The right angle
It seems that all Ryzen processors have the same OC limits from the fab…does that make this CPU the best?
While many in the media and enthusiast communities are still trying to fully grasp the importance and impact of the recent AMD Ryzen 7 processor release, I have been trying to complete my review of the 1700X and 1700 processors, in between testing the upcoming GeForce GTX 1080 Ti and preparing for more hardware to show up at the offices very soon. There is still much to learn and understand about the first new architecture from AMD in nearly a decade, including analysis of the memory hierarchy, power consumption, overclocking, gaming performance, etc.
During my Ryzen 7 1700 testing, I went through some overclocking evaluation and thought the results might be worth sharing earlier than later. This quick article is just a preview of what we are working on so don’t expect to find the answers to Ryzen power management here, only a recounting of how I was able to get stellar performance from the lowest priced Ryzen part on the market today.
The system specifications for this overclocking test were identical to our original Ryzen 7 processor review.
| Test System Setup | |
| CPU | AMD Ryzen 7 1800X AMD Ryzen 7 1700X AMD Ryzen 7 1700 Intel Core i7-7700K Intel Core i5-7600K Intel Core i7-6700K Intel Core i7-6950X Intel Core i7-6900K Intel Core i7-6800K |
| Motherboard | ASUS Crosshair VI Hero (Ryzen) ASUS Prime Z270-A (Kaby Lake, Skylake) ASUS X99-Deluxe II (Broadwell-E) |
| Memory | 16GB DDR4-2400 |
| Storage | Corsair Force GS 240 SSD |
| Sound Card | On-board |
| Graphics Card | NVIDIA GeForce GTX 1080 8GB |
| Graphics Drivers | NVIDIA 378.49 |
| Power Supply | Corsair HX1000 |
| Operating System | Windows 10 Pro x64 |
Of note is that I am still utilizing the Noctua U12S cooler that AMD provided for our initial testing – all of the overclocking and temperature reporting in this story is air cooled.
First, let’s start with the motherboard. All of this testing was done on the ASUS Crosshair VI Hero with the latest 5704 BIOS installed. As I began to discover the different overclocking capabilities (BCLK adjustment, multipliers, voltage) I came across one of the ASUS presets. These presets offer pre-defined collections of settings that ASUS feels will offer simple overclocking capabilities. An option for higher BCLK existed but the one that caught my eye was straight forward – 4.0 GHz.
With the Ryzen 1700 installed, I thought I would give it a shot. Keep in mind that this processor has a base clock of 3.0 GHz, a rated maximum boost clock of 3.7 GHz, and is the only 65-watt TDP variant of the three Ryzen 7 processors released last week. Because of that, I didn’t expect the overclocking capability for it to match what the 1700X and 1800X could offer. Based on previous processor experience, when a chip is binned at a lower power draw than the rest of a family it will often have properties that make it disadvantageous for running at HIGHER power. Based on my results here, that doesn’t seem to the case.
By simply enabling that option in the ASUS UEFI and rebooting, our Ryzen 1700 processor was running at 4.0 GHz on all cores! For this piece, I won’t be going into the drudge and debate on what settings ASUS changed to get to this setting or if the voltages are overly aggressive – the point is that it just works out of the box.
I have been having conversations with AMD, ASUS and other motherboard vendors on the reporting of voltages and temperatures for Ryzen processors (in addition to even clock speed) and the answer at this point is that we just don’t know for sure. CPU-Z is reporting a voltage of 1.482v from the Crosshair VI Hero motherboard when running at 4.0 GHz but I have seen that vary downward to as low as 1.417v, a significant drop if accurate.
Click to Enlarge
Using the latest beta version of AIDA64 to try to monitor temperatures, the Ryzen processor seems to be running quite cool, only hitting 73C under a full load. Yes, the fan on the cooler is ramping up noticeably, but this result still impresses. Ryzen Master, AMD’s overclocking utility, shows an instantaneous result of 68C but that fluctuates often throughout testing. At idle we appear to be somewhere in the ~40C range, though again the variance over time leaves me to question the accuracy of the reporting. If all of this data is true, then Ryzen 7 runs hot at idle but surprisingly cool when overclocked.
What do we see in terms of performance improvements with this overclock on the lowest cost Ryzen processor? Without spoiling our upcoming 1700X/1700 review, here are a few of the more interesting results.
Cinebench R15 continues to be one of the more important benchmarks for our comparisons and the overclocked Ryzen 7 1700 does quite well, able to match the single threaded performance of the 1800X CPU, as you would expect. (The Ryzen 7 1800X has a peak clock of 4.0 GHz.) It still falls well behind that Core i7-7700K but is able to creep ahead of the Broadwell-E processors. For the multi-threaded result, the overclocked part outperforms the stock settings on the Ryzen 7 1800X (though of course you’d be able to overclock it as well), blows by the Core i7-6900K, and nearly catches the 10-core 6950X!
In Handbrake, a real-world example of multi-threaded goodness, the overclocked Ryzen 7 1700 is able to oust every other CPU in our testing at stock settings. For $329, and a little bit of trust in overclocking, this CPU will outperform the $1500+ Core i7-6950X.
For those single threaded workloads, however, even a 4.0 GHz overclock can’t bridge the gap between Ryzen and Intel’s Core architectures. In this MP3 encoding workload, the 1700 in its overclocked state is the fastest Ryzen processor of the group but is still slower than the entire lineup of Intel processors.
