Support NeoGAF

[Frankfurter]

Given the way a mathematical mean works, that is unlikely but not impossible.

Another reason why a frametime distribution chart (or even just 99% frametime if you want a single number) is a much better metric for game performance than aggregate FPS.

Strictly speaking it's not impossible, true. But it's pretty much impossible in reality. It just doesn't make any sense.

 

[BlastProcessing]

Oh my, if this benches well its going to be my next build at years end. Theres is a high chance of it being my first full AMD build.

 

[Frankfurter]

I don't know why you think that it would be "fake" or "bad benchmarking".
Averages and maximum framerates are often reliant on GPU performance, while minimums are nearly always CPU-bound.
That's why Anandtech's tests that only look at average framerates show hardly any difference between CPUs.

This was exactly what I was concerned about with Ryzen having 8 slower-clocked, lower-IPC cores.
That's great if your application can support it, but the majority right now do not.

So far, the fastest RAM advertised as being supported is DDR4-3200 for AM4 motherboards, so 4000 is probably unlikely. You might be able to use DDR4-3600 RAM with them though.
Apparently AMD are sending out review samples with 3000MHz Corsair RAM.
Kaby Lake boards support DDR4-4266.

We are looking at exactly the same CPU, just with different frequencies. It does not make sense at all that a 7700k@stock would have higher average framerates (albeit slightly) than a 7700k@5Ghz. If the situation were 88 vs. 88fps for average, I could see it being unlikely, but possible. I.e. the min and max. difference would just be like very, very short short. But the 7700k@stock has to be faster at some point during this chart than the same processor with more clock speed.

 

[AnythingBut]

So, am i going to have to chuck out my hyper 212 and stump up for a AM4 cooler or will it fit on a AM4 board?

 

[Tallshortman]

So, am i going to have to chuck out my hyper 212 and stump up for a AM4 cooler or will it fit on a AM4 board?

I would like to know this as well, seems like there's been mixed reports so far.

 

[Seronei]

So, am i going to have to chuck out my hyper 212 and stump up for a AM4 cooler or will it fit on a AM4 board?

http://www.cmstore-usa.com/amd-am4-upgrade-kit-rr-accy-am4b-r1/

Just pay shipping.

 

[Paragon]

Technically they do, but it's no slam dunk unless you have access to several CPU samples and aren't afraid to go wild with voltage. I had a heck of a time just getting some good 3866 RAM going.

Is that on a board which actually advertises 4266 as a supported speed, and has the RAM you're trying to use on the QVL list?
I'm not saying you're wrong, just that if I bought those parts I would expect them to work simply by enabling the XMP profile.

We are looking at exactly the same CPU, just with different frequencies. It does not make sense at all that a 7700k@stock would have higher average framerates (albeit slightly) than a 7700k@5Ghz. If the situation were 88 vs. 88fps for average, I could see it being unlikely, but possible. I.e. the min and max. difference would just be like very, very short short. But the 7700k@stock has to be faster at some point during this chart than the same processor with more clock speed.

If the average framerate in the test is dependent on the GPU rather than the CPU, 1 FPS is well within margin of error despite the CPU being overclocked 20%.

 

[Weevilone]

Is that on a board which actually advertises 4266 as a supported speed, and has the RAM you're trying to use on the QVL list?
I'm not saying you're wrong, just that if I bought those parts I would expect them to work simply by enabling the XMP profile.

Yeah your expectations are out of whack, and I thought just like you when I bought it.

The motherboard is rated for 4133 I think, and the RAM is rated for 3866. Memory above 3866 is crazy expensive. The memory is on the motherboard manufacturer's QVL list at 3866, and the motherboard is on the memory manufacturer's QVL list at 3866.

It wouldn't even think about a successful POST at XMP settings. I took it to several forums where people overclock and such, as I guess I'm out of practice. It was explained to me many times (with snickers I think) that the 7700k is rated for 2400 memory, so anything above that depends on the quality of your sample. If you get a golden CPU it's not a big deal. If you get anything less, you'll almost certainly be tweaking above 3600. I feel my CPU is certainly better than most, but if I was going to do 4.8 or 5.0 with anything faster than 3600, I'd definitely need to replace the thermal interface between the die and heat spreader to handle the voltage I'd need.

