Official AMD Ryzen Benchmarks, Reviews, Prices, and Discussion

[krumme]

BWE without heavy avx stuff does not draw 230 watt at 4.2, nah.

Sorry that is a full load test at 4.3 if you even get there. But who cares anyway. A kbl is faster in gaming.
J/k

 

[looncraz]

I don't get this part. Are you suggesting the R5 will be better optimized somehow?

Yep. Updated microcode can do wonders. Those fixes will be back-ported to Ryzen 7, of course, via AGESA code updates (BIOS), but when Ryzen 5 shows up matters should be much different...

Windows will be updated with proper support, more games will be patched, BIOS support will be improved, etc...

People who are into buying at this level are usually "nerds," so it's pretty safe to sale to us - some of us will reverse engineer the code running on the CPU and send very detailed reports.

 

[YBS1]

AMD tricked me into thinking this was faster than a 6900K. I also believed it would OC at least decently due to the advertised 4ghz stock clocks. AMD also told me that XFR was a celebrated auto overclocking feature, which convinced me that these chips were at least decent overclockers. I was shown cherry picked benchmarks that showed very high IPC despite clock speed deficits. I fell for it, got snagged by the hype train and was dragged along by my pants about 10 miles before finally falling off and smacking headfirst into the great wall of reason.
I knew it was a stupid thing to do and I admitted it when I said I was preordering. I did it partially for the thrill, knowing it was a gamble. Well, I seem to have lost that gamble.

I find it odd you feel disappointed that you preordered, I'm in the opposite camp. I kind of regret passing up the chance to preorder while Newegg still had the 1800X, I passed partly to wait on reviews and partly to see how the motherboards ended up shaking out. This would be a 3930K replacement as well. It's pretty obvious some of the results that could be viewed as disappointing are due to early teething and software issues. The remaining simply comes down to differing architectures having different strengths and weaknesses, for every bench in which the Ryzen trails by a noticeable amount there is another where it leads by such a degree. One thing is beyond doubt though, the all around price/performance champion is now a Ryzen cpu. It's the rough equivalent of a 6900K at half the price (or better). I'm more anxious at this point to see how the chips available six months from now end up clocking under high end cooling, I'm betting they will start pushing up more towards the 4.5GHz mark.

 

[AMDisTheBEST]

I game 3440x1440@100hz. The Ryzen CPU failed to break 100fps in some games compared to the intel chips. Some of those game results were really terrible. I don't care about 200fps vs 500fps, but I do care when Ryzen can't even feed a GPU to max out a 100hz monitor. I'm planning on a 1080ti and I don't want 1070 performance from it because my CPU can't keep up. I'm not alone here. A lot of people are kind of freaking out about this.

Feel free to ship me your unwanted 1800x if you dont want it. I'll pay for shipping.

 

[.vodka]

Yep. Updated microcode can do wonders. Those fixes will be back-ported to Ryzen 7, of course, via AGESA code updates (BIOS), but when Ryzen 5 shows up matters should be much different...

Windows will be updated with proper support, more games will be patched, BIOS support will be improved, etc...

People who are into buying at this level are usually "nerds," so it's pretty safe to sale to us - some of us will reverse engineer the code running on the CPU and send very detailed reports.

This, so much this.

When Ryzen 5 gets launched, and Ryzen 7 gets retested in these much more stable and polished conditions, well, I don't think we'll even remember this launch day.

By mid year and a dozen BIOS updates and Windows patches plus games patched to properly support Zen, AM4 should be a solid platform. Gotta start optimizing for something different than Intel's architectures, you know.

Thing is, Bulldozer was an unsalvageable disaster, optimizing for it made a difference but it wasn't worth it. Zen on the other hand...

 

[formulav8]

AMD tricked me into thinking this was faster than a 6900K

You serious? They didn't trick you in any way. What they showed is perfectly repeatable as all the AMD specific benches are in line with the reviewers benches of the same apps.

 

[dahorns]

Feel free to ship me your unwanted 1800x if you dont want it. I'll pay for shipping.

Mighty generous of you.

 

[Rifter]

I find it odd you feel disappointed that you preordered, I'm in the opposite camp. I kind of regret passing up the chance to preorder while Newegg still had the 1800X, I passed partly to wait on reviews and partly to see how the motherboards ended up shaking out. This would be a 3930K replacement as well. It's pretty obvious some of the results that could be viewed as disappointing are due to early teething and software issues. The remaining simply comes down to differing architectures having different strengths and weaknesses, for every bench in which the Ryzen trails by a noticeable amount there is another where it leads by such a degree. One thing is beyond doubt though, the all around price/performance champion is now a Ryzen cpu. It's the rough equivalent of a 6900K at half the price (or better). I'm more anxious at this point to see how the chips available six months from now end up clocking under high end cooling, I'm betting they will start pushing up more towards the 4.5GHz mark.