Obviously one of the big discussion points around Ryzen has been on gaming performance, particularly at 1080p. I’m not going to waste time arguing with anyone about the validity of testing games at 1080p (43% of Steam users run at 1080p, only 1.81% at 2560×1440 and 0.69% at 4K) but let’s follow up from our initial gaming results from my Ryzen review and see how the overclocked 1700 performs.
In all three cases, the performance of the Ryzen 7 1700 in its overclocked state is higher than the Ryzen 7 1800X at stock, though the gains are minimal. (It’s under 2% on Rise of the Tomb Raider.) These changes are nowhere near the necessary deltas required to make up for the gaps presented both by the Core i7-7700K Kaby Lake CPU or the Broadwell-E based parts.
One more thing: power consumption SKY ROCKETS when you take this Ryzen 7 1700 to 4.0 GHz through the ASUS preset!
At stock settings, the Ryzen 7 1700 system draws 108 watts but when overclocked, that peaks at 214 watts! That’s a gain of 106 watts over the 65-watt TDP that the part is rated at. Even compared to the Ryzen 7 1800X at stock settings, a 95 watt rated processor, the overclocked CPU pulls 55 watts more power. That kind of increase in power draw is one of the main reasons I doubt the temperature readings from AIDA64 and the BIOS/UEFI itself. But our system was stable at these speeds and settings for entire gaming and rendering sessions.
This kind of power consumption level is a product of the high voltage ASUS has the CPU running at. But, if this is the voltage requried to meet 4.0 GHz on all cores, then the results are an important data point for Ryzen going forward.
It seems likely that we are seeing the limits of the Zen design on the Global Foundries process technology and it explains why even the best case scenarios with standard air or water cooling are showing limits of 4.1-4.2 GHz on Ryzen CPUs.
Closing Thoughts
I still have a lot of questions to be answered surrounding the current performance and infrastructure for AMD Ryzen CPUs, but my time overclocking with the Ryzen 7 1700 has given me reason to believe AMD has a solid solution on its hands. For those enthusiasts and gamers that are willing to get their hands dirty with some overclocking, and that it seems all Ryzen processors perform similarly when it comes to headroom, buying the lowest priced part for your next productivity rig is an excellent option. Thanks to the ease of access to a 4.0 GHz preset from ASUS, the $329 Ryzen 7 1700 was able to match or slightly exceed the single threaded performance of the $499 1800X. For multi-threaded workloads, it offers even more of an advantage.
Yes, you could easily overclock the Ryzen 7 1800X to 4.0 GHz (or 100-200 MHz higher) but saving $170 that you could put toward a GPU, an SSD or any other purpose for likely the exact same experience is enticing. The non-gaming performance is damned impressive with this part for workloads like encoding, rendering, etc. The gaming performance at 1080p is still in question, but if you plan to game at 2560×1440 or 4K, your results should be much better.
- Ryzen 7 1800X – $499 – Amazon.com
- Ryzen 7 1700X – $399 – Amazon.com
- Ryzen 7 1700 – $329 – Amazon.com
- ASUS ROG Crosshair VI Hero – $254 – Amazon.com
- ASUS Prime X370 Pro – $169 – Amazon.com
- ASUS Prime B350-Plus – $99 – Amazon.com
- ASUS Prime B350M-A – $89 – Amazon.com
If you are planning on investing in a Ryzen 7 system any time soon, the Ryzen 7 1700 is probably your best option.
Awesome part. Thanks for the
Awesome part. Thanks for the article. Looks as if this is the one I will be building systems with for friends and family.
Efficiency looks amazing.
Efficiency looks amazing. ~25% more efficient than the best Intel chip!
Uhh..
Uhh..
Wow!! that power draw!
I an
Wow!! that power draw!
I an contemplating upgrading from my 4790k due to streaming requirements. I wonder how much higher you could overclock the 1700 with an AIO…..
If it is one of the larger
If it is one of the larger ones and you have good sp fans probably 4.1 to 4.2. Anything more is going to be tough from what I've seen online.
~ 4.0 – 4.1 is essentially
~ 4.0 – 4.1 is essentially max for Ryzen. its intensely unstable at 4.1 and sometimes at 4 depending on the bin
Honestly given the fact intel
Honestly given the fact intel has had 8/16 cpus and a 10/20 one now. I would wait to see what intel does and even what amd might do in return. Intel has been able to get away with milking their cpus for years to the point where high end nvida gpus are getting bottlenecked in some intense cpu games. AMD has allowed them to do this by not releasing a competitive cpu until now. 8/16 cpus should of been standard in consumer/gaming intel cpus by now. The i7 5960x was release in 2014 (3 years ago costing 1000$US roughly on 2011 v3 board so everything is more expensive). The i7 6700k should be 8/16 or at the very least 6/12 not 4/8. The i7 7700 definitely should’ve been 8/16. Given AMD Ryzen is 8/16 and are all mainstream level cpus Intel is going to be forced to release some 8/16 mainstreams or better of their own. You never know how these will compare to AMD Ryzen and you never know if AMD has something they holding onto in case of that so given you have an i7 4790k I would not be so quick to pick up a ryzen cpu just yet. Give it time say end of year. That is if you can wait or not. Personally I have the i7 4790k and OC to 4.6ghz and its perfect for my gaming needs right now. I may upgrade to a amd ryzen but I am definitely waiting to see what happens on both sides before I do. Srry for long post.
…yup .. straight out of the
…yup .. straight out of the box the R7 1700 clocks straight up to 4.0 Ghz with no effort … only had the stock cooler so did not try harder or for too long … but it wants to go there.
The beast is out of the box and the Intel tax on my wallet is gone for now.
