Official AMD Ryzen Benchmarks, Reviews, Prices, and Discussion

[Blake_86]

If you want your computer to run fastest in fa4 and not bf1 by all means get a 7700k instead. Please.The 2667 vs 3200 ram doesnt change that a all. Just stop the whining.

I m not whining. Dont think i m a hater or something. I d realy like to build a ryzen sistem cause i m aware of it' s potential.
But you can agree that with ryzen there is a noticeable difference from 3200 and 2660, this not happen with intel cause of the very architecture of the cpu.
Ram compatibility should be a priority for amd, more then it is for intel.

 

[Mopetar]

I've been constantly bumping it: https://forums.anandtech.com/threads/ryzen-w-amd-gpu-performance-in-fallout-4.2501467/

A draw call benchmark does NOT mean a game benchmark, just your word for it. I don't take your word since all these other benchmarks say otherwise.

Please listen to what I am saying DRAW CALL BENCHMARK DOES NOT EQUAL GAME BENCHMARK

I think that you're lacking some reading comprehension or aren't paying attention.

Even if Ryzen is particularly bad in this limited case, it's still a workload on a game engine built out of another engine that dates back at least as far as Oblivion which is about a decade old at this point.

You've singled out one aspect of a particular game engine that isn't terribly relevant. Why you keep harping on it as though it's important or overall indicative of a problem when there are dozens of other general benchmarks that show Ryzen does comparably to other CPUs in its price or performance category is beyond me.

So MajinCry is on to something, then?

Or just on something. If this were a particular piece of software for some professional workload, it might be a reasonable point to make. However, no one sits around their office waiting on draw calls to get completed in order to get paid so its beyond pointless.

Even in a game like FO4, it's a tiny corner case. How much time do you actually spend at the Corvega area staring off into the distance? Even comparing it to Diamond City it's pretty obvious that its an egregious case. It's interesting to look at worst case performance (which is why some benchmarks even include the .1% frametimes) but a bigger deal is being made out of this than it deserves. We have an edge case of a corner case for something that isn't particularly relevant for the future.

At least the Quake 2 benchmark was amusing.

 

[inf64]

I m not whining. Dont think i m a hater or something. I d realy like to build a ryzen sistem cause i m aware of it' s potential.
But you can agree that with ryzen there is a noticeable difference from 3200 and 2660, this not happen with intel cause of the very architecture of the cpu.
Ram compatibility should be a priority for amd, more then it is for intel.

You can bet AMD is working hard with board partners to get the better support for high clocking DDR4. It is a process that takes time, it needs validation because nobody sane would claim to support something without being 100% sure it works without problems.

 

[krumme]

I m not whining. Dont think i m a hater or something. I d realy like to build a ryzen sistem cause i m aware of it' s potential.
But you can agree that with ryzen there is a noticeable difference from 3200 and 2660, this not happen with intel cause of the very architecture of the cpu.
Ram compatibility should be a priority for amd, more then it is for intel.

The difference is 2%. And the half is probably due to faster ram. The issue is blown out of proportions. Its not anything remotely in comparison to the fo4 or bf1 differences between the 2 arch. Go look at eg computerbase frametimes for 1800x vs 7700k. It looks like what it is; 4c vs 8c with the 8c having the 20% st perf deficit in a game engine that uses all the cores. Its a huge difference for playability. The 7700k looks nasty. Its not potential its real world bad performance right there in what have historically been far yhe most important engine. And thats even for dx11.

Excactly what fo4 bm also shows but from an older engine. Aprox 20% (18%) average uplift for 7700 for both avg and min.

In this context 2% is utterly uninportant. The meaningfull differences is to what degree the game engine uses more cores effectively. Lets not get lost in the woods.

My take on it is then. In bf1 (or crysis3 to alesser degree) the frametimes for 7700 gets some dips into like 40fps times while the 1800 stays a 80fps. On fo4 there is less variability i asume but the min never gets lower than 90.
When a 7700k doesnt deliver on the 0.1 min or frametimes its very low. When the 1800x gets low - and that happens for all the old games - its doesnt have the same deep impact because fps is 100 fps anyway.

 

[Blake_86]

You can bet AMD is working hard with board partners to get the better support for high clocking DDR4. It is a process that takes time, it needs validation because nobody sane would claim to support something without being 100% sure it works without problems.