Yeah after the bios gets sorted and the first new stepping is released thats when the prime overclocking will be happening.

 

[looncraz]

When bulldozer launched, people said to wait for OS and game optimizations because performance would improve. It never happened.

It did happen, actually... just VERY slowly. Piledriver reviews looked all the better as a result. The difference with Ryzen is that there is a MUCH larger discrepancy in performance when working within a CCX and between two CCXes than when working two threads on the same module vs between two differing modules.

Why?

Dual L3 caches.

If you move a thread around inside a CCX it will most likely still find its data in the L3. If you do that between the CCXes that data will not be present and will need to be fetched from RAM... which, insult to injury, has latency issues (which have been getting better). This process happens reliably and frequently while gaming. A Windows patch can fix that - but game patches on top will help as well (keeping data local to a CCX).

This is AMD's first SMT implementation - I fully expected for it to slow down single threaded performance in at least some scenarios. The degree is more than I expected, but it makes sense given the dispatch queue and retire queue are statically partitioned. That, most likely, can't be fixed with microcode - BUT Windows updates should be able to make it a very minor issue.

If you plan to game with Zen... disable SMT. You'll still have eight real cores to give you an advantage over Intel quad core offerings. If you need to do a lot of rendering or something of that nature, turn it back on. Too bad this can't be done from software (yet?).

 

[JDG1980]

It's somewhat disappointing that AMD still hasn't figured out how to do a polished launch. Crashes on the testbed systems are especially inexcusable.

By the time the Ryzen 5 series launches, the BIOS and Windows issues should all be worked out, so we will have a clearer idea of which performance issues can be fixed with simple software patches and which issues are the result of architectural limitations.

AMD obviously optimized Ryzen 7 for workstation/server use rather than gaming. That may be disappointing to a lot of people here, but from a business perspective it should work just fine. Naples will be a good product assuming they get all the firmware/driver kinks worked out before it launches. Most gamers will be better off sticking with the 7700K for now. Zen+ may change that, especially if AMD works closely with studios to ensure future titles are properly optimized for their architecture.

The_Stilt's chart on Cinebench scores at various wattages indicates that Raven Ridge is going to make a big splash into laptop markets when it arrives. Ryzen 7 is already more efficient in most apps than Broadwell-E, and the desktop chips are running well above their sweet spot for perf/watt in terms of clock speeds. The_Stilt measured 850 points in Cinebench R15 MT at only 30W - go down to 4C/8T with the same architecture and you should have 425 points at 15W. In comparison, the Surface Book with i7-6600U at 15W only gets 316.06 points. Kaby Lake may be a bit more efficient than that, but not much. Ryzen will hold its own quite well, especially since the AMD chips will have far better iGPUs.

All in all, though the gaming results are somewhat disappointing, AMD laid a solid foundation which can be effectively built on for the next couple of years. TechPowerUp hasn't done a full review yet, but their summary of Ryzen's launch is one of the best I've read.

 

[formulav8]

Otherwise the CPU is great and will have a huge impact in Server market the coming months.

That's is the bread and butter for AMD. The 4/8? channel 32 threaded Naples will be a high-end beast at a much cheaper price than Intel.

 

[Absolute0]

That's is the bread and butter for AMD. The 4/8? channel 32 threaded Naples will be a high-end beast at a much cheaper price than Intel.

Yea, also TheStilt had a good post showing how the architecture is super efficient at slightly lower frequencies where Naples is likely to be poised.

Big savings in those serverfarms... Naples is where AMD needs to eat into Intel's marketshare and get that real sweet $$.

HEDT is so dominated by gamers that I think AMD can really only attract a thin slice of the broader Desktop market.

 

[Doom2pro]

Ohh this gets even better, over at Semi Accurate there is some juicy tidbits about what is going on with the benchmarks being all over the place...

Apparently, some reviewers were not using a fresh Windows 10 build (they were taking an Intel build drive and letting it "see" the new AM4 Platform - WHAAAA??? WHY???) in some attempt to save time I guess? This lead to Boost being BROKEN, which apparently escaped several reviewers.

