AMD's Ryzen CPUs (Ryzen/TR/Epyc) & Vega/Polaris/Navi GPUs

An interesting take on the testing methodology fixation with 1080p testing (to stress CPU) vs 1440p / 4k (how "most" people will be using it), with a historical empirical take on the reasoning and how it hasn't actually proven out over time as thought it will moving forward - faster GPU will extend the Intel advantage as bottleneck moves back to GPU - when in actuality the gap shrinks.

It's going to be anyone's guess how the future GPU's will "tax" the CPU, so we can't really be sure we need to create an artificial test environment as the only acceptance testing for the Ryzen CPU's.

I do like that the 1700 might just as well be purchased as the 1800x given the OC performance of the 1700 compared to the OC performance of the 1800x - this from another video posted earlier.

 

No, I know what I said, and it is what I meant. There is zero point to getting an octocore CPU for the purpose of GAMING alone. The vast majority of games that use over 4 cores top out at 6 cores, even if they use hyperthreading. In this case, it's just wasting money for no benefit.

If you're an overclocker, then 8-cores makes sense. If you're someone who does things other than game, like stream or use CPU-based recording or do lots of video rendering or run programs like Blender, CFD, Maya, etc? Then 8-cores make sense. But for someone who's just gaming, buying an 8-core would be more akin to say... buying a 3TB HDD then formatting it using MBR, which means the max partition size is 2TB.

Buying the best is fantastic if you can, but there does come a point where money is wasted. Like I know someone who has a 3970X and ran it at stock for 3 years then changed out the system. Then they got a 6950X and they're running it at stock, and using it just to play video games on. I could play games better than that with this laptop if I tune it and OC the CPU. A 6800K would literally perform the same to that person. It's just wasting hardware and money at this point for them.

Did I mention that they also have a streaming PC? With a mainstream intel quadcore, I think the 6700K (also at stock)? They're using the 10-core at stock to game but the quadcore with higher speed to stream. It's one MASSIVE waste of hardware.

As for the firestrike, does the combined test use all the CPU? I don't think it does, and I think the relatively poor single thread performance and lower speeds would produce a worse combined score; but even then... that's rather bad still. I think maybe chipset drivers would really help them.

The point as I stated above, is less wasted cash. Though the arguement could be made "better to have than to not" but it really won't make a difference in the majority of tasks. If you wouldn't buy a 6900K over a 6800K for your usage scenarios, it's basically the same scenario.

 

That can be true under the right circumstances, but avoiding something until it is working correctly is equally smart. Clearly, even though those are the top scores for Ryzen 1800X that I linked, something isn't right. It may work itself and turn out amazing, but waiting for execution instead of taking it on faith that it will happen is the smartest approach and the safest way to evade regrets.

 

chew* @xtremesystems has been grinding away

 

For reference, 24k made by my system.

http://www.3dmark.com/fs/11909005

 

Yup.

Something is wrong for the Ryzen chips. As far as I can tell, IPC is a little better than Haswell (which is why @tgipier's Haswell-E firestrike benchmark didn't do much better in the Physics test in Firestrike even though the RAM is better and it's 400MHz faster, accounting also for the slight perf hit that W7 has in Firestrike Physics) but the combined score... it makes no sense. I think chipset drivers are needed. Buuuut that'll come later.

However my original point about the 8-core does stand. Until using 6 cores and 8 cores is the absolute norm for games (and it will be about 5 or 6 years from now; the high end engines will be using it I'm sure) I'm gonna say you only need 6 cores at most for a top notch gaming experience. See on Intel you could make the arguement that better RAM and higher clockspeeds can prevail in more single-threaded workloads, yes, but in this case, you're not even looking at the unlocked i5s in price range (checking Ryzen 5), and if you're looking at locked i5s then the 6-core 12-thread chips are just going to be better hands down no rematches. Even if you compare an unlocked i5, you'd need to be clocking the 5GHz range for more multi-thread-heavy games to compete with a ~4GHz hexacore.

