Review: Ryzen 5 1500X and 1600X processors

Because I expect everyone who was still arguing about it to UTFG after my prickly comment.

So just for you. IPC is interpreted in few ways:
Sometimes like "integer per clock".
Usually as "Instruction per cycle".
But in each situation it is always "per frequency".

So, here you have Loopohole35's quote showing proper assessment of IPC for one application:

Now, please read part of your quote which I turned bold, and realize that you have some UTFGing to do.

For me, Ryzen is great CPU from day one because AMD did catch in terms of IPC to intel. They just need to be able to get higher clock and iron out few kinks.
Clock, that likely comes from intel's superior fabs. And as for those kinks, Zen+ will have at least worst of them ironed out.
1st Ryzen is like Nehalem and Ryzen+ will be like Sandy.
= = = =
So, where comes gaming advantage of intel CPU from? Considering that Ryzen has like 5% lower IPC and 15% lower frequency than 7700k, one should expect one Ryzen core to perform like:
0.95 * 0.85 = 0.8075 ~= 81% of 7700k core.

In other words: In purely single threaded applications OC of Ryzen will yield close to 20% worse results than OC of i7700k. That's to stay till use of 4 full cores.

But Ryzen has one advantage even if you look at applications using less than 4 cores. Its full data throughput and mathematical throughput is quite higher than i7700k. So there will be situations where threads on i7700k get crumpled to one core and application will see stutter because one of threads did not deliver in time. Ryzen has a lot of power to spare and choking it will be much harder as result.
= = = =
BTW. Have anyone seen 1C/2T Skylake vs 1C/2T Ryzen comparison? I would like to know how good IPC of SMT on each is to have bit better overview of total CPU extractable performance.

 

I have no idea what you are trying to prove there. I stated that Loophole35 is right in way he assesses IPC. I wrote IPC and not IPS. And you confirmed his assessment of IPC.

But maybe you wanted to just re-confirm my statement One never knows.

 

Upgrade from 2500K

Welp, there is a lot of discussion here so though't id dumb it back down a bit and say I've just built my new 1600X rig. Upgrading finally from my trusty 2500k (felt bad butchering old rig for parts, sad day )

Excited to press the power button and start BIOS flashing when home from work, still some cabling to sort.

Good to see AMD back on the map and really hope AM4 is a stable platform for a good few years so I can slot in a RYZEN 2 later down the line.

Thanks for the review Hilbert, long time Lurker!

 

You're right, I'm messing things up, sorry.

He bought his PC under my recommendation (I actually assembled it, like I did mine), and said he will buy the same GPU as me (probably RX570, depending on prices here). Mostly we play War Thunder, it's a GPU game and only uses 4 of his CPU cores :bang:

I envy you. Have fun dude!

 

My 2500k has to leave too. It will become network server.

I hope that your new build will serve you well.

 

If it was down to "game" specific IPC I would think Ryzen was stronger then what is shown in some cases. As it is strong in floating point math for example. But then again the architecture is vastly different from bulldozer or intels offerings. Tomb Raider is the worst case scenario when paired with nvidia card and dx12.

We can easily pin some of the lesser performance to frequency for sure. But all things considered 6900k and 1800x in games should be roughly equal. All being stock. We have seen ryzen specific optimisations bring it to par with 6900k where it kind of should be. Of course 6900k can be then overclocked to something like 4.3 and voila.

I just might change my 1800x to ryzen 2 next year. If I can actually use the motherboard I currently have

 

Why would you change your CPU so soon, you just bought it :3eyes:

Motherboard support is supposed to last 3 years, it's a plus of going AMD and avoiding the others who change socket every new generation.

 

Because I can is the right answer. I might not but if I can I just might. And build a cheaper 1800x rendering machine with lesser board and whatnot

 

Thanks Fox, I am hoping to also re-use the 2500k as a 2nd rig. Just needs a power supply and a gfx card. Going to ditch the electric heater that is the R9 290. Hope yours works out well as a network server. Honestly can't believe how good a chip the 2500k has been. 4.4GHz since 2012 at permanent high voltage (no offset). Hope my new AMD chip can live up to it.

 

I guess this is a good read up for you:

http://www.overclock.net/t/1564781/...-dreamer-reporting-ipc-instructions-per-clock

The whole new thing that everyone is talking about is that if you overclock RAM you will get IPC increase on Ryzen. You didn't miss that right? That's the easiest example for when overclocking increases IPC.