ASUS 370 PRO MB
32GB GSkill F4-3200C16Q
NVIDIA 1080
Well, I don’t get your point.
Well, I don’t get your point. The “Intel Tax” on the 4790k at launch was $340. The launch price of the 1700 is $329.
So… You sure that ‘tax’ is gone?
Yes…
Because we’re
Yes…
Because we’re comparing an 8C 16T CPU (1700) to an 8C 16T CPU like the 6900K.
That’s $329 vs. a $1,050 CPU.
That’s the Intel tax right there.
And x99 boards are $$$
And x99 boards are $$$
My MSI X99A Raider was $159
My MSI X99A Raider was $159 in November of 2016. Not all that expensive really…
No one buys the 6900k. That’s
No one buys the 6900k. That’s just a joke to even talk about.
We compare the *actually purchased CPU here*
The dude said he bought a 4790k and is going to ‘stop paying the intel tax’. But he paid the same price for his 4790k as he did the Ryzen 1700.
The point is really easy: He’s paying the same price for a CPU.
And I get it, if you keep comparing costs to what happened to be the price in the past, it looks great. The 6900k had no rival, now it does, so it’s price has already come down. But who buys the 6900k? The purchase rates of that CPU are extremely low.
I fully expect the 1700 to be a sweet spot for content creators, until Intel readies another salvo. But let’s be honest, the 7700k is the direct competition here. You either go 8 core 1700 or 4 core 7700k. They both have their clear advantages and they cost nearly identical.
Surely the “AMD Tax” is anyone buying a 1700x or 1800x is paying for a better binned CPU. You want to complain, ask why AMD released 3 versions of the same CPU.
You and the others are
You and the others are arguing in circles. The whole reason nobody buys the 6900k is the price. If Intel had offered it at $500, a lot more people would own one. Even with the recent price drops, buyers are paying an awful lot for the 6900K for marginal gains over an R7 1800X.
I agree with the rest of your post statements, though.
This.
This.
No one buys the 6900k because
No one buys the 6900k because the 7700k stomps it.
“I fully expect the 1700 to
“I fully expect the 1700 to be a sweet spot for content creators, until Intel readies another salvo. But let’s be honest, the 7700k is the direct competition here. You either go 8 core 1700 or 4 core 7700k. They both have their clear advantages and they cost nearly identical.”
That is kind of backwards. The current Ryzen line-up competes very well with the high core count Intel chips. In fact, given the performance differences in games vs. more threaded content creation workloads, I would say that Ryzen competes better with the 6900k than it does with 7700k performance-wise. The fact that the price is so different doesn’t mean that they don’t compete. Although, the only people who would buy the Intel parts are those that don’t care about price. For anyone who cares about price in the slightest, the AMD parts are the obvious choice.
Why do Intel’s parts on the
Why do Intel’s parts on the top end “K” SKUs have any unnecessary extra overclocking headroom to begin with. Is Intel maybe under binning its “K” series SKUs! Intel still has a sizable lead in the 14nm tweaking/processes maturity stage! So Intel has more room to under bin its K series SKUs and still retain its higher relative base and boost clocks than AMD can on AMD’s GF/Samsung 14nm process.
The goal of the overclocker is to purchase the lower binned part at a cost savings and manually overclock the part to make the part perform more like the non overclocked top binned part.
This K series branding from Intel is a marketing driven design result of the classic market monopoly that has no competitive market peer to drive any innovation or improvement of the market monopoly’s product lines.
So Intel is/has been in the product segmentation stage in order to obtain more revenues from the creation of an ever increasing line of offerings that only offer the smallest amounts of incremental performance over the lower tiered offerings. AMD has always offered standard unlocked parts while Intel has its rather large and product segmented group of offerings priced accordingly top to bottom to extract the maximum revenue streams at the highest margins.
The Ryzen 7, 3 initial offerings are to market at higher clocks than where originally indicated by AMD and even better IPC improvement metrics(52% better IPC over the initial 40%) than was originally stated by AMD and still you find fault with some lower binned overclocked parts being able to reach the level of a top binned parts that are not overclocked.
You know that that is the golden goal of all real overclocking is to get the lower binned part(Overclocked) To behave like the top binned part(NON Overclocked) and get the part’s end user a great deal in the process with a little effort.
No one really wants a top end/top binned part at that high cost to have any overclocking headroom! As for the money paid for the top end/top binned part, it damn well better have base/boost clocks set to the highest clock speeds possible out of the box with no overclocking headroom to speak of!
Since you lack the brain
Since you lack the brain power to figure out this obvious concept on your own, i’ll hold your hand and attempt to break this down for you little man nate….He is NOT paying the same price for a cpu, simpleton. In the case of the Ryzen 7, he is getting a cpu with performance on par with that of an Intel chip that costs three times more money RIGHT NOW. When he bought his old Intel chip, he was not getting something with performance on par with something three times higher. This has already been painstakingly explained to you using small and simple words for your benefit. This is about cost to performance ratio, dummy. According to you, I’m buying the same computer as my Commodore 64 which with floppy drive and monitor costs as much as a budget Ryzen 1700 build today.
Just don’t mix oranges and
Just don’t mix oranges and apples. Ryzen 7 competes against Intel’s HEDT series (6900K etc), price 2-3 times lower. Ryzen 5 series will compete against Intel’s desktop processors like 7700K. Still you get two cores/four threads more than 7700K but 2/3 of the price. Ryzen 3 will have even less cores, stands against i3 and prices drops accordingly. In all cases the overall cost-performance ratio is on AMD’s side.