They also had to use non-public beta BIOSes which were sent out a few days ago, guess they didn't get the memo?

Also they needed to disable high precision event timers and set the windows 10 power option to high performance due to AM4 Windows 10 drivers being MIA at this point.

Then you have lots of the top reviewers pulling a head scratcher in regards to the memory, just flat out giving up and letting it run at gimped speeds when higher speeds were a click away (some Motherboards wouldn't post some would, I guess these reviewers forgot how to operate a PC BIOS), and then you have the gaming benchmarks where these same reviewers decided that 1080p is representative of high end PC gaming? HUH? What is this? The result of the Call Before you Write?

Some of this is obviously AMDs fault, some of it is the Reviewers fault, overall this launch has a bit of a black eye but like any black eye it will get better with time.

 

[lolfail9001]

where these same reviewers decided that 1080p is representative of high end PC gaming? HUH? What is this? The result of the Call Before you Write?

1080p is the upper limit of what should be considered for any CPU comparisons, if you want to do CPU comparisons. If you do not care about CPU performance in games, just skip that section because Ryzen won't suddenly beat 1700k by switching resolution to 8k.
The only good CPU game benches (not real-world playthroughs, but game benches) done were by [H] as always.

Ohh this gets even better, over at Semi Accurate there is some juicy tidbits about what is going on with the benchmarks being all over the place...

These juicy tidbits are all over all the reviews including the AMD trying to spin the questionable gaming results by: "Hurr durr test at 4k, Ryzen ties Intel there", so you bring nothing new here.

 

[Glo.]

This, so much this.

When Ryzen 5 gets launched, and Ryzen 7 gets retested in these much more stable and polished conditions, well, I don't think we'll even remember this launch day.

By mid year and a dozen BIOS updates and Windows patches plus games patched to properly support Zen, AM4 should be a solid platform. Gotta start optimizing for something different than Intel's architectures, you know.

Thing is, Bulldozer was an unsalvageable disaster, optimizing for it made a difference but it wasn't worth it. Zen on the other hand...

I think you underestimate the number of optimizations AMD has to do, with game developers to improve Ryzen performance.

It will take much more time, and we can not see better numbers when Ryzen 5 reviews will be out.

P.S. We are talking like the CPU is a let down. It is not. Absolutely not a let down. If you consider how it is priced, and how it behaves being new uArch, it is success for AMD.

 

[krumme]

Ohh this gets even better, over at Semi Accurate there is some juicy tidbits about what is going on with the benchmarks being all over the place...

Apparently, some reviewers were not using a fresh Windows 10 build (they were taking an Intel build drive and letting it "see" the new AM4 Platform - WHAAAA??? WHY???) in some attempt to save time I guess? This lead to Boost being BROKEN, which apparently escaped several reviewers.

They also had to use non-public beta BIOSes which were sent out a few days ago, guess they didn't get the memo?

Also they needed to disable high precision event timers and set the windows 10 power option to high performance due to AM4 Windows 10 drivers being MIA at this point.

Then you have lots of the top reviewers pulling a head scratcher in regards to the memory, just flat out giving up and letting it run at gimped speeds when higher speeds were a click away (some Motherboards wouldn't post some would, I guess these reviewers forgot how to operate a PC BIOS), and then you have the gaming benchmarks where these same reviewers decided that 1080p is representative of high end PC gaming? HUH? What is this? The result of the Call Before you Write?

Some of this is obviously AMDs fault, some of it is the Reviewers fault, overall this launch has a bit of a black eye but like any black eye it will get better with time.

This product was rushed. There is not two ways about it.
Reviewers didnt have time. The bios support for higher than 2133 is a mess at best.
Period. I booted it today and can just confirm reciewers frustration. And they have to write not play.

But i would rather have it now than waited 3 months.
The CFO of amd agreed with me.
No need to complain on either side.

 

[.vodka]

I think you underestimate the number of optimizations AMD has to do, with game developers to improve Ryzen performance.

It will take much more time, and we can not see better numbers when Ryzen 5 reviews will be out.

It will be better than today, but sure, making it perfect will take time.

Probably by the end of the year.

P.S. We are talking like the CPU is a let down. It is not. Absolutely not a let down. If you consider how it is priced, and how it behaves being new uArch, it is success for AMD.

Spot on. It's far from a failure.

If it's doing as well as it is in these conditions...

