At SIGGRAPH in Los Angeles this week, AMD released even more details about the its upcoming Ryzen Threadripper product family ahead of its retail release in August. Though readers of PC Perspective are already well aware of the Threadripper 1950X and 1920X CPUs that were announced just a couple of weeks back, along with prices, clock speeds, performance estimates, and more. At tonight’s Capsaicin event, we learned about the on-sale, preorder date, and even a surprise new SKU option.
i9-7980XE | i9-7960X | i9-7940X | i9-7920X | i9-7900X | i7-7820X | i7-7800X | TR 1950X | TR 1920X | TR 1900X | |
---|---|---|---|---|---|---|---|---|---|---|
Architecture | Skylake-X | Skylake-X | Skylake-X | Skylake-X | Skylake-X | Skylake-X | Skylake-X | Zen | Zen | Zen |
Process Tech | 14nm+ | 14nm+ | 14nm+ | 14nm+ | 14nm+ | 14nm+ | 14nm+ | 14nm | 14nm | 14nm |
Cores/Threads | 18/36 | 16/32 | 14/28 | 12/24 | 10/20 | 8/16 | 6/12 | 16/32 | 12/24 | 8/16 |
Base Clock | ? | ? | ? | ? | 3.3 GHz | 3.6 GHz | 3.5 GHz | 3.4 GHz | 3.5 GHz | 3.8 GHz |
Turbo Boost 2.0 | ? | ? | ? | ? | 4.3 GHz | 4.3 GHz | 4.0 GHz | 4.0 GHz | 4.0 GHz | 4.0 GHz |
Turbo Boost Max 3.0 | ? | ? | ? | ? | 4.5 GHz | 4.5 GHz | N/A | N/A | N/A | N/A |
Cache | 16.5MB (?) | 16.5MB (?) | 16.5MB (?) | 16.5MB (?) | 13.75MB | 11MB | 8.25MB | 40MB | ? | ? |
Memory Support | ? | ? | ? | ? | DDR4-2666 Quad Channel |
DDR4-2666 Quad Channel |
DDR4-2666 Quad Channel |
DDR4-2666 Quad Channel |
DDR4-2666 Quad Channel | DDR4-2666 Quad Channel |
PCIe Lanes | ? | ? | ? | ? | 44 | 28 | 28 | 64 | 64 | 64 |
TDP | 165 watts (?) | 165 watts (?) | 165 watts (?) | 165 watts (?) | 140 watts | 140 watts | 140 watts | 180 watts | 180 watts | 180 watts? |
Socket | 2066 | 2066 | 2066 | 2066 | 2066 | 2066 | 2066 | TR4 | TR4 | TR4 |
Price | $1999 | $1699 | $1399 | $1199 | $999 | $599 | $389 | $999 | $799 | $549 |
TR 1950X | TR 1920X | TR 1900X | Ryzen 7 1800X | Ryzen 7 1700X | Ryzen 7 1700 | Ryzen 5 1600X | Ryzen 5 1600 | Ryzen 5 1500X | Ryzen 5 1400 | |
---|---|---|---|---|---|---|---|---|---|---|
Architecture | Zen | Zen | Zen | Zen | Zen | Zen | Zen | Zen | Zen | Zen |
Process Tech | 14nm | 14nm | 14nm | 14nm | 14nm | 14nm | 14nm | 14nm | 14nm | 14nm |
Cores/Threads | 16/32 | 12/24 | 8/16 | 8/16 | 8/16 | 8/16 | 6/12 | 6/12 | 4/8 | 4/8 |
Base Clock | 3.4 GHz | 3.5 GHz | 3.8 GHz | 3.6 GHz | 3.4 GHz | 3.0 GHz | 3.6 GHz | 3.2 GHz | 3.5 GHz | 3.2 GHz |
Turbo/Boost Clock | 4.0 GHz | 4.0 GHz | 4.0 GHz | 4.0 GHz | 3.8 GHz | 3.7 GHz | 4.0 GHz | 3.6 GHz | 3.7 GHz | 3.4 GHz |
Cache | 40MB | ? | ? | 20MB | 20MB | 20MB | 16MB | 16MB | 16MB | 8MB |
Memory Support | DDR4-2666 Quad Channel |
DDR4-2666 Quad Channel | DDR4-2666 Quad Channel | DDR4-2400 Dual Channel |
DDR4-2400 Dual Channel |
DDR4-2400 Dual Channel |
DDR4-2400 Dual Channel |
DDR4-2400 Dual Channel |
DDR4-2400 Dual Channel |
DDR4-2400 |
PCIe Lanes | 64 | 64 | 64 | 20 | 20 | 20 | 20 | 20 | 20 | 20 |
TDP | 180 watts | 180 watts | 180 watts? | 95 watts | 95 watts | 65 watts | 95 watts | 65 watts | 65 watts | 65 watts |
Socket | TR4 | TR4 | TR4 | AM4 | AM4 | AM4 | AM4 | AM4 | AM4 | AM4 |
Price | $999 | $799 | $549 | $499 | $399 | $329 | $249 | $219 | $189 | $169 |
Let’s not bury the lead here: the Ryzen Threadripper 1900X is the third entrant into the Threadripper family and will consist of 8-cores, 16-threads, a base clock of 3.8 GHz and a Turbo clock of 4.0 GHz, while still supporting XFR for as much as 200 MHz of additional clock speed. It will still have 64 lanes of PCI Express, four channels of DDR4 memory support, and will come with a price tag of $549.