If sides were reversed AMD
If sides were reversed AMD would be doing the same thing. Intel gets away with doing things like this because AMD is always so behind. Ryzen should’ve released 2-3 years ago given the fact intel had release an 8/16 cpu 3 years ago. What I mean is if Intel was the one behind so bad it was losing customers they would have the lower prices when they finally release a cpu that is competitive enough. The same shit happens with AMD and Nvidia. AMD will charge a ridiculous price for something if they can like with the R9 295×2 or Radeon Pro Duo for example release date price of 1500$ US or the R9 290x 750 US$ or the Radeon 7990 1000$ as you can see AMD is no stranger to charging a fortune when it can.
…the 4790k is a 4 core chip
…the 4790k is a 4 core chip .. methinks you have put your apple in the orange bin 😉 .. ( no hate )
As there were reports that
As there were reports that SMT/HT on R7 is causing problems, and that disabling said virtual threads improves gaming performance. Any plans to visit that aspect in a separate article?
Some sources say that’s due
Some sources say that’s due to a bug of the scheduler of Windows 10
If it is bouncing threads
If it is bouncing threads between core clusters, then I could see that being an issue for the AMD parts. Intel doesn’t have separate core clusters. The 4 core Zen based parts will only be a single cluster, so if that is the issue, it will not be a problem for those parts.
This thermally
This thermally targeted/binned for 65 watt usage 1700 part will need to be compared to some other 1700 samples. But looking at your tested part’s initial performance metrics maybe one of the best things to do with all the 1700 parts in general is over spec the cooler. Doing this to maybe to keep the thermals/temps lower and any thermally related deficiencies/effects reduced that may have caused the part to be binned to the lowest in the Ryzen 7 series offerings.
Overcool it so any thermal transistor leakage is never allowed to get in a runaway thermal state through negative feedback of any temperature/leakage related issues on the lower binned 1700 SKUs that are overclocked. I’m sure there will be some 1700 samples in the population that failed other tests in the binning regimens and thus perform well with thermals but have failed one of the other binning tests and your part may be one. There are variances in the diffusion process that make some transistors leak more than others if the QA/QC is not maintained on the diffusion lines, so extra cooling never hurts in the first place.
The 1700 may be that sleeper hit in the bargain market for overclockers to focus on rather than any top binned parts that really should have not so much overclocking room to begin with. As if I paying top dollar for the top binned SKUs, I want the part to have the highest base/boost clocks out of the box. It sure helps that AMD’s is using the better thermal transfer solution for all its SKUs also so that just makes things all the better for overclocking on AMD’s Ryzen 7 1700 SKUs. No crappy thermal paste on the Ryzen parts.
Windows has thread
Windows has thread scheduling/placement problems with Ryzen atm. Might want to wait a few weeks and see what happens.
I saw this video and was
I saw this video and was wondering the validity of what he was saying.
https://youtu.be/40h4skxDkh4
Any thoughts on how windows 10 handles the core scheduling for Ryzen?
Also any thoughts about performance issues on the Asus board versus other manufacturers, there has been mention that the current gigabyte board performs better due to better handling in UEFI.
I figured I should ask because you mentioned a bios version.
I like to know if by just
I like to know if by just raising the turbo boost clk alone so to increase the single thread IPC without stressing all cores work? I did this on my Fx8370 and FX8310 so it matters when needed for apps like 3d cad and will not waste energy having it in all cores
I have the 1700 (non-x), and
I have the 1700 (non-x), and find that I can overclock to 3.9GHz.
In my BIOS (ASUS B350 Prime, third or so release of BIOS) you can
easily set this. But there is a loss of voltage (to stock 1.18V)
during Windows10 Boot. This means, I can only set BIOS at 3.7GHz.
Use Ryzen Master after Win10 bootup for full voltage and clock.
I also discovered a mechanical issue between AMD’s Spire cooler
and the ASUS B350 Prime backplate. Threaded hollow posts rising
above the motherboard stand too tall. You can put the Spire on
with no problem, but forget about ever taking it back off.
The problem is you can’t clear those four posts to twist the
sink off the heatspreader. Thermal goo holds well enough that
levering the sink off won’t work either. My Ryzen popped right
out of the levered down AM4 socket, fortunately with no damage.
I would never want to risk that again.
So, even though the Spire is a great sink, I suggest you spend
the extra 30$ for a clip-on sink instead.
Apologies for posting anonymously, its just a warranty thing.
Footnote: Use Ryzen Maser to
Footnote: Use Ryzen Maser to set 3.9Ghz at 1.25V
The power increase with
The power increase with OC’ing is disturbing. Hopefully these things can be OC’d without the sky-high voltages as we learn more about them. Might go with 7700 or 6800 for now and wait for Zen 2.
Well like he said in the
Well like he said in the article, that’s not a finely-tuned OC with the voltage as low as can go and still be stable – it’s the UEFI doing some auto overclocking. If the voltage can be bumped down to 1.35-1.375, even, it’ll drop power consumption a decent amount.
7700k uses even more power
7700k uses even more power when overclcoked.
1700 on stock uses less power than stock 7700k FFS lol
I bought a 1700 for this very
I bought a 1700 for this very reason: I had a feeling that out of the gate, they’d overclock to 1800X speeds, anyway. I’m afraid to go too far with mine right now since I’m using the Wraith Spire cooler. I stopped at an all-core speed of 3.7, so now my max boost speed is my actual speed. It gets a little toasty, but it didn’t need any extra juice from the B350 Tomahawk I got to go with it. Once I get the AM4 kit for my Kraken X31 (not shipping until the 15th or later, though) that I’ll get closer to 3.9-4GHz like you did.