So, when Ryzen works well? Ryzen 5 looks to be my go-to. If it can OC (safely for 24/7 use) to over the "standard" 3.9GHz that the 8-core chips arrive at, that's even more of a benefit. But as-is, I don't see much use for the flagship 1800X chip. If the 1700 chip can OC to 3.9GHz the same way and use the same RAM and has the same IPC and everything, then that's the only one worth a buy. Because at the $500 price point and the 3.9-4GHz OC limit, an entry level hexacore Skylake-E (later this year) clocked to ~4.7GHz with quad channel 3200MHz 14-14-14-34 RAM (reduces CPU usage quite a bit compared to the 3000MHz 18-18-18 dual channel limit I've seen Ryzen has to use if they want over 2666MHz bootable) might compete with, or beat, the Ryzen 8-core in real-world workloads. CPU tests don't affect RAM too much (like Firestrike Physics; though that LOVES cache like we saw with L4 128MB DRAM in Broadwell chips) but more real-world usage (rendering, recording, streaming, etc) takes RAM into account heavily. I think the net gain will put them on par, for slightly less cash on the intel side (assuming the entry-level hexacore comes out to about $350 considering the recent price drops intel put out) and the gaming performance on the Skylake-E at those speeds will be HEAVILY improved in CPU-bound scenarios (considering that games will only use 2-6 cores from either CPU).

 

@hmscott Love the beginning! Funny as sh*t.

 

Microcode optimizations. And obviously system / software level tweaks and optimizations.

 

honestly with all the delays this should have been ironed out long ago... but oh well.

 

Silicon Lottery has the Ryzen sku's up, pretty tight spread for speeds:
https://siliconlottery.com/collections/pga-1331

Interesting what the % is at each bin:
As of 3/6/17, the top 20% of 1800Xs were able to hit 4.1GHz or greater.
As of 3/6/17, the top 67% of 1800Xs were able to hit 4.0GHz or greater.
As of 3/6/17, the top 97% of 1800Xs were able to hit 3.9GHz or greater. (3% under / fail ?)

As of 3/6/17, the top 33% of 1700Xs were able to hit 4.0GHz or greater.
As of 3/6/17, the top 77% of 1700Xs were able to hit 3.9GHz or greater.
As of 3/6/17, 100% of 1700Xs were able to hit 3.8GHz or greater.

As of 3/6/17, the top 23% of 1700s were able to hit 4.0GHz or greater.
As of 3/6/17, the top 70% of 1700s were able to hit 3.9GHz or greater.
As of 3/6/17, the top 93% of 1700s were able to hit 3.8GHz or greater. (7% under / fail ?)

 

3 Wheel-barrows of cash not withstanding, it's really tough to be a Ryzen reviewer right now

 

Gaming follow-up to the Build video + Orange Pastel liquid swap...

Does Ryzen 7 REALLY suck for gamers?

 

Doesnt matter.
Here is my take. People are blowing the whole gaming performance thing out of proportions. Like wayyy too much and jumping on the hate train. How low is the performance in games? 10%? 15%? that barely translates to 5-10 fps at most.
I dont see people going crazy over the fact the 7700k performs way better than the 6900k in games? Specially at the price the 6900k retails at.

 

Oh yeah, some are taking them way out of proportion. Though it has disappointed me greatly thus far.

Sent from my OnePlus 1 using a coconut

 

Out of curiosity, can you think of a specific game in which a 6 or 8 core processor from either Intel or AMD outperforms the 7700K? My impression is that right now, the optimal CPU for gaming is a quad-core clocked at high frequencies.

 

Frostbite 3 games
Cryengine 3 games
Dying Light
Deus Ex: Mankind Divided
ROTTR in geothermal
Witcher 3 in that one town that people use as a benchmark
Black Ops 3

Those should all do it when aiming for higher FPS.