Essentially buying all matching RAM and MB only ensures that your CPU will be the weak link, short of an ideal processor sample.

 

[AnythingBut]

http://www.cmstore-usa.com/amd-am4-upgrade-kit-rr-accy-am4b-r1/

Just pay shipping.

Cheers.

 

[TC McQueen]

So, am i going to have to chuck out my hyper 212 and stump up for a AM4 cooler or will it fit on a AM4 board?

I would like to know this as well, seems like there's been mixed reports so far.

I decided to pony up for a Cryorig H7, just to avoid the hassle of dealing with the 212's mounting system, especially after reading someone's horror story of reapplying thermal paste and having the thing slip on the CPU heatspreader.

 

[Nostremitus]

I decided to pony up for a Cryorig H7, just to avoid the hassle of dealing with the 212's mounting system, especially after reading someone's horror story of reapplying thermal paste and having the thing slip on the CPU heatspreader.

Noctua's coolers need a new mounting kit. I assume other coolers will too.

 

[Thraktor]

Given the way a mathematical mean works, that is unlikely but not impossible.

Another reason why a frametime distribution chart (or even just 99% frametime if you want a single number) is a much better metric for game performance than aggregate FPS.

Part of the problem here is they don't seem to be defining what their "minimum" FPS is, which makes it more difficult to interpret. Is it the 95th percentile frame time? The 99th? Is it one second of game time that simply had the lowest number of frames rendered? Each of these have different implications for the smoothness of the gaming experience.

A frame time distribution chart would be ideal (along with some measure of short-term frame time variance, as that isn't captured in a simple distribution), but if you're going to reduce it down to a single number, then I think for most people the most straight-forward statistic would be "percentage of frames over 16.7ms". For people with 60Hz monitors (i.e. most people) it tells them how frequently they could expect to experience frame-rate drops, which should be pretty easily relatable as a measure of the smoothness of a gaming experience.

Of course it'll become less relevant as more people move to high refresh rates and adaptive sync, but if we want to simplify (which I don't necessarily think we do) then that's the simplification I'd go with.

http://wccftech.com/amd-ryzen-1700x-benchmarks-leaked-beats-kaby-lake-ipc/

What on earth is Userbench?

 

[TC McQueen]

Noctua's coolers need a new mounting kit. I assume other coolers will too.

There seem to be very few coolers that won't require a new mounting kit.

 

[papersleeves]

What on earth is Userbench?

I would assume this is it.

 

[Durante]

Part of the problem here is they don't seem to be defining what their "minimum" FPS is, which makes it more difficult to interpret. Is it the 95th percentile frame time? The 99th? Is it one second of game time that simply had the lowest number of frames rendered? Each of these have different implications for the smoothness of the gaming experience.

That's usually how people define "minimum FPS". (And yes, it does make very little sense, since the arbitrary "second" discretization it introduces can have a significant impact on the final number while having no relation to the game experience).

A frame time distribution chart would be ideal (along with some measure of short-term frame time variance, as that isn't captured in a simple distribution), but if you're going to reduce it down to a single number, then I think for most people the most straight-forward statistic would be "percentage of frames over 16.7ms". For people with 60Hz monitors (i.e. most people) it tells them how frequently they could expect to experience frame-rate drops, which should be pretty easily relatable as a measure of the smoothness of a gaming experience.

That's actually one of the metrics techreport uses (or at least close, I don't think they use percentages). The present the frames over 50 ms, 33.3 ms, 16.7 ms and 8.3 ms. So you get a good idea of severe stutter (>50) and performance consistency at 30, 60 and 120 FPS.

 

[General Lee]

Gibbo at OcUK seems to have some interesting tidbits to share. Apparently he's managed to get R7 1700 to around 4 GHz.

But we just tested a 1700, it hit 4.0GHz stable in everything, but ONLY in the Crosshair mainboard, the lower-end boards it was hovering around 3.80GHz as the VRM's were cooking with extra voltage. It however was maxing around 4050MHz, so I'd say 1700 can do 3.9-4.1GHz, of course the 1800X will probably do 4.1-4.3 as no doubt better binned, but if your clocking the motherboard has a big impact on the overclock and so far Asus Crosshair and Asrock Taichi seem the best two.