Super cool that it just worked, though.
Serious question: isn’t the
Serious question: isn’t the Ryzen TDP more about the heat dissipation and not power draw? Meaning, you need a cooler rated for at least 95W for the 1700X and 1800X but only 65W for the 1700? Based on temps for the 1700s I’ve seen in various reviews, comapared to the 1800X, this seems to be the case. I could be missing something…
So, not really all that surprised that overall power consumption is so high when you overclock the 1700.
Power draw and heat
Power draw and heat dissipation are nearly the same thing. TDP is related to the heat being given off, but the heat generated is based on how many watts you pump into it. At stock settings, AMD claims that the XX watt CPUs should generate an *average* of XX watts under load by limiting the voltage and clockspeed. Overclocking usually involves locking the voltage in at a much higher value than it would normally be, and watts are proportional to volts according to Ohm’s Law.
Heat and temperature are not the same thing. As temperature increases, the driving force goes up, and more heat is given off until the amount of heat being given off equals the amount of heat being generated. Theoretically, a “65W cooler” could dissipate 100W or more, but it will run at a higher temperature than a “140W cooler” since the bigger cooler will reach the equilibrium point at a lower temperature.
I notice that all the overclocking articles are using high end coolers. I hope people don’t expect to achieve the same results and low temps using the stock cooler.
The Ryzen 1700 non X Part is
The Ryzen 1700 non X Part is only rated at 65 watts because it was tested at 95 watts and most likely exhibited some stability issues. So AMD created the 65 watt bin to take any of the die samples that failed any stability, clock, and other binning tests for both the 1800X binning regimens and the 1700X testing regimens that are both targeted for 95 watts stable usage. So any 1700 SKUs are made up of die samples that failed both the 1800X and 1700X binning test regimens and those die samples trickled down to the bottom 1700 bin to be branded the Ryzen 7 1700 parts.
TDP(The rating) is more about the heat dissipation as a factor of processor die stability and durability at a set TDP top operating metric more than it is about power draw in and of itself. It’s Power draw related to power leakage at a clock speed that is what creates the heat and on some die samples the Diffusion process varies. The variability in the diffusion process creates some transistors/dies that have poor efficiencies relative to other dies and those dies create more heat at the higher clock speeds relative to other dies made on other diffusion lines or even the same lines.
TDP is a rating that is the result of some statistical testing sampling done on the entire production of processor dies to allow for the proper binning of all the parts in the runs built up before the parts’ first release to market. So the processor’s maker sets a reasonable lowest TDP/best clock speeds for its best performing parts and that top bin sets the bar that is used to create the other lower bins of parts in that line of processors for that makers final RTM product. The parts’ maker does engineer for TDP in its initial designs for mostly the lowest operating temperatures that do not affected the products intended engineering goals for TDP, stability and clock speed range at the 14nm/other fabrication process size the chip was targeting in its initial design/engineering specifications.
Wait for the GF 14nm process to become more mature and there will probably be more 1700 parts that can be clocked a little better on average than these first release SKUs. Ditto for the 1700X and 1800X parts. And there is always the chance that some micro-code update and other tweaking will improve things across the entire Ryzen 7 series line of SKUs also as time passes. The Zen 2 products are scheduled for 2018 so AMD’s is now on a constant upgrade path up until the Zen micro-architecture is replaced with a newer one.
You should be using at least
You should be using at least DDR4 3200 with Ryzen. It makes a big difference.
Even with that low
Even with that low temperature, I worry about the long term consequences of such a high voltage on the chip. Is that a valid concern, or as long as you don’t melt the damn thing is it safe to assume you will get a good few years out of the chip at those sorts of voltages?
If they’re anything like
If they’re anything like Bulldozer then they could take a beating. I tried to kill my FX-8120 by raising ungodly amounts of voltage through it. 1.6v on an air cooler and almost 6 years later it still crunches the same numbers in Cinebench. Only time could truly tell you the answer of how Zen holds it’s ground.
14nm is definitely going to
14nm is definitely going to be less robust than 32nm SOI. AMD recommend a long-term maximum of 1.35v and temporary peak of 1.45v.
Thanks, it’s good that
Thanks, it’s good that they’re giving that information. Intel has this tendency to be tight lipped on what voltages will actually damage your chips, and OCing has felt a bit like rolling the dice to me. Although I buy my PC for the long term (5-7 years) so I’m perhaps more concerned about longevity than most.
That’s because Intel may be
That’s because Intel may be intentionally under binning its K series SKUs to make them appear to “Overclock” above the nominal metrics! Intel Has a sizable 14nm process maturity/tweaking advantage so Intel has enough lead in clock speeds to have that extra latitude to maybe under bin/under clock a little and still maintain its higher relative clock speed advantage over AMD’s GF fabricated Ryzen SKUs.
With Intel any Loose Voltage specification lips may sink any pre-engineered under binning for Overclocking schemes ships. Who really needs any top binned expensive part with too much over clocking headroom, it’s better to get the best base/boost clocks out of the box for any top end SKU.
The real overclocking value comes with the lower binned parts that cost much less and may still be able to be clocked as high as the pricy top binned part. Look at the Ryzen 7 1700 it’s getting up there with the 1800X. And AMD’s current Ryzen 7 1800X samples do not appear to have too much more “overclocking” headroom above 4.1-4.2 GHz without exotic cooling methods.
The High Precision Event
The High Precision Event Timer on Windows 10 can be disabled. Admin Command Prompt – Command bcdedit /deletevalue useplatformclock
Nice overclock on air!