Though, ideally, getting a 6-core to at least near the speed of the quad would be best. I'd take a 5GHz 4c/8t over a 4GHz 6c/12t of similar IPC if all I was doing was gaming.

But I would NOT take a 5GHz 4c/4t over a 4GHz 6c/12t for any real reason.

 

chew* @xtremesystems was allowed from HQ to share a beta bios for those that own the Gigabyte AX370 Gaming 5 AMD AM4 Motherboard.

F5c approved. Gaming 5 only....

https://drive.google.com/file/d/0B6A...ew?usp=sharing

 

So, the kind I enjoy most and play most.

 

4.3GHz was impressive back in 2011-2012. Today, not so much. That should be the non-turbo core speed. To see 3.9GHz as the stock turbo clock is pretty sad. That is wimpy turdbook territory in a desktop package. Clock is king, and adding extra cores/threads that work efficiently just makes it extra juicy.

A run-away train traveling 50 MPH is very dangerous. The same run-away train traveling 100 MPH is devastating and will leave a much greater wake of death and destruction.

 

Black Desert Online I believe uses 6 threads as well

As for clockspeed, I could care less about it, what I can do with it, is far more important really. But as a Benchmark enthusiast I can see why chasing numbers is fun, I gave that up a while ago.

Function over form for me.

 

What makes it any different from the [email protected] Turbo? Just curious. The Intel 8 cores seem to have a little more headroom, but they still top out at around ~4.5 on air.
http://hwbot.org/hardware/processor/core_i7_6900k/

 

I agree. Unfortunately I do not have the deep pockets some others have. I chased numbers also, back in the day, but trying to keep up with the "Jones", did not sit well with my wife. Unhappy wife=unhappy life

Besides, I have fun doing other things. Yes, that is me@45 Years old

 

Well I think a lot more games use Cryengine and Frostbite than FPS ones (Dragon Age: Inquisition, Mass Effect Andromeda, newer Need for Speed games, the MMO AION, etc) but lots of FPS titles do use them, yes.

Clockspeed DOES matter here, though. Granted, stock v stock, the current intel -E CPU line has similar stock speeds and does not overclock as well as previous lines (1st through 3rd gens) and VERY likely not as much as upcoming 6th gen, but either way, it matters because not everything uses all the threads. As I even pointed out above, for the same kind of cash, the 1800X chip and the upcoming 7800K (supposedly named) could butt heads due to a difference in overclocking and memory usage, and if they manage to butt heads in the multi-threaded game, then in gaming the 7800K would oblivion it in most scenarios (especially ones that use 4 cores or less).

Function over form is indeed the way I go, but figuring out something's function is also important. These Ryzen 7 chips will not ever be a great choice for a pure gamer or an enthusiast, all things considered. They'll be a good choice for someone who needs good CPU power on a budget and is certain they'll never overclock, but that market is smaller than people think. And that makes me rather annoyed at AMD, because they had a golden opportunity.

 

One word: overclocking

That's what I am waiting to see from Ryzen. AMD GPUs suck at overclocking. Historically, AMD CPUs have been good at overclocking, but were inefficient and produced bad results at any clock speed. Looks like they are doing better at efficiency now with Ryzen, but they need to be good at overclocking.

I do not run my CPUs stock, so what 6900K does at stock is totally irrelevant, just as what Ryzen does at stock is irrelevant to me. I don't buy CPUs that are not good at overclocking. At least, not on purpose I don't. I have no interest in CPUs or GPUs that suck at overclocking and do not allow me to place much higher than normal in overclocked benchmark rankings. I can run them at slower clocks if I want to as well. It's the overclocking limitations part that I am unable to forgive.

 

AMD Naples server processor: More cores, bandwidth, memory than Intel.

https://arstechnica.com/information...essor-more-cores-bandwidth-memory-than-intel/

 

That's why I keep disregarding Broadwell-E though. It barely OCs. 4.2GHz is considered a good OC and isn't even a "guarantee" for a 6900K like 4.7 is for a 6700K or 4.8 is for a 7700K. Haswell-E is not as bad, but still bad. I think @TBoneSan can't get 4.4GHz out of his 5820K even?

