Threadripper

Threadripper is just several smaller processor dies on one chip, while Ryzen is just one (or latter two). So it will just get more cores.

Coco Chanel

Attached: week-5-coco-cover-pic-1024x624.jpg (1024x624, 51K)

But TR already has quad channel support.

LOOK IT'S A CHART ON THE INTERNET MUST BE REAL

Butthurt fanboy?
If it's not accurate (real), there's no loss, since it didn't exist in the first place.

Eight memory channels, 128 pcie 4.0 lanes, 64 cores and 128 threads, infinite virtual memory address.

Attached: 1506977929561.png (200x200, 45K)

not going to happen

Here are the Singaporean prices.

Attached: file.png (1844x802, 600K)

A free oven.

Attached: 1544385127448.jpg (1536x2048, 470K)

QUICK INTELBROS DAMAGE CONTROL HARDER

12 cores at 1.25 is pretty comfy and its a great backup chip for kikeripper gen 3

With Ryzen going the chiplet road I see no reason for Threadripper to continue existing. Ryzen and Ebin should cover all use cases.

HEDT is a meme Intel created to rip people off anyway, I hope it dies

>so what will TR get?
a significant discount on the older models

It's going to be nonworthwhile gains to not step on EBYNs toes, so it'd have to be between 16 cores and 32 cores. Probably have some aspects from EBYN2, maybe DDR5 support.

simply ebin ;DDD

except for the collective hours AMD fangirls have been screeching online using this list, telling me to BTFO cause I didn't feel like selling my shit to upgrade(downgrade) my computer.

It moves upmarket and cedes market share to Ryzen. Especially with PCIE 4.0 having twice the bandwidth per lane, the fewer lanes won't be as big a problem.

> more memory channels
> More PCIe lanes
> More cores
Like it always has. There are some rigs I'd run increased core ryzen on and others that would stay on TR4 because I max out the PCIE lanes on those builds and ryzen has insufficient memory/pcie lanes

Except that EBYN2 will also benefit from ALL THOSE PRIME SILICON CORES m8.

If Ryzen3 is going to 16, then TR3 is 20 to 28, EPYC2 is starting in the 30's instead of topping out at 32 per chip.

Or they'll delete TR all together.

TR3 is already at Max(32)
TR stands alone on
> increased memory channels
> pcie lanes
Cores need to be fed.

Its literally not real, its literally the fake adoredtv list reposted word for word. It literally isnt real. If amd was sitting on 5ghz chips, we would know. The whole fact of the matter is that they didn't discuss clockspeed in their CES benchmark at all, so it's pretty obvious that they were overclocking a binned chip. Zen already has an IPC and power lead over intel, so the chip they used was probably clocked around 4.7 compared to the 9900k's 5.0. Which is honestly reasonable progress over the 2700x.

Im and amd fag but ppl need to chill the fuck out. The leaks are fake, the memes about $500 16 core 5ghz cpus need to fucking stop because its not gonna happen this generation. That said zen 2 is still going to be really good.

>Boost 5niggahertz
Nah

TR3 will probably still cap out at 32c and not anything outrageous like 64c, given the absurdity of trying to feed the latter on 4 channels alone.
That said, compared to their 14nm predecessors, they will:
- have much better clocks in their 250W TDP envelope due to 7 nm
- will have 2x the L3 per core
- >= 2x AVX throughput per clock
- have 2x IO bandwidth from PCIe 4.0 (should any consumer-tier peripherals ever materialize)
- will be UMA, are more importantly won't have any no-mem nodes for the >16c parts, so NUMA-unaware software won't be fucked anymore

well...
1) you want more pcie lanes, this shit is expensive, so a clear consumer v prosumer deliniation
2) you need more ram/faster again, a clear deliniation
3) You need a fast cpu, here is a point away from epyc, with ryzen, you get a fast consumer product, with rpyc you get a fast in multithread product, with threadripper, you get a middle between the two, if not faster then ryzen on the thread ripper side.

there is a clear use case for separating consumer and hedt, cost of manufacturing, and a clear reason to go hedt over server, faster single core.

keep in mind, server does not get the best dies, what it gets are the best dies at the lowest voltages. this doesn't necessarily mean they get fast dies, they just get dies that perform VERY well under lower voltage/power, you know, power, that thing that is one of the biggest factors of consideration for data centers?

thread ripper and top tier ryzen get the best cpu cores in terms of peak speed

epyc gets the best cores in terms of lower power for highest speed.

no, there are clear server use cases that demand lower core counts but just higher io, they will sell chips that are aimed at that segment.

also will be fully using butter doughnut.