The 1900X becomes a very interesting part for a number of reasons. Its price puts it between the Core i7-7820X and the 7800X (8-core and 6-core parts from Intel’s Skylake-X family). Even with a base clock speed of 3.8 GHz it will find itself slower than the 7820X due to lower IPC and similar clock rates. However, AMD is counting on the appeal of 64 lanes of PCIe, countering the 28 lanes on the 7820X from Intel, along with a slight cost advantage, to help it shine. The 1900X will have the same core and thread count as the Ryzen 7 family, though at higher clock speeds, a higher TDP and double the DDR4 memory channels and more than 2x the PCIe lanes. For just $50-100 more, the 1900X is a compelling option against the 1800X if you are a connectivity, storage, or multi-GPU junkie.
Continue reading about the newly announced Ryzen Threadripper 1900X!!
With these three processors in place, AMD is confident that its family will be able to compete with the Intel HEDT products with several advantages. The 1950X has a significant core/thread, cache, and PCIe advantage over the 7900X, the current highest end part from the Skylake-X family. Even the 1920X has more cores, more cache, and more PCIe lanes, though it will be much closer in performance for 20% lower consumer cost. Finally, the 1900X shares a core count with the 7820X but has 114% more PCI Express lanes at a price that is 8% lower.
We are still waiting for full reviews before coming to a final answer on the power consumption and efficiency of Threadripper against the Skylake-X family, but AMD gave us a bit more information in the slide above. Comparing the Core i9-7900X to the Threadripper 1950X, AMD shows a 24% advantage in performance with Blender while running at nearly the same power draw (2% lower, for the full system, at the wall). The result is a 29% edge in performance per watt. Obviously, this is a single data point with a single benchmark and scenario – much more testing is needed to see how efficient the high-core-count Threadrippers are going to be.
I know a lot of our readers have been trying to figure out when Threadripper is going to go on sale: now we know. The 1950X and 1920X will be on sale on August 10th with preorders starting on July 31st! The newly announced 1900X will be a bit tardy, targeted with an August 31st sale date.
- Amazon.com – Ryzen Threadripper 1950X
- Amazon.com – Ryzen Threadripper 1920X
- Amazon.com – X399 Motherboards
Other Details
AMD shared a couple of other new bits of information as well.
Rather than leaving consumers with just a single test to measure performance by, AMD has pitted both the 16-core 1950X and the 12-core 1920X against Intel’s 7900X processor in POV-Ray, Premiere Pro, Handbrake, 7-zip, and Veracrypt. The 1950X is showing 20-30% advantage while the 1920X seems to be +/-5%. Considering the $200 price advantage the 1920X has over the Intel CPU, as well as the PCIe lanes increase, this will be a popular selection for a lot of users.
AMD was touting its partners and the preparedness of the ecosystem ahead of the Threadripper availability on the 10th. AMD is very aware of the issues that haunted them for weeks and months after the Ryzen 7 release with poor motherboard implementations and the teams are eager to avoid that fate this time around. To be fair, this is still a very rushed launch from what the board partners are telling me, but initial testing appears to be more stable.
Finally, AMD did mention that there would be a vibrant ecosystem of air and liquid coolers ready for the Threadripper release on August 10th. AMD is including a bracket with the CPU that will adapt the Asetek-built water blocks (with the tooth-style retention) for the TR4 socket.
Die Count Mystery Solved!