Skylake-E should change that heavily though... while using less power and generating less heat

Sent from my OnePlus 1 using a coconut

 

All of the bickering and drama reminds me of politics too much. It smacks of fanboism rather than objective analysis of results. I think we should just let the chips fall where they may. AMD has made great progress from where they were, but they may not end up displacing Intel as the performance king. It's too early to tell though. I'm willing to give them another 60 to 90 days to iron out the bugs and optimize drivers before deciding who the winner is.

 

Well both my 6850k's could do 4.5Ghz , at different voltages obviously. But the one i kept can even do 4.8ghz. I have yet to own a 6 core that hasn't been able to at least do 4.5Ghz. Even my super old 3930k was easily doing 4.8Ghz under water.

 

I think der8auer took 1800X to 5.8Ghz. Thats mighty impressive.

 

I'm honestly inclined to say you got lucky, but I have no statistics. There are apparently lots of people who cannot get 4.2+ though. Even the X chip doesn't outpace the rest like in the past. I can't give someone a general guarantee like with the other chips

Sent from my OnePlus 1 using a coconut

 

Yes it is. And, I am looking for more evidence of that. I hope it is not a fluke. What is the highest stable/benchable clock? Has der8auer published any of that, or was it only a CPU-Z validation? Let's hope it's not a one-off example of a rare chip.

If 1800X could easily run 5.0GHz with stock voltage like both of my 7700K can, and do that stable in benching and gaming, I'm sure it would make mincemeat of any Intel i7 Quad CPU with double the cores and threads.

I think people that are focusing on 95W TDP are missing the point. The TDP is not as important as the CPU being able to use as much or as little as it needs to get the job done. Back to my old argument that doing more with more is more impressive and important than doing more with less. More, more, more, more, more... please. Don't cut any corners or try to do something silly and ordinary (and boring) like settling for similar results using less. Just give us more in mass quantities. If it takes higher TDP, so be it. Who cares? Using lower TDP and lower voltage doesn't win any battles for the performance crown. If they are going to beat Intel, doing that is job #1. If they don't, they will remain in second place. Since there are only two contestants, second place is last place.

 

That was under LN2. You're not hitting close to 5GHz under air/water. It was also using over 1.9v.

The highest bench I've seen was 5.368 or so, in Cinebench. No report on the voltage that time though.

Sent from my OnePlus 1 using a coconut

 

That sounds pretty grim. Maybe I was too quick to acknowledge good progress. I'm looking for something to be excited about. Seems like that's not it.

 

I think 5.8ghz was validation only AFAIK and that too was under LN 2. Someone else set a Cinebench record @ 5.36ghz under LN2 as well.

But IMHO expecting anything with 6 Cores to run at 5Ghz is pretty much impractical, let alone an 8 Core. We wont be able to do that anytime soon without breaking the laws of physics or without some sort of breakthrough in processor technology. And that is one of the reasons why processors like the 7700k and the 6700k exist.
Even now though the the 1800X is performing better than any of the Intel 4 Cores or 6 Cores at productivity applications @ 4Ghz.

End of the day we needed something like this. Heck i know for a fact that next time someone want me to help assemble a at home productivity desktop, im going to slap an 1800X in there.

For example -

 

On the other hand though the 5.3-something clock took world record in Cinebench from a 5960X @ 6GHz... so it might still be benchable. But for air/water it does not seem so.

