Civ 6 Multicore performance

Gorbles, >>. I am laughing because you are absolutely correct. We have NO idea what the code is in either Gal Civ III or Civ VI and my statement would and should send the folks who are at Firaxis into a tizzy. I hoped to stir the pot a bit and churn up some discussion.

Thanks!

 

If rendering is already the bottleneck, then the indication is that the AI is finished with its deliberations and waiting around for the graphics to catch up. To me, this implies that there is no "give the AI more time and it will get smarter" effect to take advantage of. The algorithms and decision processes don't appear to be do any deep dive analysis with all that free time. Desires to have a supercomputer play against you won't really help. I tried running a little word search generator on a supercomputer once. It sucked at the task. Big hardware is not always the answer.

I really don't understand why rendering is such an issue, though. The map is complicated textures and loaded with sprites in tiles, units, improvements, everything, but there are games out there with much more insane graphics demand and keeping up with that rendering with the right hardware.

Even with generous hardware, the between turn times are unimpressive for what I can envision is going on: rendering, and a complicated once through decision engine run sequentially once per player. If they insist on running the player turns sequentially and it is truly a once through decision tree, then there is little to be gained from multi-threading. So, I am not expecting a lot of performance changes from Civ6 as it goes forward. There is no indication that the AI is "thinking" in the background while I am dithering about or they are waiting for screen updates. I really hope to be in the wrong about this, though. We'll see.

 

Did anyone watch the AMD Ryzen presentation today? They seem to think this new chip will kick Intel into touch. Of course it's hard to demo a chip when Intel has no mainstream 8 core CPU that is affordable. Should make next few months interesting.

 

So could someone summarize whether Civ VI uses multiple threads for the AI? Kind of tired to go through the thread to find it.

I don't particularly care about multithreading for the graphics for, while nice, it's AI turn times more than pretty graphics that I care about in strategy games. I'd much rather have 7 threads for AI and one for graphics than vice versa.

 

Eh, we'll see. So far, the only demos are ones that take advantage of more cores and rely less on IPC and clock rate. Let's see how Ryzen does in games.

 

I read some articles about it. If it delivers what they've been saying, it should indeed make things interesting. The key is that 40% improvement in IPC. If they get that, a 3.4 GHz 8-core CPU would be pretty interesting, particularly if it's not priced quite so steeply as Intel's 8-core CPUs (even their 6-core ones are pretty pricey currently).

It would certainly be good for consumers and the industry in general if they do get that; CPUs have stagnated for quite awhile. Every year I see Intel's new release, and every year I think, "meh, that's not worth upgrading from Sandy Bridge" - and in less than a month, my CPU will have been available for 6 years. We won't know if it lives up to expectations until launch day, of course, but I'm hoping it is the first actually exciting CPU in a long time when it launches.

 

The big question is the release date. it was suggested Jan 17th. However AMD keep saying Q1 2017. The suggestion on price is $500 for the 8 core. With $350 for the 6 core which will follow in 2 months after release. Intel has it's own 6/8/20 core CPU to follow in August. Interestingly Skylake X will apparently have a TDP of 140w.

I am looking for a new system in the new year. Just need to wait for the right parts. A big drop in price from Intel would be nice. Have to see what AMD chips can actually do on a single core. All very well competing multi core but many things will only run 1-2 core.

 

I'm expecting an announcement of some sort, likely including an official release date, on or close to CES in January. Hopefully with at least some retail availability later in January. Hadn't seen price suggestions, but $500 for the octo would be half of what Intel charges for their Broadwell octo core, and $350 for a hex would be a good $75 less than Intel's pricing. So if the performance is on pace with Broadwell (which is what they've typically been comparing it against), those are decent savings. The 8-core is also supposed to be 110W TDP, or 30W less than Intel, so it could save power too.

Do you mean 6/8/10 from Intel in August? They just introduced a 10-core consumer part for the first time last year, and it costs $1500+; I'd be surprised if they went to 20 outside of servers. Maybe 12, but no more than that. I'd say definitely wait to see what Zen can do before buying when we're this close and it could shake things up a lot, but I probably wouldn't wait till August, especially given Intel's average per-generation gains have been single-digit-percentage for years.

I'm curious which ones you mean in August as well. Kaby Lake on desktop is supposed to be Q1, but that should be another ho-hum minor improvement release not worth waiting for. The best reading of the tea leaves I'm making currently is Cannonlake (10nm) for 15W and less laptops in Q4 2017, and Coffee Lake (14nm, 6-core) for desktops and high-power laptops in H1 2018, which would be a decent increase to hexacore being mainstream, but too long to wait. I don't see anything specific for August on the consumer side, but may have missed a rumor. Hence why I'm thinking it makes sense to wait for Zen (and maybe Vega on the graphics front if you're looking for high-end graphics, as it's also rumored for January, and current high-end prices are high due to nVIDIA not having any real competition there), but not anything beyond that.

