Finally a 3800X review

As expected, it is the 1800x of this generation. The extra $80 can be spent on a better cooler, MB, GPU, etc with far more meaningful gains.
 
I really feel like amd dropped the ball by not having at least dual thread turbo boost at the max value. 3700x seems like a much better buy.
 
Don't mean to thread hijack but Hardware Unboxed did a test with all cpus locked at 4.0 ghz.


The 3900x was consistently 3-4% faster. That doesn't seem like much, but it was enough to put the 3900x right between the 3700x and 9900k.

Again, everything was locked to 4.0 ghz, so please don't misinterpret. It seems the extra cache helped and the 3900x surprisingly had lower latency.

It would be interesting to see what a 12 core (edit: TR) does if they make one with even more cache, more surface area and quad pumped memory.

Back to reality though, the 3800x is a waste.
 
Last edited:
Nightfire said:
Don't mean to thread hijack but Hardware Unboxed did a test with all cpus locked at 4.0 ghz.


The 3900x was consistently 3-4% faster. That doesn't seem like much, but it was enough to put the 3900x right between the 3700x and 9900k.

Again, everything was locked to 4.0 ghz, so please don't misinterpret. It seems the extra cache helped and the 3900x surprisingly had lower latency.

It would be interesting to see what a 12 core Ryzen does if they make one with even more cache, more surface area and quad pumped memory.

Back to reality though, the 3800x is a waste.
Click to expand...


The weird part here is when your looking at the productivity benchmarks, AMD had a higher IPC, clock for clock, core for core, than the 9900K

But in games, same clock speed, same # of cores, all things being equal... the 9900K was faster.

So that is really weird to me. Basically it only goes to show that something strange here is going on with the software & optimizations.

Curious if the problem is in the game engines themselves, or if this is a Windows specific issue.
 
IMO, we need to wait to make full judgement on the 3800x until we get confirmation that PBO is operating as expected, currently.

I have a 3800x, and even with PBO enabled, the board isn't using more than the default 95w limit for the 105w CPU's.

Even stock, my 3800x occasionally hits 4.6ghz boost, and is 4.2ghz all core. PBO doesn't change this at all, but given that I'm sitting at 70c and only 95w I would suggest my 3800x still has more thermal room to throttle up clocks if it would let me. My 2700x with PBO enabled would use 130w, but again, with the 3800x I can't get it to use more than 95w no matter what settings I set in the BIOS.

We need to wait a little, I believe, before we give a complete judgement on the 3800x. I believe PBO is bugged right now, and the 95w cap is keeping it held back. Given I have at least 10c more of thermal headroom and i'm already hitting such high frequencies with a 95w cap, I can't imagine this thing not being able to clock up more.

What's weird is that the 3700x is allowed to hit the 95w limit when you enable PBO which then brings it on-par with the 3800x. So it's like PBO works for the lower-wattage parts, but not the 3800x/3900x.
 
Sindalis said:
The weird part here is when your looking at the productivity benchmarks, AMD had a higher IPC, clock for clock, core for core, than the 9900K

But in games, same clock speed, same # of cores, all things being equal... the 9900K was faster.

So that is really weird to me. Basically it only goes to show that something strange here is going on with the software & optimizations.

Curious if the problem is in the game engines themselves, or if this is a Windows specific issue.
Click to expand...

Engines by and large are optimized for the dominant vendor (or the one whom pays them the most)
in this case Intel is WAY larger and has supported/been supporting far far more developers etc for a much longer period of time (not to mention, many of the benchmark software to the best of my knowledge is directly influenced by Intel and to a lesser part Nvidia leaving AMD in an uphill battle to be that much better (in terms of build quality at least) at a lesser price with more features thrown in for good measure.

For AMD to even be competing with if not dominating much of what Intel has at this point says FAR MORE than AMD being behind by a bit here and there .. no other company I am aware of besides maybe IBM have built CPU to this high of a level (for many OS not a specific one)

^.^
 
dragonstongue said:
Engines by and large are optimized for the dominant vendor (or the one whom pays them the most)
in this case Intel is WAY larger and has supported/been supporting far far more developers etc for a much longer period of time (not to mention, many of the benchmark software to the best of my knowledge is directly influenced by Intel and to a lesser part Nvidia leaving AMD in an uphill battle to be that much better (in terms of build quality at least) at a lesser price with more features thrown in for good measure.

For AMD to even be competing with if not dominating much of what Intel has at this point says FAR MORE than AMD being behind by a bit here and there .. no other company I am aware of besides maybe IBM have built CPU to this high of a level (for many OS not a specific one)

^.^
Click to expand...

I'd do a different take.
It's same ballpark but I chose to blame latency for most of it.

Latencies, every industry see higher latencies as bandwidth and capability increase, may that be internet, the playing a song on a modern device, display standards like DVI, DP, Hdmi vs good ol d-sub.. even my stereo system adds latency!
My fiber internet connection at 500/500 still has higher latencies than my ISDN connection back in 1997.

Now to cpu's, physics are catching up to them too, Coffee Lake see lower performance than Kaby Lake, CFL-R see even higher because the lovely ring is larger - not by much but a little by little, they're already struggling with bandwidth and increasing bandwidth just makes it even worse.
Mesh is also slower, chiplets and whatever intel plans will likely follow suit!
But when was the first time we actually saw a cpu not follow the norm when it came to memory latency? Zen.
Every CPU prior have had consistent 50-70 ns latencies, there was absolutely No issues assuming that latency.

This is even shown in amd's press deck, L3 cache just masks memory latency issues.. very well that has to be said

qMcEgXTegugq8vyW.jpg


But even so, you can never ever not rely on dram memory. :)

So that being said, it's only a matter of time before Intel mainstream platform also gets latency issues to system ram, but my assumption is that both AMD and Intel has on die L4 cache which is at a magic 50-70 ns latency at little higher than dram speeds (probably ddr5).
It won't fix it, just make it better.

I checked latencies on my Xeon at work and it has about 100ns about 25ns worse than my previous xeon.. Latencies are getting worse and game devs will optimize for it.

End comment, the more I've looked into design on the zen family the more amazed I am, and I would not count on latencies getting better but these damn engineers are stubborn at times :)
 
You must log in or register to reply here.
Back
Top