For memory speeds it seems only the high end boards like Asus Crosshair will go past 3000 MHz, at least for now.

A comment from Asus:

"
I’ve decided to provide some recommendations on DDR4 limitations concerning AM4 currently.

As it stands the AMD code has restricted RAM tuning options which means many RAM kits at launch will not be compatible. This is the same for our competitors also.
What we recommend is the following:
If fully populating a system with 4 DIMMs (2DPC), use memory up to a max of 2400MHz.
If using 1DPC (2 DIMMs) ensure they are installed in A2/B2 and use memory up to max of 3200MHz.

The indication I have received from HQ is that AMD has focused all their efforts on CPU performance so far and will release updated code in 1~2 months when we expect improved DDR4 compatibility and performance."


In short if filling all 4 DIMM's set your speed to 2400MHz and work up from there.
If using 2 DIMM's put them in the A2/B2 slots and a max of 3200MHz should be possible.

In our testing only the Crosshair board achieved 3000-3200MHz, the others were in the 2400-2666MHz range.

BIOS updates will come!

https://forums.overclockers.co.uk/t...ou-need-to-know.18770248/page-3#post-30533843

I guess it won't come as a surprise pre-orderers have no trouble going straight for the top end.

Over 500 Ryzen pre-sold, 1800X is the best seller.

 

[Coulomb_Barrier]

This is great from Computerbase, on the frequency vs core count debate.

https://videocardz.com/66354/core-count-vs-frequency-what-matters-for-gaming

They benched these games:

Anno 2205 – FPS
Ashes of the Singularity (DX12) – FPS
Battlefield 1 (DX11/DX12, Multiplayer) – FPS
Deus Ex: Mankind Divided (DX11) – FPS
Dishonored 2 – FPS
Doom (Vulkan) – FPS
F1 2016 – FPS
Gears of War 4 – FPS
Project Cars – FPS
Rise of the Tomb Raider (DX11/DX12) – FPS
Shadow Warrior 2 – FPS
The Witcher 3 – FPS
Total War: Warhammer (DX11) – FPS
Watchdogs 2 – FPS

And here is how the processors stacked up:

It appears that core count is more important than frequency after all. The only exception being 6900K, which is 200 Mhz higher clocked than 6950X. Of course high-frequency still matters, and Kabylake vs Haswell is a good example (22% faster with 20% clock difference).

This minireview proves that Ryzen has not much to fear from higher-clocked Kabylake, but let's be honest, no one really knows how SMT & XFR will work for games. A mystery which can end with a big disappointment or Intel marketing crisis.

 

[Profanity]

No minimum framerates or frametimes, and stock clocks only. Oh boy.

 

[General Lee]

https://www.youtube.com/watch?v=VXtDEF-LubM

Some Korean video with Overwatch comparison. Can't make out the FPS counter, but it looks like the 1700 runs at 3.4 GHz with all cores loaded. Overwatch seems to utilize all threads quite well paired with a 1080. edit: Then again it looks like he might be also viewing several streams on the side?

 

[Coulomb_Barrier]

No minimum framerates or frametimes, and stock clocks only. Oh boy.

Minimum framerates - a figure showing a likely solitary lowest spike in framerate is not as important as average framerate I don't care what anyone says.
Frametimes - I'm sure at least one outlet will address this.
Stock clocks - I don't understand the point, a 7700K (4.5Ghz turbo to 4.8-5Ghz) actually gains less from overclocking compared to a 6900K (3.7Ghz turbo to 4.3-4.4Ghz) or 6950X (3.5Ghz turbo to 4.2-4.3Ghz) so overclocking would show the gap even larger in favour of more cores.

 

[LordOfChaos]

Any thoughts on the highlighted block?

 

[Coulomb_Barrier]

Any thoughts on the highlighted block?

Looks like the blade head of a Mach 3 razor. ?

 

[tuxfool]

Any thoughts on the highlighted block?

Why does the official die shot look like it was decapped by some lab?

 

[Nostremitus]

Any thoughts on the highlighted block?

Where the Machine Learning takes place? (SenseMI) In other words, no idea...

 

[LordOfChaos]

.