Sent from my OnePlus 1 using a coconut

 

So, let me explain the cache problem. It takes two forms. One person ran a core path utility showing that, instead of using the 16MB L3 as a universally shared cache, it gives each thread its own 16MB of L3 cache. If true, Windows is basically barraging the L3 cache, constantly kicking out info in line, which would be devastating to performance. Part 2 involved communication between CCXs. If the data is loaded in one complex, but is needed in the other, instead of staying on the CPU, it takes a round trip visit to the ram. Considering the low ram speeds, etc., this also takes a heavy toll. Combined, this means AMD needs to work on the microcode and M$ **** the bed on cache appropriations on its OS.

But the level of chicanery is not over, as in comes the MB manufacturers. They expected a horrible fail of AMD Ryzen. This is why they said they spent 5-6 weeks developing the firmware. They also are running out of MBs for Ryzen. AMD held back at release to send units to wherever ran out, meaning since release, they have adopted more to different retailers based on their demand so that there was a steady supply. MB manufacturers, on the other hand, under-produced, making supplies tighten, bottlenecking the ability to buy a Ryzen platform. Taken together, the MB manufacturers had no faith in AMD and basically said **** you to them. Only after seeing larger than expected demand did you see a flurry of updates for firmware and companies start to be more attentive.

Then comes the ram manufacturers, which didn't even announce AMD specific ram until it became obvious that Ryzen was selling fairly well. Instead, they just repurposed Intel optimized products (which the new lines are still that, but contain optimized SPD timings and AMP (AMD's version of XMP) profiles).

So don't say that AMD is why it wasn't ready, as the rest of the industry gave the finger to them before release.

With that said, I still stand by my hold recommendation on purchasing a Ryzen setup until these issues are dealt with. During this interim, software companies will also work on optimizing their code, which many game partners have not yet released their optimized codes yet. So, wanted to give context and spread that hate toward AMD to others that also richly deserve it!

Sent from my SM-G900P using Tapatalk

 

All I am looking for is for AMD to soundly beat Intel at their own game, stock and overclocked. I overclock everything for normal everyday use. Pretty simple really. Not looking to give partial credit or get excited about caveats qualified by yeah-butt scenarios. Just give me something that is better than the Intel equivalent no matter what I want to do with it. If it wins at everything, it's a winner.

 

Seems they are really good at screwing things up. We can always count on the Redmond Mafia for a mess.

I agree AMD has an uphill battle. That is also why I agree with your approach to wait and see what happens. It would be a shame for AMD to fail again due to the incompetence of others, but that is entirely possible. Intel and Micro$loth have had a headlock on the industry for a long time. There is a reason I call Micro$lop the Redmond Mafia and Nazis all of the time. It is fitting based on their command and control approach to things. NVIDIA sort of has a strangle hold on us, but they have been winning by default due to a no-show year over year by their only competitor, to a lesser degree due to innovation and excellence.

 

They don't even let you switch the primary graphics adapter without a reboot... this is why MUX switches are so convoluted. Mac OS has been doing that for years to my knowledge.

Sent from my OnePlus 1 using a coconut

 

Aren't they doing that already
Specially in benchmarks that actually matter.

 

The Nazis at Micro$loth are legends in their own minds, and in the minds of the Kool-Aid drinkers that believe they can do no wrong. It's easy to call yourself the best when you're the only player. It takes a hero to always do the right thing. They even stoop to digital genocide to try to force us into using their newest garbage by killing their own products that worked better. Why? Because they make more money. These clowns are not heroic by any stretch of the imagination.

Nope, I don't think so. Everything means everything. If an 8C/16T powered desktop gets beaten or even matched in 3DMark benchmarks by my monster laptop, they are not winning at the things that matter most to me.

 

They are lazy as Hell! AMD, without optimizations for it and drivers for Win10, gets like 16% more performance from Windows 7!

I agree, but benchmarkers, because of low frequency, never checked what the 8-core could achieve (6900K), even though the IPC was improved over the 5960X. So, getting over the frequency part is hard for many. Not only that, AMD did not leave OC headroom at all. The chips ramp heat above those clocks as usual with former AMD products, while being larger voltage jumps to get over the gap before you get more speed. But, to their credit, the solder technique is much better than Intel's!!! Intel soldered chips still give 6-8C better by delidding. Ryzen gives 1C difference on max temp while not just delidding, but going to bare die!