I realy hope you are right and in the next week things grows better

 

[Dresdenboy]

Memory seems to give a big boost to Ryzen, but that's about it, far as I can tell. And no, sorting by graphics card would only make sense if we were using 6670-class hardware. But once we're past the super low end, we're at CPU bottleneck territory.

I think, there is still a GPU vendor dependency, as the CPU part processing the drawcalls is in part happening in the GPU driver stack.

 

[Mopetar]

Memory seems to give a big boost to Ryzen, but that's about it, far as I can tell. And no, sorting by graphics card would only make sense if we were using 6670-class hardware. But once we're past the super low end, we're at CPU bottleneck territory.

I think, there is still a GPU vendor dependency, as the CPU part processing the drawcalls is in part happening in the GPU driver stack.

Also, doesn't FO4 use PhysX or some other Gameworks features that get offloaded to the CPU in the case of using an AMD card?

 

[MajinCry]

I think, there is still a GPU vendor dependency, as the CPU part processing the drawcalls is in part happening in the GPU driver stack.

NVidia's driver is meaningless for synthetic draw call workloads. This was discussed in the thread; when the same draw call, with no additional calls (just drawing the object, no shadows, no lights, no other objects), NVidia's driver plows through them a couple times faster. than AMD's. Outside of Asteroids, however, that driver behaviour is useless for games. It's purely an optimization for synthetics.

 

[PhonakV30]

Review at 3Ghz! Ryzen really is amazing CPU.
https://www.planet3dnow.de/cms/30155-amd-ryzen-7-1800x-teil-1/subpage-benchmarks-bei-3-ghz-aida64/

 

[Agent-47]

But it seems to just be stock FO4 even if in a town. It does show that most of the Ryzen cores are nearly idle though and (not sure how they are numbered) it appears that both CCX have stuff happening even if the first one still has mostly unused cores.

you know this argument is getting old. there is a very easy fix for that.

get a high speed ram to spin the CCX at higher speed. since lower latency does not seem to be as big an improvement, get a high latency high speed ram to keep the prices in control.

 

[DrMrLordX]

you know this argument is getting old. there is a very easy fix for that.

get a high speed ram to spin the CCX at higher speed. since lower latency does not seem to be as big an improvement, get a high latency high speed ram to keep the prices in control.

Sadly not all boards support speeds above DDR4-3200 right now. That new AGESA code from AMD isn't proliferating quickly.

 

[Burpo]

 

[Agent-47]

Sadly not all boards support speeds above DDR4-3200 right now. That new AGESA code from AMD isn't proliferating quickly.

Yes. I agree, specially for mATX boards which is the type of board I use. But MSI has listed 3200 MHz ram support for its mortar B3500 using corsair, g skill and hyperx. Sadly only on single channel so far. But for gaming ryzen is more bottlenecked by CCX bandwidth rather than memory bandwidth so it should be alright I think.

 

[DrMrLordX]

Unfortunately, for now, 1/2 RAM speed = CCX speed. Those of us with DIMMs that can handle speeds of DDR4-3733 or higher want to push for higher memory clocks.

 

[beginner99]

get a high speed ram to spin the CCX at higher speed. since lower latency does not seem to be as big an improvement, get a high latency high speed ram to keep the prices in control.

About htis RAM speed / ccx issue. Do we have any data about ma RAM speed vs. sku? Maybe the 1700 also has been binned with a worse IMC than say a 1800x. I have seen people having issues to get above 2600 mhz on the asus board while others managed 3200. Interesting to know if it is silicon lottery only or also sku dependent.

 

[ultima_trev]

For some reason 3DMark Fire Strike combined test scores better when I have two cores disabled in Ryzen Master.

My two best 8 core runs:

http://www.3dmark.com/fs/12114242

http://www.3dmark.com/fs/12082722

2 cores disabled:

http://www.3dmark.com/fs/12136472

http://www.3dmark.com/fs/12136467

http://www.3dmark.com/fs/12136449

Could this be due to the fact that I'm only running DDR4-2400, i.e. not enough memory or CCX bandwidth for all those threads/cores?

 

[Malogeek]

Unfortunately, for now, 1/2 RAM speed = CCX speed. Those of us with DIMMs that can handle speeds of DDR4-3733 or higher want to push for higher memory clocks.

It really needs to be considered as 1:1 ratio as the fabric is running at the speed of the memory controller, but the DIMMs are double data rate.