 

[Lactose_Intolerant]

Exactly. 7700K is 4.2ghz base frequency, turbo boost to 4.5ghz. 1700 is 3ghz base frequency, 3.6ghz with boost. Assuming ryzen can be overclocked to 4.5+ ghz, this should be very interesting.

Why do you assume it can overclock that high?

edit: I responded before even reading the report about only the 1700 only being able to hit 4.1 if you are lucky.

 

[pestul]

Why do you assume it can overclock that high?

I don't think 4.5GHz is going to be easy if it takes liquid nitrogen to reach 5+GHz. And it kind of makes sense for a 8core cpu.

 

[Delusibeta]

I just ran the benchmark for GTA V at 1080p on my system for the first time. I am using Kaby Lake at 4.8 GHz currently, with 32GB 3600 RAM. It is really a strange benchmark. It ran like 188FPS almost all the time, but would occasionally crater to 50 for a split second. I didn't really see a good reason for this, but almost like it was hitting the hard drive for some data that held things up, which wouldn't be very good design.

There are a lot of settings too. Who knows how they had it setup.

Gamers Nexus took GTA5 off their CPU benchmarking course because they saw major stuttering when using i5s and the game reaches 160 FPS ish. Would explain the other benchmarks if you're getting similar issues with an i7.

 

[papersleeves]

Why do you assume it can overclock that high?

edit: I responded before even reading the report about only the 1700 only being able to hit 4.1 if you are lucky.

Merely assumption based on speed of current fx series. Of course I hope it does it, but maybe it won't. Someone managed to do 4ghz on the 65w TDP 1700. I just hope we can get better results with the 95w TDP variants.

 

[RenegadeXV8]

Don't set yourself up for disappointment regarding overclocking.

Most Broadwell-E users can only manage around the 4.2-4.3GHz mark, as anything beyond that requires a substantial increase in voltage.

So there's little reason to believe that Ryzen, on an inferior 14nm process, will clock reliably beyond that as well.

 

[Nostremitus]

Don't set yourself up for disappointment regarding overclocking.

Most Broadwell-E users can only manage around the 4.2-4.3GHz mark, as anything beyond that requires a substantial increase in voltage.

So there's little reason to believe that Ryzen, on an inferior 14nm process, will clock reliably beyond that as well.

Broadwell-E is also a lot more power hungry at stock than Ryzen appears to be.

A the Boadwell 6 core is at 140W.. An 8 core Ryzen is at 95W...

Unless I missed something...

 

[LelouchZero]

Broadwell-E is also a lot more power hungry at stock than Ryzen appears to be.

A the Boadwell 6 core is at 140W.. An 8 core Ryzen is at 95W...

Unless I missed something...

It's TDP not power consumption, which is often more to do with the heat in recent years than power consumption IIRC.

We still don't know much about that if anything at all until we see some reviewers testing it.

 

[Beautiful Ninja]

Broadwell-E is also a lot more power hungry at stock than Ryzen appears to be.

A the Boadwell 6 core is at 140W.. An 8 core Ryzen is at 95W...

Unless I missed something...

When tested previously, the actual power use of the two chips was within 5W of each other. I think the Broadwell E's only hit that 140W mark under heavy AVX use, which is something Ryzen cannot do as it only offers half AVX performance compared to Intel.

 

[LelouchZero]

When tested previously, the actual power use of the two chips was within 5W of each other. I think the Broadwell E's only hit that 140W mark under heavy AVX use, which is something Ryzen cannot do as it only offers half AVX performance compared to Intel.

Wow, where have you seen power consumption? I don't think I've seen anything about it.

Heavy AVX? That stuff be trying to cook my CPU, damn. I ran Aida 64 for 1 second on a i7 4790K and saw the temperature hit 100C in an instant. I'm sure my reaction speed topped e-sports players with how quick I clicked 'stop'.

 

[ndawgdrake]

Is the reason gaming performance is better on 4c8t than 6c or 8c because of the higher clock speeds of the 4c?

 

[Nostremitus]

When tested previously, the actual power use of the two chips was within 5W of each other. I think the Broadwell E's only hit that 140W mark under heavy AVX use, which is something Ryzen cannot do as it only offers half AVX performance compared to Intel.

See,I missed something, lol.

 

[dr_rus]

Broadwell-E is also a lot more power hungry at stock than Ryzen appears to be.