Sent from my SM-G900P using Tapatalk

 

Partly because Micro$loth is lazy, and possibly because they are so dishonest and manipulative. I cannot help but be suspicious they might try to sabotage AMD. The mere fact that anything works better on Windows 7 (almost everything in fact) says a lot. The only things that work better for Windows 10 are the things designed with an intent to make Windows 10 look better than Windows 7.

Now, that is impressive. Delidding is a cool geek thing to do, but having to do it in order to get good temps even stock is a pretty negative thing for Intel. It's good for us that delidding is effective, and very bad for Intel that we need to.

 

In case people are interested, AnandTech has a preview of Naples, Ryzen's big brother. This is a server CPU for 1 or 2 socket systems coming in the second quarter of 2017. It has 32 cores (so a 2 socket system will have 64 total) and should be very competitive with similarly priced Xeons.

 

I might actually sell my Xeons and build an AMD based workstation if the performance is better for actual productivity just like the 1800X.

 

Everyone has a critical eye of AMD. But, we must remember how closely Intel and M$ work together, the ****tier performance of later OSes generally and helping intelligence agencies generally (see today's wikileaks on CIA spying in relation to OSes), and Intel fighting for dear life on keeping Apple as a client. Also, Ryzen is being used for initial performance comparisons for Naples, which would cause the most damage to Intel's bottom line. So, the thought of purposeful collusion to "conveniently" not optimize for AMD and blame it on "growing pains" sounds like an Intel move. Plus, M$ incompetence is known world wide, so easy to do without any chance of charges for collusion!

Sent from my SM-G900P using Tapatalk

 

Well in that respect my 6850k is also pretty bad at 4.5Ghz since @Johnksss@iBUYPOWER managed to pretty much match my scores using his DM3 in single GPU mode.
The point that I'm trying to make is that there is much more to all this than we have at our disposal at the moment. And that 6+ core processors aren't really and I doubt will be, OC behemoths. End of the day what will matter is the scores right? And we are getting those right now even without the optimizations.
I personally am disregarding anything other than 3DM 12 and Vantage, because benchmarks like firestirke are heavily skewed towards GPU power.
And that the 1800X is one of the most bang for the buck processors out there for people like me, who need one for actual muktithreaded work. And over the months once the optimizations and proper codes kick in. Things will and should improve.

 

http://www.zerohedge.com/news/2017-03-07/wikileaks-hold-press-conference-vault-7-release-8am-eastern
Off topic, but notice how the article mentions the CIA using M$ Windows update to spy on everyone. Lulz!

On topic, after the bugs get smoothed out, as @bloodhawk mentioned, the performance will increase. Also, so far AMD has revised the frequency for boost on the six core to 4.0. This may continue to increase as we get closer to release...

Sent from my SM-G900P using Tapatalk

 

Yeah, somewhat agree. Where we differ is in application only. If Ryzen doesn't overclock very well and I can't overclock the crap out of it to beat not only Intel but other Ryzen owners as well, then it becomes boring. I am not interested in doing what is ordinary and get no pleasure out of a cookie cutter experience. That has been one of the issues that has made AMD GPUs so unattractive for so long and I hope it does not end up being applicable to their CPUs as well. Better performance than before is really great. Better gaming performance than before is great. But, it cannot end there and be worthy of being called excellent. Not being able to push it and accomplish something unique, that others lack the skill or knowledge to do, without a reliance on getting lucky from an abnormally good or capable chip, is where I lose interest in the product. That would be like drag racing with every car having a governor that limits their maximum speed to an equivalent value. The sport loses relevance and becomes meaningless to those that once enjoyed it.

This is the part I am watching and hoping we can have a reason to get excited. I am hoping it has a lot of headroom to do more as things are optimized.