Threadripper is not necessary anymore.

Attached: SIXTY FOUR.jpg (480x360, 16K)

AVX 1024 coprocessors (2 of them)

is there even a use case for TR?
it's only good at cinebench wanking or unzipping folders.
adding 32 cores with shit memory bandwidth and shit overclocking accomplished nothing.
professionals all use X299 anyways.

>X299
Dead socket, not sure professionals are gonna buy shit chips at crazy price together with mobos again.

A tripper is good for any serious computing be that 3d stuff, video stuff or engineering or scientific stuff since it also supports ECC.

And by good i mean roughly 3-4 times better than incel shit chips that cost more.

Just look at reviews anywhere. People are glad they can move from Xeon EPs and Huanans on 1366/2011 to an equally cheap and modern platform. Although by
> professionals all use X299 anyways
I can tell you're a shill.

I'm glad you added the part of you being an amd fag so we don't have to point out rest of the retardation in your post.

>Its literally fake
Its unconfirmed and might be not entirely accurate BUT the ballpark figs appear reasonable.
>If amd was sitting on 5ghz chips we would know
The chips dont currently exist, amd will tell you when they actually get them.
HOWEVER 5Ghz single core boost on the select few best quality chips looks absolutely believable.
2700x gets 4.3-4.4 on a 14nm+ LOW POWER MOBILE node.
Ryzen 2 is fabbed at the bleeding edge high power nose.
>They did not discuss clocks at CES
They only demonstrated EPYC and clocks arent a factor or a marketing point in server segment.
>The leaks are fake
The leaks are neither true or false until confirmed.
However i expect them to be close to truth.

Threadripper will probably be offered in 32, 48, and 64 core. Epyc Rome has a 64 core variant, and if history repeats itself, we will have a 64 core Threadripper. The problem is, you'll be limited to 4 RAM channels again, hindering performance massively in comparison. Then again, a 32 core (4x8 zen 2 chiplet) with 4 RAM channels should be monster!

Attached: cheers.jpg (225x225, 9K)

>128 pcie 4.0 lanes
I came

delid dis

Jow Forums you dissapoint

yes, thank, you

dunno about octo channel memory, but l1, l2 and l3 might have a nice bump, also pci lanes will probably increase from their current 60 up to anywhere under Rome's 128

en.wikichip.org/wiki/amd/ryzen_threadripper

That and pcie lanes will be the only major difference really

but those are just the disclosed, already known, Rome specs.

Attached: ef3a4ec33d.png (853x457, 54K)

24c and 32c - TR
32c, 48c, 56c, 64c - EPYC Rome

not to mention that epyc 7000 series already had 128 lanes, gen 3 though.

TR - up to 48 cores, maybe even 52 or 56
I doubt they'll repeat max core count, and that's just because they can afford to not repeat it and it looks cool -- i.e., more for marketing purposes rather than technical reasons.

>Epyc Rome has a 64 core variant, and if history repeats itself, we will have a 64 core Threadripper

the Epyc 7000 series already had a 64c variant, why would history be repeating itself in this case? Rome will also be up to 64c/128t, so if there was any repetition we would have another 32c/64 tTreadripper 3 at it's max core count.

I wanna see some boards with a bunch of U.2 connectors

I'm just on the edge of pcie lanes
i might upgrade my NAS to epyc pretty soon because a fuckload of u.2 storage is interesting

KANTOOOOOOOOOOOOOO

24 and 32 cores.

>up to 2666
AHAHAHAAH AYYMD FAGS BTFO

It already had QC.

>make moar corez

That is ECC memory you fucking dumbass. Memory error correction is more important than memory speed for servers.

Attached: 1551034123589.jpg (777x578, 40K)

>Memory error correction is more important than memory speed for servers.
why not both you mongoloid? unless you're doing actual rocket science, memory speed is much more important as errors in memory almost never occur.