 

[unseenmorbidity]

It did happen, actually... just VERY slowly. Piledriver reviews looked all the better as a result. The difference with Ryzen is that there is a MUCH larger discrepancy in performance when working within a CCX and between two CCXes than when working two threads on the same module vs between two differing modules.

Why?

Dual L3 caches.

If you move a thread around inside a CCX it will most likely still find its data in the L3. If you do that between the CCXes that data will not be present and will need to be fetched from RAM... which, insult to injury, has latency issues (which have been getting better). This process happens reliably and frequently while gaming. A Windows patch can fix that - but game patches on top will help as well (keeping data local to a CCX).

This is AMD's first SMT implementation - I fully expected for it to slow down single threaded performance in at least some scenarios. The degree is more than I expected, but it makes sense given the dispatch queue and retire queue are statically partitioned. That, most likely, can't be fixed with microcode - BUT Windows updates should be able to make it a very minor issue.

If you plan to game with Zen... disable SMT. You'll still have eight real cores to give you an advantage over Intel quad core offerings. If you need to do a lot of rendering or something of that nature, turn it back on. Too bad this can't be done from software (yet?).

So, is there no chance Zen's SMT will be an asset for gaming?

I was thinking of getting the six core,and then upgrading to an 8 core a few years from now.

 

[RussianSensation]

And really who is buying a $500 CPU to run 1080p on low anyways, my 6 year old GTX 460 i had before my RX 480 could run 1080p on low so if you are trying to run a 5-7 year old GPU with zen and then complaining of low gaming performance i dunno what to tell you other than bring your GPU into this century. This just shows how most review sites are seriously biased as well, only a few games reviewed and at 1080p low setting is a complete disgrace to reviewers everywhere and the sites doing this should be ashamed of themselves. Its actually hard to find 1440p and 4k results which is just sad in todays world, so much bias even in reviewers. Seriously who is spending $500 on a HEDT CPU to game at 1080p low setting, its just unreal how some people go so far out of their way to confirm there bias.

X10000!

Seriously, most professional review media is completely clueless nowadays (or are in bed with Intel for product samples and marketing dollars). Who the hell buys a $330-500 8-core 16T CPU that they want to be a well-rounded processor for encoding, rendering, well-threaded office applications, streaming, and still be great at games, but then pairs it with a GTX1070/1080/1080Ti (or in reviews $1200 Titan X Pascal) and then uses a POS $90-200 1080p 60Hz monitor? Give me a break! I will continue ripping into 1080p 60Hz gaming as we are now in 2017, and 1070, a 1440P card, is only $349. Not only is 1080p outdated for PC enthusiasts, but most 1080p monitors are small in real estate (24" or less) and tend to be budget in terms of IQ quality (a lot of them are mediocre IPS or TN panels).

Joker is one of the few true PC gamers on YouTube who did a Real World Ryzen vs. Intel PC gaming review.

The idea of testing the CPU at low resolutions or low GPU settings to "test CPUs" is marketing drivel that Intel has used for a decade, despite it not making any sense in a decade. If 95% of games are GPU limited at 4xMSAA 1080p/1440p/3440x1440 or 4K, those are the real world results us gamers actually should care about since those are the scenarios we use our PCs in.

I did not buy my 6700K and 6800K CPUs and GTX1070s to play games at 1080p low-medium settings. To exacerbate the matters, these so called professionals reviewers used Titan XP for gaming benchmarks. In fact, if there is ANY extra gaming performance on the table, I, and I am sure many of you, will increase all IQ settings to the max, and then if more performance is available, we woudl raise MSAA to 2-4X or even enable SSAA. The fact is a 4Ghz R7 1700 thrashes my 6700K in so many other applications outside of games that it's a MUCH better well-rounded processor. Outside of professional gamers who need 200-300 fps, who wants to play games with tearing?

The comparisons of 7700K vs. R7 1800X are so stupid, it's beyond any reasonable logic. That's like claiming i7 6900K is a failed CPU because it loses to the 7700K in gaming. It's OK to claim that 7700K is a better CPU for gaming, but that doesn't suddenly make R7 Ryzen a bad CPU. I would hope so that someone buying a $330-500 CPU does something other than gaming with it; and especially not gaming at the peasant 1080p 60Hz resolution. Otherwise, you do not need to spend that much on a processor in the first place. I am sure you can pick up a used i7 4770K/4790K and that would be more than adequate for gaming.

There also seems to be a lot of criticism coming from i7 4770K/6700K owners that Ryzen didn't really change the landscape for them. Let's look at Steam survey and see just how many Steam users have a CPU as powerful as the i7 4770K/6700K? For someone who has an i7 920/860, i5 2500K/2600K, R7 1700 @ 4Ghz would be an excellent upgrade.

