the various threads for a week or so. The above post drew my attention, especially given the responses that followed didn't really address it adequately from my POV.

CB15 scores posted are multi-thread, right? With results being linear, right? I assume so, given some of the calculations that I've seen in these threads.

If new record is 2449 @ 5.14GHz, even if R7 1800X hit 1601 @ 3.6GHz, the numbers don't add up.

2449 / 1601 ~ 1.53
5.14 / 3.6 ~ 1.42

So, for 1.42x base clock there was a result of 1.53x the base result. The implication being that the 5.14Ghz cannot have been the actual clock during the record run, unless AMD's implementation of SMT sees improved SMT as clocks increase. That seems to lack logic to me.

For me, that LN2 run was clearly run well above 5.14GHz.

Thoughts?

Disclaimer: I make no claims to having any knowledge about this kind of stuff.

This was my point also. The 5.14/5.2 can be idle clocks or the bus float a lot during benchmark or worse the RTC clock used during time calculations by CB has drift because is under very low temperatures.

i repeat myself: to test the reliability of the RTC, time measures should be done under heavy LN2, like the one made by handbrake and concurrently the same measure must be done with an external clock...

 

[bjt2]

AMD's SMT is quite a lot better we have seen it on countless benchies these last weeks.

We are talking at same cores and threads. The only difference is clock. Apart the faster execution, the IPC should be similar.