And for completeness.
Yes, you can influence IPC even by OCing your CPU ... and I know it's hard to understand because at first glance it contradicts the "per clock / per cycle" logic. But considering certain code just can be executed more efficiently because caches are cleared a certain time earlier for example (you really have to look at the whole CPU / RAM / Cache subsystem) it means that your per clock value of instructions can be higher on the higher clocked same CPU.


BTW:
the benchmark in the link could be very interesting for comparing Ryzen with the 7700K because it aims EXACTLY at reproducable ST and MT IPC values.

 

Technically speaking, OC does not change IPC and never will. On instruction level IPC is static. OC of Infinity fabric decreases some latencies, but integer unit still does same number of instructions per cycle.

From external point of view OCing memory increases number of instructions done per second.
But there is no increased number of ticks, that transistors in integer unit or any other logic processing part of CPU does per second.
If unit processes 4 add operations per cycle, OC of infinity fabric(memory) will not increase it to 4,2 add ops. per cycle.

What OC of memory does is simple, it removes bottleneck caused by slow communication.

 

It is NOT static.

Less pipeline bubbles (NOP) means higher IPC.

More pipeline bubbles means lower IPC.

 

It seems like most reviewers are fawning over the Ryzen 5 - a very different reaction compared to the Ryzen 7. I think the Ryzen 5 is more in tune with what people expected, and can be compared with the Core i5 on an appropriate price level (by comparison, I think it was difficult to judge where Ryzen 7 fit). Also, I think some reviewers may have been keen to avoid the kind of backlash that resulted from the Ryzen 7 reviews.

I built a second system when Ryzen 7 was released, and it's now crunching data alongside my 4790K. As I had hoped, the Ryzen 7 is producing almost twice the computing numbers in WCG - and I couldn't be happier . When a second-gen Ryzen (or perhaps a 16c/32t Ryzen 9) is released I'm thinking of retiring my 4790K and going full AMD. I have to say, it's an exciting time to be a PC enthusiast.

 

It is good theory. But let me expand on it.
You want to include latency from system memory into play. It is fine. Then let's say CPU core does 4 add ops per cycle when it executes from L1.
It does 3.8 add ops. per cycle when it has to fetch that from L2.
Then 3.2 add ops. per cycle when L3 is in play.
And 2.8 add ops per cycle when data was in system memory.

Why not to go one step further and add data on SSD, or HDD, or flash drive?
In the end, you want to add external latency into calculation of static CPU design just for sole purpose of having dynamic IPC.

No problem with Ryzen, it has direct I/O for SATA and USB ports. Same as memory.

No? Looks stupid, right? Only thing which belongs to calculation of IPC is size of data set and cache misses connected to this data set. Else I can take any CPU and say that with lowest speed memory its IPC sucks.
But that would be false statement. IPC remains same, it is just buttlenecked by memory subsystem.

If you ever designed any ASIC, even smallest and simplest one, you would know exactly how many cycles it takes to change state after receival of signal/trigger.
You imagine elevator with variable ability to move from floor to floor based on sleepiness of lift attendant while human factor is not part of core elevator logic and mechanical operation.

 

I still see it a little different.
Because the moment you start measuring IPC is the moment you need to make a compromise as to what external factors you need to put into the equation.

Sure you can leave out HDD, because basically you can run code purely from a RAMdisk to prevent any interference.

But, you won't be able to make an isolated IPC premised measurement
- without any code (software)
- without RAM
- without chipset/subsystem

So you have to take these into considerations and all the consequences they bring to affect the IPC.

 

BIOS.

Btw, when memory bottlenecks CPU, it is analogy to CPU bottlenecking GPU.
But, I can agree that while your view is different and fundamentally wrong for my low-level-design point of view. If one looks at it in very macroscopic way, you can even take it your way and be correct. (It just blurs line between IPC and IPS. Which is blurring of design and real world result.)

 

Nice review...Sold the 1700x on Our Family PC and will buy 1600/1600x..The 1700x was overkill on Our Family PC for the price and intended use..

 

Hey Hilbert nice review,
i have one question though, i tried searching here in the thread but couldnt find it either here or in the article.
How is it with a chance of unlocking the 2 locked cores? like you know you can lock cores on the r7 models in the bios, is there the option to even try to unlock them for the r5 models (perhaps on x370 boards only)?
could you check that out please?
Thanks

 

I don't think it would be a good idea, they're turned off for a reason or they'd be selling for R7 1700.