A the Boadwell 6 core is at 140W.. An 8 core Ryzen is at 95W...

Unless I missed something...

Intel's TDP is the maximum the platform allows for, same for AMD most likely but AMD used to calculate it differently (as in not actual maximum but something less like a sustained peak or something). Since 140W is the peak envelope for S2011-3 which can house a 6950K with its 10 cores it's safe to assume that a 6 core one doesn't hit this peak number most of the time unless heavily overclocked.

 

[·feist·]

So... when's the OT?

Last I recall (this was a from a long time ago, though) OTs were not supposed to be posted more than 24-48 hours ahead of the on-sale date.

Someone let me know if this has changed. Either way I should have the Ryzen OT up in a few days.

When tested previously, the actual power use of the two chips was within 5W of each other. I think the Broadwell E's only hit that 140W mark under heavy AVX use, which is something Ryzen cannot do as it only offers half AVX performance compared to Intel.

Wow, where have you seen power consumption? I don't think I've seen anything about it.

Heavy AVX? That stuff be trying to cook my CPU, damn. I ran Aida 64 for 1 second on a i7 4790K and saw the temperature hit 100C in an instant. I'm sure my reaction speed topped e-sports players with how quick I clicked 'stop'.

A bit over two months ago Canard PC/CPCHardware released some figures from a Zen engineering sample.


https://twitter.com/CPCHardware/status/811883423371526144
https://www.cpchardware.com/cpc-hardware-n31-debarque-kiosque/
http://library.madeinpresse.fr/samples/MPqY2Vg2I71P-f

Core i7-6900K @ 96 Watts

AMD 2D3151A2M88E - Engineering Sample @ 93 Watts
Base clock: 3.15GHz
Turbo all core: 3.3GHz
Turbo one core: 3.5GHz

 

[longdi]

I dont know how accurate hwinfo.exe is, but my overclocked 5960x at 4.5ghz, shows 200w processor power use when running aida64. This is at 1.19v only

I guess tdp/power really spike when you past the threshold

 

[tuxfool]

I dont know how accurate hwinfo.exe is, but my overclocked 5960x at 4.5ghz, shows 200w processor power use when running aida64. This is at 1.19v only

I guess tdp/power really spike when you past the threshold

Is it really power usage or the current package TDP?

I know for sure the latter is reported, I didn't know the former was tracked.

 

[longdi]

Yeap it is probably the package tdp, I forgot the exact term

 

[papersleeves]

Interesting gameplay comparison.

GTAV, max setting. 7700k vs 1700X.

https://youtu.be/LdqIsqwDRig

Edit: This might be fake.

 

[Nostremitus]

Interesting gameplay comparison.

GTAV, max setting. 7700k vs 1700X.

https://youtu.be/LdqIsqwDRig

Edit: This might be fake.

Was about to ask about it's legitimacy.

 

[Coulomb_Barrier]

Is the reason gaming performance is better on 4c8t than 6c or 8c because of the higher clock speeds of the 4c?

It's the other way around with newer games. More cores = better performance. See my post above. 4 cores 8 threads is the bare minimum anyone should buy in 2017.

 

[Mr Swine]

Interesting gameplay comparison.

GTAV, max setting. 7700k vs 1700X.

https://youtu.be/LdqIsqwDRig

Edit: This might be fake.

It looks very fake tbh. But it's only a few days left before we get to see real benchmarks so...

 

[Tomasoares]

I hope Ryzen gets to perform smooth 60 fps on games with 1080p

 

[Vipu]

I hope Ryzen gets to perform smooth 60 fps on games with 1080p

That would be terrible flop if it didnt, doesnt any cpu from last few years do that?

 

[Scotch]

I hope Ryzen gets to perform smooth 60 fps on games with 1080p

Setting your expectations low, huh

 

[Paragon]

That would be terrible flop if it didnt, doesnt any cpu from last few years do that?

Not at all. Even a 7700K struggles with some games.
The question is whether Ryzen's additional cores make up for its lower per-core performance in those games, and how it handles other games which don't scale well beyond 4 cores. (not well from looks of it, based on that GTA benchmark)

 

[Renekton]

Setting your expectations low, huh

Because lower resolution means higher chance of CPU limited ~~