It's amazing how the pro-Intel media never criticized 6-8 core Intel CPUs starting with i7-990X and 3930K eras, but the minute AMD's Ryzen makes 5820K/6800K/6850K and 6900K irrelevant and frankly flat out horrible in value, all of a sudden ALL the focus shifts to 1080p and lower PC gaming benchmarks? What a joke our tech-review industry has become. How come the media is barely discussing that other than Thunderbolt and SLI/CF support, the $90-100 B350 boards are significantly cheaper than the X99 board despite offering all the latest modern features from PCIe 3.0 x4 M.2 support, to USB 3.1 Type-C to Ryzen overclocking support? The total platform cost rises even more with X99 for those who want a multi-threaded powerhouse of a CPU.

With Intel, we basically get amazing mainstream gaming CPUs that are slow for multi-threaded tasks OR extremely overpriced X99 parts. With Ryzen, we actually get the most well-rounded processor to keep for the next 4-5 years. With B350 chipset, a mobo can be purchased for $90-100 and Ryzen 2.0/3.0 7nm parts will likely be backwards compatible with AM4 socket come 2019-2020. OTOH, LGA1151 Z170/270 or X99 platforms are completely dead. It's interesting how most "professional" reviewers have no clue as to the forward looking advantage of the AM4 platform as well. It means a builder can pick up a 4C or a 6C Ryzen and just get the 2019-2020 8C Ryzen when he/she needs more performance. With Intel, not only are you forced to pay more for X99 chipset board, but these boards have no upgrade path at all.

If I had to build a new PC system now, I would purchase an R7 1700 over the 6700K/7700K. I'd rather have a CPU that's 95% as good for games but is 30-60% faster in anything else I want to throw at it today or in the future. I would also get a guaranteed upgrade path for faster Ryzen models. Win-Win.

Techno-Kitchen already showed that a 5Ghz 7700K is barely faster in games than a 4Ghz 7700K. Most games today are GPU-limited once we start using Real World gaming settings and anti-aliasing settings!

7700K@4.9 vs 7700 (4.0) in all actual & new games (GTX 1070)

 

[Glo.]

You should've seen professional forums.

Extasy about the performance, and value.

 

[Doom2pro]

1080p is the upper limit of what should be considered for any CPU comparisons, if you want to do CPU comparisons. If you do not care about CPU performance in games, just skip that section because Ryzen won't suddenly beat 1700k by switching resolution to 8k.
The only good CPU game benches (not real-world playthroughs, but game benches) done were by [H] as always.

These juicy tidbits are all over all the reviews including the AMD trying to spin the questionable gaming results by: "Hurr durr test at 4k, Ryzen ties Intel there", so you bring nothing new here.

So 7700K only runs good at 1080p? Why would I gimp my HEDT rig to get it to perform better than Ryzen at lower resolutions?

I care about realistic performance, pushing my rig to it's limits, not gimping it so Intel has a better chance.

 

[krumme]

The freq is 5-8% below what i gunned for. But looking at the professional bm results at anandtech its striking how powerful this cpu is. Its a brute. The fpu is especially a good deal stronger than i expected.
Who expected this fpu perf level?

Looks a bit like k7 and k7 launch. At that time gaming wasnt the number one. Productivity professinal bm and multimedia in a broader sense what was driving reviews.

Its the youtube age and a lot of jungsters that force all this gaming craze now. Like its a console.

 

[formulav8]

just like Intel

Yeah Intel showed us their thoughts by charging over 1k for a cpu, targeting a certain market, that a $500 cpu could do just well overall.

 

[formulav8]

If you do overclock, then you paid too much for an 8C/16T RyZen chip if you bought an 1800X or a 1700X

By far, most people don't even enter the bios, much less overclock.

 

[french toast]

1080p is the upper limit of what should be considered for any CPU comparisons, if you want to do CPU comparisons. If you do not care about CPU performance in games, just skip that section because Ryzen won't suddenly beat 1700k by switching resolution to 8k.
The only good CPU game benches (not real-world playthroughs, but game benches) done were by [H] as always.

These juicy tidbits are all over all the reviews including the AMD trying to spin the questionable gaming results by: "Hurr durr test at 4k, Ryzen ties Intel there", so you bring nothing new here.

Totally agree, all benchmarks should be space invaders in 240p, why overly stress the gpu.