About htis RAM speed / ccx issue. Do we have any data about ma RAM speed vs. sku? Maybe the 1700 also has been binned with a worse IMC than say a 1800x. I have seen people having issues to get above 2600 mhz on the asus board while others managed 3200. Interesting to know if it is silicon lottery only or also sku dependent.

It's down to the specific models of RAM as well as access to beta bios' for certain boards at the moment.

 

[Agent-47]

For some reason 3DMark Fire Strike combined test scores better when I have two cores disabled in Ryzen Master.

My two best 8 core runs:

http://www.3dmark.com/fs/12114242

http://www.3dmark.com/fs/12082722

2 cores disabled:

http://www.3dmark.com/fs/12136472

http://www.3dmark.com/fs/12136467

http://www.3dmark.com/fs/12136449

Could this be due to the fact that I'm only running DDR4-2400, i.e. not enough memory or CCX bandwidth for all those threads/cores?

more likely due to the fact that 8 MB cache is not shared between 3 cores making it 2.7Mb/core instead of 2.

 

[ultima_trev]

more likely due to the fact that 8 MB cache is not shared between 3 cores making it 2.7Mb/core instead of 2.

Huh, then shouldn't disabling 4 cores be even better? However when I tried that it made the combined score a bit worse than the 8 core config. I just achieved my personal best Fire Strike score overall, with a 3+3 config no less:

http://www.3dmark.com/fs/12139687

Even when I run all 8 cores at 4.05, it doesn't score this well.

 

[Puffnstuff]

What is even more interesting is the work AMD are doing with Bethesda to optimize their games as announced with their partnership.

This could all be hilariously blown away if the Fallout engine gets updated to utilise Zen and Vega

Its a well known fact that software always lags hardware which is exemplified when considering how long multicore cpu's have been on the market and how many software applications still utilize single thread performance. Even so there's a waterfall effect in Ryzen's cache system that needs to be addressed to speed up the flow of data between them and the cpu and I hope that the next iteration of it is much better in this area. While its still a magnificent leap forward for AMD it cannot dethrone Intel's finest and no amount of fanning will make it happen.

 

[french toast]

Some may say im bat skit crazy BUT Isn't the consoles effectively using a prototype/mini ryzen? 4x2 jaguar cores setup very similar to 2 CCXs.
I would suggest getting 80% of games patched wont be as difficult as some make out, the engines are inherently designed to be numa aware are they not?

 

[alexruiz]

Huh, then shouldn't disabling 4 cores be even better? However when I tried that it made the combined score a bit worse than the 8 core config. I just achieved my personal best Fire Strike score overall, with a 3+3 config no less:

http://www.3dmark.com/fs/12139687

Even when I run all 8 cores at 4.05, it doesn't score this well.

I think it is all bandwidth.
Higher speed RAM would help even more with the scores.
Here is my best run in a PC I built for a customer:
http://www.3dmark.com/fs/12079421

Your system has a higher clocked GPU, higher clocked CPU, but slower clocked RAM, and that offsets it.
My personal observation is that physics test loves bandwidth.
The 1700X was overclocked using Ryzen Master. I could ave gone to 4.0GHz in that chip, but the voltage required was insane (1.425V) so the 3.9Ghz at 1.3875V was a nicer balance.

 

[zinfamous]

My problem with MajinCry's graph is that there is little control for those results. Assuming there is validity to the results, wouldn't it make far more sense to sort them by graphic card? Also, (and I haven't looked at all the people who contributed) shouldn't other hardware be controlled like memory etc.?
Since the graph is based on FSP/(GHz * 10) where does the other variation come from? The 2500K scores at 4.5GHz and 5.1GHz for instance? Is there more going on there like memory speed or messing with the FSB (unlikely)? Why is that Haswell Pentium G3258 by far the fastest CPU?

While I think the results could be meaningful all they really show is that Bethesda have once again used a very old and poor engine for FO4 (and SSE I guess too). So yes, I do plan to go back to SSE once SKSE is out but I would not base my choice of CPU on that one game.

What? That dang thing still isn't out? I've been patiently waiting for this, pretty much ignoring the slow progress, assuming it would be ready if I just stop paying attention...that was about 2 months ago.

 

[imported_jjj]

CPU power consumption in Metro Last Light. Ofc as always tested at high FPS so not relevant for most.
Fewer cores end up with higher load per core and fall behind.

 

[guachi]

I wonder how preposterously low a 1700 would be on that chart.