Just before the Threadripper systems started shipping to reviewers, overclocked der8auer delidded a Threadripper CPU and found four dies under the heatspreader. The assertion was that all were real dies with two of them disabled to give us 16-cores. That would have been a HUGE WASTE if in fact AMD was simply taking the EPYC design and fusing off two dies.
Source: Techpowerup
As it turns out, AMD confirmed to us that only two of the dies on that substrate have working transistors on them and the other two are simply blank wafer pieces to help with the structural support of the heatspreader! In otherwords, there are two blank, worthess chunks of silicon on the part to keep the top from caving in. (Note that the two working dies are always going to be diagonal from each other.)
We are just days away from the full release of reviews where we will be able to discuss performance in all aspects, including gaming, lightly threaded applications, mega-tasking, and more. It might be early to say it, but I believe AMD has a winning product on its hands and a large community of consumers that is ready to take on a newcomer into the HEDT field.
Can you imagine 2017 without
Can you imagine 2017 without AMD …. what a boring year it would be.
Curious to see how an
Curious to see how an overclocked 8-core compares on desktop vs this 8-core.
Theoretically we might gain a slightly higher overclock since half the cores are separate and cooling is easier.
On the other hand, the looser integration should create potential latency issues which sometimes can have a HUGE impact on performance in things like games.
The 1900X doesn’t appeal to me much beyond the potential CPU upgrade if you need even more cores. I’m not sure many people would need the extra PCIe lanes, and DDR4 isn’t a big issue either for an 8-core CPU (3200MHz Dual is sufficient).
Anybody know the difference
Anybody know the difference between EPYC and THREADRIPPER?
I realize EPYC is server, thus will go up to 32-core because lower frequency with more cores makes sense for server, whereas on desktop we still need the performance per clock which limits the cores we can feasibly cool.
I would GUESS that they are nearly identical with perhaps some cache changes to optimize for the platform they are on?
Also curious if the 16-core Threadripper will be 4x4C, or 2x8C.
EPYC is locked. Threadripper
EPYC is locked. Threadripper has an unlocked multiplier.
EPYC core clocks will be lower (due to higher core count and thermal considerations. Threadripper will boast the highest clock speed overall…not counting overclocking “potential.”
EPYC is supported in 2 socket configuration. Threadripper is 1 processor socket only.
EPYC has 128 lanes of PCIe. 64 are used for processor to processor communication when two processors are installed. Threadripper is limited to 64 PCIe lanes, 4 of which aren’t usable by PCIe AICs.
EPYC is up to 8-channel DDR4. Threadripper is max 4-channel DDR4.
EPYC motherboards will be available in multi-socket configurations. Threadripper motherboards are 1 CPU socket.
Threadripper 16-core should be 2 dies at 8 cores each for 16 cores. Nothing at all has indicated they will produce any model with 4×4, as Infinity Fabric would technically slow the processor down in highly multithreaded workloads where more than 8 cores are in use.
The 12-core would therefore be 2 dies at 6 cores with 2 cores
disabled per die.
Epyc is not Up to 8 channels,
Epyc is not Up to 8 channels, it’s 8 channels across all of the Epyc SKUs!
Yes Epyc has 128 PCIe lanes available no matter the single socket(1P) or the Dual socket(2P) motherboard SKU!
Yes that Epyc 7401P’s 24 cores/48 threads are going to be clocked lower than Threadripper’s(Unlocked cores are), but that Epyc 7401P comes with a 3 year warrenty and the professional level of extended support that AMD has to give for the professional market. And the Epyc 7401P at $1075 is the workstation Price/Performance leader for an SKU with 24 cores/48 threads feeding from 8 memory channels and over those 128 PCI lanes to all manner of multi-GPU and PCIe NVM Raid configurations that any Threadripper SKU can not match.
That includes the ablitiy on the part of the Epyc 7401P’s owner to populate the Epyc motherboard’s 16(2 per memory channel) DIMM slots with single Rank 4GB or 8GB DIMMS across half the 16 slots(One DIMM per memory channel) for some 2400MHz/2666MHz at great RAM timings performance configurations for maximum bandwidth and the best letency on those memory timings metrics that any 4 channel Thread Ripper Motherboard SKUs can not match.
And the Epyc 7401p CPU SKU only cost $76 dollars more, Threadripper 16/32 is beat by the Epyc’s 7401P’s 24/48 cores for any workstation Blender, Solidworks, other rendering workloads. And The 7401P and its little(Big Actually) mothernoard friends are all fully certified for ECC usage, that’s FULLY Certified and Tested to work with ECC memory, plus longer warrentys.