Still curious on how Civ VI does with multiple cores for the AI.

 

Octo core S7 Ryzen chip is 95TDP. Well that is what they said at presentation on Tuesday.

http://www.fudzilla.com/news/proces...d-kaby-lake-x-cpus-could-be-coming-in-q3-2017

Basically follow up Broadwell chips. Yep definately waiting till new year. Was always plan. Only thing hitting this plan for 6 is the continually dropping value of the pound. The 140TDP is a joke if AMD have managed 95TDP.

Few sites are reporting AMD somehow limited the performance of the Intel chip and in fact it was not actually running at 3.4ghz. Even so the performance seemed pretty close. Of course on many games now the CPU is not the limitation. So maybe reason Intel seem to be hitting a cliff is CPU have gone as far as they can without a big innovation.

 

In fairness, Broadwell-E debuted in May, 2016. We're comparing AMD's still unreleased processors with processors that Intel's already been selling for seven months. Let's see how Intel responds!

 

Ah, Skylake-E. Or Kaby Lake-E if they skip Skylake-E. Yeah, that would be later this year.

I know in some of the first Zen benchmarks (in the summer, maybe late spring), AMD was comparing a 3 GHz Zen to a 3 GHz Broadwell, in order to demonstrate that they got better performance clock-per-clock, while also not revealing what clockspeed Zen would be released as (they likely hadn't nailed that down themselves at that point either). So both chips were not running at their top clocks, but the point of those early demos wasn't that, but that they could actually compete in instructions per cycle, since they'd been getting destroyed in that for years with Bulldozer and its successors. They were also clear at those demos that both chips were set to the same 3 GHz speed.

While yes, eventually it will be interesting to see how Intel responds, as mentioned their typical upgrades haven't given reason to wait for the new release unless it's imminent (which Intel's aren't, and AMD's soon will be). There's always something to wait for in PC hardware, but sooner or later you have to make a call or you'll be waiting forever.

As for the TDPs, if Intel can sell a 3.3 GHz quad-core Broadwell at 65W (and the same for Skylake), you'd think they could manage their 3.4 GHz hex-core Broadwell somewhere around 100-105W, but that's still listed at 140W. So's the 10-core, (admittedly 3.0 vs 3.4 GHz) $1723 processor though. So maybe they're just giving the whole E-line the same TDP? Despite the slower clock speed I'd expect the 10-core to consume some more power. Not saying Intel can manage 95W for an octo-core at desktop frequencies and mainstream prices - although they have 45W laptop quads that, if doubled, would amount to 90W octos at not that far below desktop clocks, and they've done more impressively in some server chips - but they may well be at least a little below 140W for the octos already if the 140 is a whole-line figure.

On the other hand, if they're increasing the voltage a fair amount on the 6/8 core Broadwell-E's over the slower-clocked 3 GHz ten-core, maybe they are using similar amounts of power. I know through overclocking my laptop's Core 2 Extreme that a relatively small amount of megahertz can require a relatively significant amount of additional voltage, and power consumption is related to voltage-squared, so it wouldn't be inconceivable that the extra 200 MHz on the 8-core over the 10-core really does keep power use similar, and again that at least on the 3.6 GHz hex-core, the extra 400 MHz might require enough additional voltage to negate most of the power gains. It would have to be significant to cancel out 40% fewer cores, but seeing the difference my CPU needs between 2.4 GHz and 2.8 GHz (25% more power from just voltage, before adding in the difference due to frequency), let alone 3.0 GHz, maybe it is that significant.

Now I kind of want to buy a 10-core Broadwell-E and a motherboard that would allow disabling cores to experiment with, but that definitely would blow a hole in my budget.

 

http://wccftech.com/amd-ryzen-7-1700x-gaming-benchamrks/

Gives first insight to Civ 6 stats for 8 core Ryzen vs a 6 core Broadwell. Gains of 15.17 and 20.43%.

You might argue the difference is purely due to the 2 extra cores. If these figures are right. Of course this is a CPU that has not even been released yet where Broadwell driver and other support is completely bedded in. Will wait and see for the reviews and the hype to die down here. A 8 core chip with 95tdp looks good. The Intel chip seems to be running faster ram. Broadwell 140tdp vs 95 TDP for Ryzen.

Overall figures look very promising given this is not even the high end model of the 3 8 core Ryzen chips coming out. Does this prove more cores for Civ 6 lead to better performance? Pends how they tested the game I guess. The waits during turns might still be the same?

Would a 400 dollar Ryzen 8 core chip blow a hole in your budget?? Hmmmm.

Personally AMD Ryzen 7 1700 one at $329 looks better value. 65tdp. Wait to see if Intel drop their prices in a few days.

 

FPS is very much secondary to late game turn times....so does anybody have a late game, huge map save I can use for benching the 1800X?