Nice overclock on air! 😀
Power draw is a non-issue IMHO, not just because it doesn’t really matter anyway, but AMD also makes GPUs which can consume even more power than that and are engineered to work that way.
My ASUS B350 Prime BIOS is
My ASUS B350 Prime BIOS is missing any option to disable SMT.
Since this is easily dismissed as a Microsoft Win10 problem,
not holding my breath for ASUS to provide a temporary fix.
But SOMEONE with the skills should provide a temporary fix.
And I have a half-baked clueless suggestion: Dummy threads.
Consisting mostly of NOP, running at highest priority and
fixed affinity to the eight SMT cures we want to suppress.
Could do nothing loops be enough to trick the scheduler
to avoid “busy” SMT cores, while minimally competing for
cache and internal resources of the processor?
Unfortunately, this wouldn’t
Unfortunately, this wouldn’t accomplish what you’re after. There isn’t a “real” thread and a “fake” thread where you can occupy the fake one to keep the real one open. Instead, you could think of each core as being dual-ported with respect to threads. Each is equal to the other. Either thread gets the full resources of the core so long as it isn’t otherwise occupied.
NOPs would just keep the cores busy and slow down everything. At high priority, you’d basically be rendering the system essentially unusable.
The only way to accomplish what you want is for either the software to set its own affinity and use only one thread from each core or for the scheduler to be updated to prioritize underutilized cores before the second thread of a more occupied core.
Don’t dismiss NOP too
Don’t dismiss NOP too quickly, just because a CPU would.
Look to the “Zen Microarchetecture” slide at AMD, and I
doubt NOP would get past decode into the micro-op queue.
Imagine a simple do-nothing loop padded with 64 NOPs to
be fed to eight SMT virtual cores, and maybe even real
cores of the alternate CCX.
I totally pulled that count of NOPs from nowhere, not a
scientific number. Lower count of unrolled NOPs wastes
less shared cache, higher count less frequently needs to
decode a real micro-op to repeat the loop. And certainly
fits entirely in cache, not wasting external bandwidth.
Now, if you care to use task manager to assign affinity of
your benchmark, game, or application to real cores on one
CCX, that’s fine for one use. But preference won’t stick.
And that assumes the full screen app will let you switch
view to task manager without crashing.
Was hoping for a more Ron Popiel “Set it and forget it.”
IF you are wanting to thrash
IF you are wanting to thrash the cache subsystem on Zen/Ryzen then you are going to need the AMD Zen micro-arch optimization manual/s from AMD. And also the AIDA64 folks are still in beta with their benchmarking software for Zen/Ryzen.
There is a lot of new Zen/Ryzen Cache IP that is new as is AMD’s SMT implementation.
Maybe go over to Anandtech and read the entire “Ryzen: Strictly technical” forum posts there are about 18 pages of posts covering all things Ryzen related because the core’s threads are relatively equal as far as the hardware is concerned and one thread is the same as the other with respect to access to the core’s resources.
In the SMT section of their hot chips paper/presentation AMD/senior fellow states that:
“Front End Queues are round robin with
priority overrides”(1)
(1)[See page 15 of PDF]
“A NEW X86 CORE
ARCHITECTURE FOR THE NEXT
GENERATION OF COMPUTING
MIKE CLARK
SENIOR FELLOW”
http://www.hotchips.org/wp-content/uploads/hc_archives/hc28/HC28.23-Tuesday-Epub/HC28.23.90-High-Perform-Epub/HC28.23.930-X86-core-MikeClark-AMD-final_v2-28.pdf
Does serve to clarify that at
Does serve to clarify that at least at least 4 sequential
NOPs (Ryzen decodes 4 instructions at once) are required
to assure no actual work gets decoded to micro-op queue.
Why would a small sequence of do-nothings thrash a cache?
A 64 byte loop would clutter just 0.1% of 64K i-cache, and
only one way out of four that share the same lower address
bits. Needing no reference to any other cache or bus.
I’m not sure what goes on with op-cache, but NOPs don’t
create micro-ops, so only our jump back to the beginning
might cache there…
You did not go to the Ryzen:
You did not go to the Ryzen: strictly technical forum and look at the new graphic/post that shows what may be going on. Also maybe you should go read up on CPU execution pipeline technology and not focus on any assembly op code level NOPs and look instead at the execution pipeline variety of NOP(Pipeline Bubble) and how SMT helps to reduce that NOP injected into the pipeline stages(non Programmitcally done) by the scheduler/pipeline hardware to a minimum. Go to your local college/university and start reading the Microprocessor Report from the beginnings with the 4 bit processors to the 8 bit, 16 bit on up to the current designs.
And you are clueless as to how the Cache subsystems work on the modern microprocessor designs because there are lot more specialized almost microprocessors in their own right controller units on a modern CPUs that manage the cache/memory and cache hierarchies on the microprocessor designs that are used today. The cache subsystems on computing systems are complected affairs that have to manage the pre-staging of lots of code/data into the cache levels from the memory level/s below to keep the many cores fed and the latency/bandwidth issues hidden from the processor cores so work can carry on unimpeded!
So that’s where the Ryzen CCX unit(4 cores) to CCX unit(4 cores) problems are may be happening with the coherency traffic going over the Infinity Fabric a little to high. This is because the OS is not really made aware of the existence of the CCX unit construct in the software and the scheduling of work is more haphazardly dispatched to the 8 cores/16 processor threads. These workloads are in the form of software threads(in the thousands) that are multi-tasked(preemptively multi-tasked software threads) by the OS and OS dispatched to the hardware’s the available hardware processor threads on the 8 CPU cores(2 processor threads per core) across 2 CCX units.