Threadrpper can not touch the Epyc 7401P’s 24 cores/48 threads at only $76 more for the Epyc SKU over Threadripper’s 16/32 half the memory channels/half the PCIe lanes. Epyc is Epic for the workstation users!
And the Epyc 7401p CPU SKU
And the Epyc 7401p CPU SKU only cost $76 dollars more, Threadripper 16/32 is beat by the Epyc’s 7401P’s 24/48 cores for any workstation Blender, Solidworks, other rendering workloads.
I would like to see the benchmarks supporting this claim.
50% more cores doesn’t compete with 70% higher clocks unless you have saturated the memory… threadripper probably supports over 2666 in overclocked mode.
Yes EPYC will have it’s place in the server realm… but to say that it is better than threadripper for workstation tasks… show me the numbers.
Threadripper is not Fully
Threadripper is not Fully TESTED/certified for ECC memory! So that’s enough to remove Threadripper from any workstation branding! Workstation usage implies professional usage, and professional usage requires fully certified ECC support in the CPU and motherboard SKUs. See that ->Work<- in Workstation name, that's what implies the fully certified for ECC support needed. 24 cores and 48 threads are going to do great on non gaming graphics rendering workloads for the Epyc 7401P and its full 8 channels to DDR4 memory, ditto for those 128 PCIe lanes on some very large SSD based NVM RAID arrays and for pleny of available PCIe 3.0 x16 slots populated with Radeon WX branded GPUs. I TOO await the benchmarks but the 24/48 cores/threads to 4 more channels(8 total) worth of memory bandwidth says that the Epyc 7401P is the workstation SKU of choice at a very affordable price relative to Intel's costly 24 core Xeon SKUs. AMD has really made its real professional Epyc SKUs near as affroable as its consumer Zen/Ryzen/Threadripper SKUs. So there is no monotary savings need to use Threadripper as a pseudo workstation SKU just to save money with that 24 core Real Professional Epyc 7401P's 24/48 cores/threads SKU only $76 more than a 16 core Threadripper CPU SKU's cost! The Epyc 7401P's price/core metric even beats Threadrippers price/core metric. Do that math.
No. It’s up to 8 channels.
No. It’s up to 8 channels. You must populate memory banks in order to gain access to those channels. You don’t automatically get 8 channels. No system works that way unless it requires all slots to be occupied by DIMMS.
Thus, “up to 8 channels.”
Total system cost is going to be higher on EPYC, without a doubt.
It’s not up to for the
It’s not up to for the motherboard maker to offer the end user, that’s not what up to implies. And the devices’ end user can affordably populate theos 8 memory channels with the more affordable unregistered 4GB ECC UDIMMs and still get at least 32GB of usable RAM, Threadrpper will have to use non certified for ECC regular 8GB DIMMs across its only 4 memory channels to get a not so useful for ECC workloads 32GB on RAM. IF It’s not Fully certified for ECC then it’s not a real workstation SKU IT’s only a Pseudo Workstation SKU.
The Epyc 7401P price at $76 dollars more for 24 cores/48 threads says AMD’s customers don’t have to worry about not having the money to get a real workstation SKU like Intel’s Xeon customers sometimes have to attampt to do because Intel’s Xeon SKUs are so overpriced relative to Intel’s consumer offerings at that 24/48 core/thread count range.
The Epyc 7401P’s price/performance value beats Threadripper and Relly beats any 24 core Xeon Proce/performace metrics.
Total system cost will always be higher for workstation usage because of ECC memory pricing, but factor in the 8 channel memory population options on any Epyc motherboard and those 128 PCIe lanes and that professional 3 year warrenty (Epyc CPU and Epyc motherboard pro warrenties) and that seals the deal for the Epyc 7401P and professionl workstation usage that is hands down the best professional workstation deal going for graphics pros with not to much dosh to spend.
You have issues.
You have issues.
Lots of issues.
…and no.
Lots of issues.
…and no. I’m not wrong about “up to 8 channels.” You’re arguing and then changing the argument because you’re a troll.
There’s no need for this. I answered a question, you’re arguing…for what reason, I have no idea.
Nowhere can I find where it
Nowhere can I find where it says ECC is a must for a workstation… I do see where its basically useless for anyone who uses the workstation for CAD.
