Originally Posted by briank
Did Zen1 or Zen1+ improve over time?
I am a partner and Founder in an IT Firm here in Vegas. We do a lot, but my Main Focus is designing, building, maintaining, and servicing Servers, Workstations, and Desktops for Mid to Small Size Business (anywhere from 10 to 60 Workstations/Desktops in a Business, and anywhere from 0 to a Cluster of 8 Servers, depends on the Client.) I also have a huge focus on Networking Infrastructure, but thats not relevant here.
Since the Launch of Zen 1 I or my team have either built or bought over 150 Zen based systems, not including the Epyc Servers. About 60 Of those were Zen 1 Based Systems, and the Other 85 or so before the launch of Zen 2 were Zen+ Systems (Because of a error in our CRM/Asset Tracking Software the 2200G and 2400G are counted amongst Zen+ Data, meaning it balance between the 2 are actually far more even as we rolled out ALOT of 2400G based systems but I don't have an exact number on that right now). For those curious by the end of this week we will have already Rolled out 8 Zen2 Based systems, which is a lot of systems for us in a Month, especially considering I personally built them all as our OEM Partners have still not given us Zen2 Options yet.
Anyways, I only say all this to give you an idea of the amount of Zen Based systems I have dealt with. Even for the systems I don't personally build, it is part of my job to stability test the systems with the clients specific workloads, which means each PC generally spends anywhere from a day to a week in my lab before being put into production. I have become the Zen expert in our company just because of the amount of time I have spent with them, and considering our business consisted of almost 100 Percent Intel Chips before the Launch of Zen and Epyc, its become quite the asset to know these chips so well.
Anyways, the whole point of this post is to tell you that definitively, without a doubt, with both Zen, but even MORE so with Zen+, the average quality of the chips increased over the products life cycle, specifically with the X based chips. With the Original Zen for the first 4 to 6 Months, the average Clock that I could achieve with the X Chips that was AVX2 Stable was 3.8Ghz. There were some that could hit 4.1Ghz, but only a few, and none of them would remain stable with a heavy AVX2 Workload. After about 6 Months the average Clock For the 1800x specifically was increased to 4Ghz. I only had 1 Chip after the first 6 Months that could not remain stable at 4Ghz, the rest all did it with Ease or Higher. The 1700x almost all hit 4Ghz, but the 1700 Remained the most hit or Miss from the 8 cores. We did a lot less 6 Cores, funny enough for our clients it was either 2400G, or 8 Core Chips, we've rolled out less than 10 6 core based Zen 1 Chips. Because of this the sample size is much smaller for those chips, but I would imagine the trend holds true with that segment as well.
With Zen+ it was almost identical, except it took about 7 Months this time before the X chips started consistently hitting higher clocks then chips received in the first few batches. After about that 7 to 8 Month Mark, 2700x and 2600x Consistently out performed their non X counterparts, and also were able to in almost every case perform just as well as the best 2700x I received in the first few months. It was incredible how well their binning process has become. If you go and you buy a 2700x still in stock now, and bench it either against the 2700s bought at the same time, or a 2700x that was purchased at launch, I can almost guarantee you that you will be able to hit that 4.25Ghz All Core Frequency that the 2700x usually topped out at, at least 50mv Lower Voltage or More. And you have a much higher chance of your chip being one of the chips capable of going above 4.25Ghz. Its the same with the 2600x.
So knowing how refined AMD was able to make their binning process (and keep in mind, Zen+ was fabbed at GloFlo Not TSMC, But the Chips are sent to their Factories in Texas and New York after Fabrication, and they are Binned there) I have no doubt in my mind that in about 6 months time we will see the X parts specifically start to, at the very least, consistently hit the same clock speeds of the top performing early X Chips, and more then Likely will be hitting those same top clocks at lower voltages. There is still obviously silicon lottery involved that will dictate if your chip will go higher than the current top chips, but your chances of it being a lower performer like my 3900x will be heavily reduced unless you buy a Non X Model (which 3900 non x's were announced btw). So those of us who have lower performing 3900x then our forum counterparts, will have a much better chance of buying a chip that performs more like theirs, if we buy another in about 8 Months time. They may even have the binning process down a bit quicker this time, so we may start to notice this process happen a bit quicker, but only time will be able to tell us that.
Anyways, I thought you guys would find this interesting.