This may be what is causing the inter-CCX unit cache coherency and data movement traffic to clog the freeway so to speak during time of high draw call activity(more draw calls at lower resolutions) in the gaming workloads or other workloads that may be affected by processor core load balancing(done in a non optimized way) with no attempt to set any core/CCX unit affinity and manage any code/data locality with respect to the CCX unit construct’s cache coherency/data/code/memory traffic/bandwidth limits and the issues that may arise from any mismanagement of resources by the OS, API or software/gaming or other.
I did read it. For the
I did read it. For the moment, you are the one not getting it.
But you do not seem clueless, and I believe in your own time
you probably will reach the right conclusion.
Instruction NOPs do not create pipeline bubble NOPs, they also
do not fill them. The empty bubbles pre-exist to be filled by
whichever SMT can take best advantage.
Linux’s scheduler has a “shield” option that windows lacks.
I think a super busy NOP loop with affinity could serve the
same shielding purpose for Win10 without doing much to slow
down an alternate SMT on the same core.
If you trust Microsoft to care about us, wait for them to
to fix the scheduler, or ASUS to add SMT to the B350 BIOS.
We can disable 4 cores to avoid program shuffling between
CCX. Or NOP those eight threads, and retain the option to
re-enable them without reboot. Not sure if infinity fabric
might be able to use the alternate CCX’s L3 to any benefit?
Could fill it with NOPs, or leave it mostly empty and see.
“Instruction NOPs do not
“Instruction NOPs do not create pipeline bubble NOPs, they also do not fill them”
Never said that, I said that the scheduler/pipeline hardware(NON Programmatically inserts the NOP/Bubble) via the hardware.
I said: “SMT helps to reduce that NOP injected into the pipeline stages(non Programmitcally done) by the scheduler/pipeline hardware to a minimum.”
If the scheduler has a second processor thread to work then that potential pipeline NOP can be replaced with work from the other thread while the first thread is stalled(usually waiting for a dependency to be resolved, or data fetch delay, or some dependent computation to complete).
But yes there are already some 4 core benchmarks being done with one only 4 cores enabled and the results anainst the 7700K from the 1800X look almost the same, if the 7700K is running at the same clock speed as the 1800X. I think the benchmarks shown where at 4.0 GHz for both systems 4 Ryzen cores(enabled) against 4 7700K cores. There are some AM4 motherboards with the BIOS options enabled to do so for Ryzen. (1) His results will have to be verified by others!
“Hello all, This is a first article I try to publish in English version. However, one thing you should know before reading this article is I’m not expert on English so about the grammar some may not correct but I hope this will help you to understand more about this article better than using Google translate or other.
Today we will test in a way that we believe many of us would like to see. We think it might be a good idea to estimate Ryzen 5’s performance by comparing Ryzen R7 1800X with the i7-7700K clock-for-clock and core-for-core. Some of you may not understand what we’re trying to do or what we’re trying to present, but stick around. It’ll be worth your while.
All tests will be done with Ryzen R7 1800X overclocked to 4.0Ghz and core/thread count reduced to 4/8 (half the number of cores and threads is disabled); and i7-7700k underclocked to 4.0Ghz.” (1)
(1)
“Core by Core, MHz by MHz with AMD RYZEN 7 1800X vs Intel Core i7-7700K [English]”
http://www.zolkorn.com/en/amd-ryzen-7-1800x-vs-intel-core-i7-7700k-mhz-by-mhz-core-by-core-en/view-all/
If a single threaded pipe
If a single threaded pipe achieves 60% utilization,
then a multi-threaded pipe might achieve 50%+50%.
But 50% is less than 60%, and it sometimes matters.
>I said that the scheduler/pipeline hardware
>(NON Programmatically inserts the NOP/Bubble)
>via the hardware.
Sonotori! Therefore programmatic NOP should be
little different than absence of another thread.
Hopefully giving us 59%+1% utilization.
Percentages are “alternative statistics” made up
on the spot, only serving to illustrate a point.
Don’t take them too seriously…
Your benchmark link with one CCX disabled seems
to indicate more gain to stop shuffling between
cores with differnt L3 than anything going on
with SMT.
I believe the same NOP trick applies to exclude
an unneeded CCX without need to reboot.
Now search “How to Create a Shortcut to Run an
Application with a Set CPU Affinity in Windows”.
This direct link might or might not work…
https://www.eightforums.com/tutorials/40339-cpu-affinity-shortcut-program-create-windows.html
Thats how I’m planning to launch those NOP loops.
Hide a batch in my startup, and forget about it…
But you could forgo making fake busywork, instead
make custom shortcuts for all your apps with the
affinity set for a single CCX and non-SMT.
I was a little worried bout programs that launch
other programs, but I observed the Valley benchmark
inherited affinity preference from its control GUI.
Now I’m not so worried…
Oh freddled gruntbuggly!
Thy
Oh freddled gruntbuggly!
Thy suppurations are to me as plerdled
gabbleblotchits on a lurgid bee.
.
.
.
Or I shall rend thee In the gobberwarts
with my blurglecruncheaon, see if I don’t!
Finally, something we
Finally, something we completely agree on!
Yes there was a lot of
Yes there was a lot of writhing in agony during those stanzas and even in-between, followed by a very unsuccessful attempt at some very sycophantic sucking up on Arthur’s part, with Ford feigning in agreement, followed not to much shortly afterward by both Author and Ford being tossed of out an air lock into the cold hard, and quite airless relatively, vacuum of space.