It’s a troll account that
It’s a troll account that mentions and tries to promote EPYC on pretty much any Threadripper article. It doesn’t need to back itself up with facts.
If those 8 core are spread
If those 8 core are spread across those 4 Zeppelin dies then that’s a lot of dark silicon with which to absorb the heat Phonons across the dies. Hell even cores spread across 2 CCX units on the same die if some of the cores are disabled on the same CCX units on the same Zeppelin die that’s still going to have more dark silicon to absorbe and transfer heat away from that active silicon parts that are producing the heat.
So that’s maybe the question and AMD still needs to put 4 dies into that MCM to keep the package/heatspreader from warping even if two of the 4 Zepelin dies are totaly dead and used to fill the space under that giant metal lid/heat-spreader.
That TR 1900X and 8 Zen cores under that giant hunk of metal with 4 full memory channels is going to have some better performance metrics on any workloads that need the extra memory bandwidth. And those cores are, with all that dark silicon and indium based solder, going to have the most non active material to absorbe and transfer heat. So some more overclocking of an 8 core threadripper SKU is going to yield better gaming results.
The CCX unit to CCX unit latency on the same die and across dies is relatively fixed and the there is the question of available L3 cache to fewer cores and what core to cache ratios can be offered to the remaining cores across the CCX units on the various die configuration possibilities that are available to AMD with that 8 core Threadripper variant.
It may be that it is just as
It may be that it is just as likely the following:
It’s cheaper to produce one package with four dies, disabling two of them prior to QC, than it is to produce one package with two “real” dies, and then either blank space or “dummy” dies in place of real dies. That would require separate tooling and line time during production.
Any Zeppelin dies with borked
Any Zeppelin dies with borked memory controllers are worthless and can be used for dummy dies, and there are probably enough Zeppelin dies with such defects even at 80% yield rates for all Zeppelin production to accout for the necessary dummy/dead dies. AMD is using the same Zeppelin die across all of the Ryzen 3/5/7, Threadripper, and Epyc production and that small fraction of the defective 20% represents enough nembers of fully unusable Zeppelin dies for that specific purpose on threadripper, if AMD is in fact needing totally dead/uselss Zeppelin dies as filler dies.
The 8 core Threadripper has a
The 8 core Threadripper has a lot more cache and memory bandwidth per core, but you only get 2 cores per CCX. If you have 8 threads that want to share data, they would hit a lot of interprocessor communication latency. The same 8 threads could run on a single Ryzen CCX (4 cores/8 threads). It may not be a clear victory for 8 core Threadripper, at least for some applications, until there is a lot more software optimization for such architectures. Some applications that don’t share data between threads should perform much better on TR due to the larger per core caches and increased memory bandwidth. We will have to wait for benchmarks. I don’t expect it to play well with Nvidia drivers, and since I have no plans to buy an Nvidia card, those benchmarks will be irrelevant to me anyway.
You REALLY need to fix the
You REALLY need to fix the tables layout
Just checked it, it DOES NOT
Just checked it, it DOES NOT happen on Chrome, on Firefox the last two columns of the tables are behind the elements of the right bar
Yeah. Please PC Per, view
Yeah. Please PC Per, view this page in Firefox and see what is happening.
We know … but are choosing
We know … but are choosing to burn the whole place down and rebuild as opposed to tweaking just one thing.
Bear with us for now, if you could.
Chur Jeremy
Chur Jeremy
Join the rest of us who have
Join the rest of us who have forsaken firefucks
Nice write up Ryan!
Super
Nice write up Ryan!
Super intrigued by the 1900X. It will be interesting to see what effect the quad channel memory has based on how memory dependent the 1800X is in gaming. 64 lanes of PCI-E is awesome. It feels like Intel has been sandbagging us with the PCI-E lanes. On Skylake-X you have to spend at least $999 to get more than 28 lanes and even then it’s only 44. I’m not at all interested in spend $999 on a processor alone.
I’ve been happy with my Intel platform, but I love seeing AMD make things interesting!
Yes with directx 12 being the
Yes with directx 12 being the API of the future there is no need for all those pci express lanes. Most games don’t even code for SLI/ Crossfire anymore. Dx11 version which is “inferior” still has much better support for multiple cards. TR better have something else going for it than more lanes.
If priced well it might take some of the market from Intel but at $550 IDK. I’m a gamer and the most I’ve spent on a processor is $330 on my 4770k. Better to spend more money on a better video card than break the bank on a processor though.
