Is anyone else incredibly excited for the 1900X model Threadripper that AMD announced today? I think I was always more interested in the X399 platform and all those PCIe lanes than TR itself. Sure 16 cores is amazing but it’s probably more than I need and I don’t want to pay $1000 for a CPU (even if it is a better deal than Intel offers).
But the 1900X MSRPs for only $50 more than the Ryzen 1800X MSRP (mind you those are widely available for under MSRP so the real world gap might be larger) and gives you full access to the X399 platform. Now that is a hard to pass on deal.
I’ve been looking for the 8 core version as well. The need for small offices to have 12 or 16 cores on a single database is not really feasible. I think the 64 lanes and quad channel memory is the selling point for small business servers. I can put so much NVME storage and GPUs passed through to a vm for remote desktops and still add 10GB network adapters without worrying about lanes. It’s a great small server platform. Let’s hope the motherboards are in the $300 - $400 range because the 8 DIMMs are gonna break the bank currently.
it does look pretty compelling. I’m going to wait for benchmarks for sure for performance, if its going to lose anything for having the cores more spread apart. I am thinking about how surprisingly the 7700k still topped gaming performance over similarly clocked 7900x which i think was attributed to a change in how the cores are connected with the 7900x, getting rid of the ring connection or something and if Ryzen’s architecture is going to have a similar problem. How the larger heat spreader it is going to overclock might be interesting as well. One of the main drawbacks of Ryzen was the lanes for sure, and the x399 is a much more compelling platform. Also so far I haven’t seen any x399 boards with more than 3 m.2s, i wonder if thats going to change. And cooler support?
I wouldn’t mind seeing more built-in M.2 slots either, though don’t forget that PCIe to M.2 adapters are a thing.
Sometimes it feels like all anyone ever puts in PCIe slots these days are GPUs, so it’s worth reminding ourselves that PCIe is a general purpose interface. It kind of makes sense for most of the PCIe lanes to go to PCIe slots, where users can add whatever they need. That said, the arrangement on these boards is pretty GPU-centric. Most non-GPU devices don’t need an 8x let alone a 16x slot. That said if you were to populate the board with 4x slots you would run out of physical space before you ran out of lanes so that would be kind of dumb.
16/8/16/8 plus 3 M.2 slots seems to be the most common configuration and that’s exactly 60 lanes so it makes sense. Some have a 5th PCIe that’s electrically shared with an M.2 slot, so not exactly useful if M.2 is your main objective. I guess my ideal board would be something like 16/4/16/4/4 plus 4 M.2 slots. Is that too specialized? Is there a real demand for 8x slots?
Yeah all 8 phase boards pretty much.
Some of them look really nice, can’t wait to take a closer look at what they have to hide underneath those heatsinks.
Gigabyte allready showed one of their boards as pre production models at computex.
On that pre production model they used IR3553 powerstages.
But i personally hope that they have re-thought about that.
And using IR3555´s instead, But yeah thats yet to be seen.
Not gonna lie, if I wouldn’t already be running my 1700 build, I would probably invest a bit more and get into Threadripper. I am totally happy with my system but that amount of PCIe for overall maybe 300,- bucks more… very tempting.
I really want to see the encoding performance. Looking through Anandtech’s ‘bench’ database you can see a few dual processor systems but the handbrake re-encoding times show next to no benefit over a single cpu. Also doing some googling shows that a bulldozer based opteron 16 core also has the same lack of scaling beyond the first die. So if the 1900X has the same problem then it wont be any faster at encoding than a quad core ryzen 5.
To get around it you just do a couple of encoding jobs at the same time on the different cpu’s/die though some VM’s.
Hopefully Wendell run run some test’s on the multi die scaling.
Newegg has some X399 boards available for preorder now. They are, as expected, a bit pricey. The cheapest so far is the ASRock X399 Taichi at $340 and the most expensive is the Asus ROG Zenith Extreme at a whopping $550.
I actually did a small happy dance to see that 1900x, was what I was hoping for, price and specs both. Now to figure out why I need it when I already have a 6 core intell from years ago. Figure I will buy in 6 months or so when the bugs are ironed out more tho.
Perhaps the intell can run a plex server, be a fun duty to give this thing, tho my $400 MB is a bit much for that, lol.
The 1900x is really the chip that the 1800x should have been in the first place. Not so much for the 64 PCIe lanes. More than the 28 that are available in x370 is great but 64 lanes are not that essential given four way sli has pretty much gone away and other than graphics cards, nothing else uses x16 PCIe lanes.
Together with the extra 100Mhz of probable overclock headroom these should get, the key thing that will let this chip be properly performant in gaming/graphics loads will be the second 32 byte/cycle infinity fabric interconnect to the additional dual channel memory controller.
Having two interconnects to access system memory will mean that each memory controller only has 8 threads and 1/2 a GPU to support instead of the 16 thread and 1 gpu load R7 chips place on their memory controller. The 1600 and 1700 both perform about the same in most game benchmarks suggesting that 8-12 threads is about as much as a dual channel memory controller can comfortably handle in a gaming load.
I wonder how many reviewers will test the 1900x with memory only in dual channel mode and repeat the same ridiculous waffle that some of them said about Ryzen
Encoding does not tend to be that sensitive to memory latency, it is sensitive to core count and frequency though. I would expect the 1900x to be similar to slightly better than an R7 at the same clock speed using. It will certainly be better than a quad core just as an R7 is better.
Used 40Gbps/56Gbps ethernet adapters (5/7 lanes worth of bandwidth per port) are starting to trickle down from various datacenters to eBay. I totally wouldn’t mind using the extra lanes to try those bad boys out.
Also, if you’re buying a system to use for the next several years, having the extra bandwidth for extra type-c ports or who knows what kind of expansion cards can only ever be a good thing.