And I, upon reading your repetitive every minute thought process detail on the setting up of your NOP loop, was moved to pondering what a merciful thing it would be if I too where to be tossed out of an air lock into the cold hard, and quite airless relatively, vacuum of space as a form of final relief from the utter agony of reading those very NOP loop minute thought process details with which you so repetitively noted in written from.
Now that’s the kind of thread
Now that's the kind of thread what makes me proud of you all.
I assemble by hand all the
I assemble by hand all the way back to PDP8e, and own more old
books on the subject than my local library. I’ve also designed
my own (for fun, not practical) look-up table processor, so you
are jumping some very silly conclusions. I don’t know much x86.
But I could probably scribble an x86 NOP loop on a damn napkin,
if no-one else with modern programming skills takes interest.
Maybe stick a Babel fish in
Maybe stick a Babel fish in your ear and go for some Vogon poetry! That may just help with getting your point across better in translation.
Translation:
I can’t program
Translation:
I can’t program my way out of a wet paper sack with a modern
compiler or IDE. Though I’m frequently forced to use Code
Composer Studio, Labview, and whatever else they throw at me
at work. I’m just not wired to think that way. The moment
it becomes a project tree, my brain just goes out the window.
I can still write small snippets of extremely efficient
spaghetti code and assemble them by hand. On paper, just
like I did back in the 70’s and early 80’s. I once did a
nearly complete re-write of the Xerox 820’s CBIOS to jam
a floppy disk cache of 4 tracks in the saved space. I’m
talking about Xerox’s CBIOS for the CP/M operating system.
That was all done on paper, and BASIC abused to save the
data to disk. Some of it had to be on the boot track…
If the fish isn’t working, thats fine. I don’t often find
anything I’ve written to make sense, even to me.
You’re not the chap that
You’re not the chap that programmed the space invaders game into the old Burroughs TD-830 CRT terminals are you! You know the one where the code was transmitted in text format with the escape codes to tell the terminal to load the code and then there was the escape code entered to execute the game. Ah those ASCII graphics rendered alien thingys working their way down towards ones movable LASER cannon. That was some ASCII-tastic gaming fun in those early days with the click click of the 029 keypunch machines going on in the background interspersed with the Thunk Thunk Thunk sounds of the card reader and the high speed tap tap tap of the line/band printer.
But do stop with the patting of ones back and maybe focus more on the concepts with as little unneeded info as possible.
I didn’t buy into Ryzen so I
I didn’t buy into Ryzen so I could debate forums.
Now if you’ll scuze, I gots to go benchmark me
some Dwarf Fortress…
And no, I’m not THAT chap,
And no, I’m not THAT chap, though his story is cool.
I’m the other guy. You know, the one that hacked his
Trash-80’s cassette relay to interrupt 20ma current
loop at 110baud, and used it to drive ASR-33 teletype.
Did make all aformentioned clunk-bang-bell sounds.
Was just trying to say even
Was just trying to say even my primitive skills would
suffice to write a NOP loop, but you may have to wait
a while for the ink to dry.
As for the lookup table processor, its more of an ALU
table magnetized to 25nS Everspin MRAM. Has only one
very long instruction consisting of two moves and an
ALU operation.
The ALU is really strange, cause it doesn’t “test” for
zero, or carry, or any other flag. Result of any tests
were pre-determined at the time the table was built.
So instead of traditional flag outputs, it has action
flags such as “skip next”.
Abused FUZE BASIC on a Pi to flip those tables to MRAM,
somehow word “burn” doesn’t seem appropriate for MRAM.
I was originally looking at MRAM to replace the missing
core memory of my Dad’s wire wrapped PDP8 clone, but then
got sort of distracted by the opportunity to design my
own processor. I never did get that old PDP8 working…
Anandtech “Ryzen: Strictly
Anandtech “Ryzen: Strictly technical” forum read post #437(it’s on page 18) by unseenmorbidity, Today(3/9/2017) at 10:21 AM.
Fourm poster, Unseenmorbidity states:
“For the past few days, there have been talks around tech forums about Ryzen’s clock domains. Specifically about how some parts of the internal fabric of Ryzen run at memory clockspeed, or half effective transfer speed. Thanks to hardware.fr, we now have a diagram of the clock domains in Ryzen.”
Go read his assessment, it’s very Interesting!
I’m aware that either SMT
I’m aware that either SMT could be considered the real one.
Yes very equal those threads
Yes very equal those threads are but fighting it out like James T. Kirk and Gorn for Hegemony of the execution resources in that dry and desolate setting! Marquess of Queensberry Rules NOT generally adhered to, mind you, with plenty of sand to fling in the opponent’s eyes and the hitting below the belt sorts of things that comes with the struggle for the core’s favor in resource allocations in that catch-as-catch-can competition of threads!
I wouldn’t actually mind
I wouldn’t actually mind knowing what the 1440p results look.
Testing Ryzen with DDR4
Testing Ryzen with DDR4 2400?
What’s up with that?
Did you test the Intel 8 cores with only dual channel vanilla clock speed memory? Think not.
Look up quad channel
Look up quad channel memory.
ONLY Intel 2011v3 X99 motherboards support quad channel memory.
That is, i7 X and Xeons cpus.
Trying to compare quad channel memory to dual channel memory is like comparing workstation graphics cards to gaming cards.
Or a semi truck to a pickup truck. Both are trucks right?
Z270’s are all dual channel memory systems.
B350/X370’s are all dual channel memory systems