Productivity you say. Most of those bought Ryzen for that. Are they going to sell them on the used market and get TR instead?
Competition is good. It might get Intel to lower their prices.
What does DX12 have to do
What does DX12 have to do with PCIe and the definite lane advantage that AMD’s threadripper offers any PC owner over Intel’s segemented paltforms with various degrees of PCIe lanes offered(More Lanes more $$$$). PCIe lanes are needed now more than ever and those NVM/SSD can make use of 4+ lanes of PCIe. So for SSD RAID usage alone the more PCIe the better and Multi-GPU managed by the DX12 and Vulkan APIs will still see the need to give those multi-GPUs plugged into the PCIe slots the with most available PCIe bandwidth, that need does not disappear with the DX12/Vulkan APIs.
For any New Vega mciro-arch based GPUs there will be maybe more need for PCIe than before if a discrete Vega GPU SKU is utilizing its own paged to SSD Virtual Video Memory page swap space on the PC SSD/s for Textures and mesh/other data. Vega’s HBCC/HBC will use the HBM2(Of whatever size) as a last level cache to the Vega GPU, so that’s has to benifit from more XDMA/DMA types of transfers over the PCIe lanes’ available bandwidth into/out of the Vega GPU’s HBCC/HBC-HBM2 video memory/cache.
The more PCIe full x16 lanes the better for having enough PCIe available bandwidth to and from the GPU/s.
Anyone know how we’re going
Anyone know how we’re going to cool these things yet? No heatsinks seem to be available as of right now, just a few CLCs.
It comes with a bracket for
It comes with a bracket for asetek coolers, it is recommended to use a CLC or custom loop.
You should not need to use
You should not need to use water coolers for these. While the power consumption is high, the thermal density is low since it is spread across multiple die with space between them. The TR parts are aimed at HEDT market though, so water coolers are probably given priority over air coolers. There are plenty of air coolers that can handle the power consumption, but the base plate and/or heat pipe contact area needs to be large enough to cover both die. The lid on the processor (not really a heat spreader) can’t transfer much heat sideways so the cooler needs to cover the area of both die completely, and some existing coolers may not do that.
I don’t know if I would
I don’t know if I would recommend the low core count versions of Epyc or Threadripper for most uses. They could have some specific use cases though. TR is 2 chips with 2 CCX each, so 4 CCX total. An 8 core TR only has 2 cores per CCX. That gives you only 4 threads per each L3 cache. If you have an application with a lot of shared data between threads, that could be a serious limitation. With the way the Nvidia driver works (lots of inter-thread communication), I wouldn’t expect it to work well. Anyone know how many threads the Nvidia driver keeps active? It has to distribute draw calls to multiple threads for processing, and then combine them back together in a single thread to submit to the GPU under DX11. That will involve a lot of data sharing. AMD parts with games that use DX12 with some optimizations might work well since any thread can submit work to the GPU independently.
The 8 core Epyc chip is even worse, with only one active core per CCX. You only get 2 threads per CCX, so data sharing would be a bottleneck without significant optimization. Running multiple independent applications or VMs would still work well. It would also be good for applications where the CPU is just throwing data at the GPUs. Once we get better software optimization for these architectures it will not be as big of an issue. If the application requires a lot of data sharing between threads though, it may take significant optimization effort to make it work well with such a small number of threads available in a zone. It will be interesting for some types of memory bound applications due to the available bandwidth and the large amount of cache available per core.
It will be interesting to compare 8 core Ryzen, 8 core TR, and 8 core Epyc, if all of them are available. Some applications might perform significantly better with the extra cache and memory bandwidth, reguardless of the interprocessor communication overheads.
So the ThreadRipper 1900x is
So the ThreadRipper 1900x is 2 4-core CCXs and not 4 2-core CCXs?
I’m curious about the 1900x’s
I’m curious about the 1900x’s TDP. If it’s lower than it’s brethren, I’ll buy it. I don’t wand to exchange the PCI lanes for twice the wattage, personally. Also, can half of them be disabled along with their memory channels for daily use, and activated when productivity is required? This is for a home PC that is used 1 or 2 days a week with heavy workload, with the rest of the week split between web surfing or gaming (GTX 1080Ti). Electricity cost is high where I live, along with a year round A/C bill.
I may opt for the 1920x over the 1800x at these prices. I really like the quad channel memory and the ability to handle my 5 hard drives and many USB devices more efficiently. Certainly a compelling option. There’s also a potential price drop for the 1800x on Aug. 31.