AMD Computex 2019 Livestream

More, comparing to Intel's technology as directly as possible as well as comparing to price.

Which is exactly what they did. The 3800X competes favorably with the 9900K on features and performance while beating it on price and power. Would you have preferred they not showcase the 3700X at all? Or would you have them line the 3800x against HEDT products so the 3700X could lose to the 9900K?

Remember that AMD still has to get the channel supplied, and we can't buy these things yet- they've given Intel plenty of time to respond if Intel hasn't already gotten that ready.

What's Intel going to respond with? If they add hyper-threading to the 9700K, then you practically have a 9900K.

So this is marketing. It's not something that we can reliably predict what will be the best price / performance etc. come retail availability as announced.

Okay. So company's shoudn't try to market their products because a competitor might have something up their sleeves?
 
Which is exactly what they did. The 3800X competes favorably with the 9900K on features and performance while beating it on price and power. Would you have preferred they not showcase the 3700X at all? Or would you have them line the 3800x against HEDT products so the 3700X could lose to the 9900K?



What's Intel going to respond with? If they add hyper-threading to the 9700K, then you practically have a 9900K.

As I mentioned in another thread, Intel should have a locked 9900 with HT, just like there was a locked 8700 with HT.
 
The 3800X competes favorably with the 9900K on features and performance

We know it competes favorably to the 9700K, according to AMD...

What's Intel going to respond with? If they add hyper-threading to the 9700K, then you practically have a 9900K.

See #162 above

Okay. So company's shoudn't try to market their products because a competitor might have something up their sleeves?

I said compare on technology and price. They've given Intel over a month to respond; we've heard that Intel has 10-core SKUs coming, and they can do a whole lot with existing SKUs and parts.

I expect Intel's 5GHz parts to edge ahead of any of the AMD CPUs detailed in single-core, and a 10-core Intel part to be faster than any of them overall, based on AMD's numbers. Intel can adjust their lineup so that they're more directly competitive by the time these CPUs are released.

You, I, and AMD know this. So yes, AMD should 'market' their products, but they (and we) shouldn't pretend that what they put on their marketing slides will resemble ground truth when their products hit reviewers.
 
I have to admit, I am still thoroughly confused about the x570 chipset and its lanes.

Prior to launch, rumours were saying 40 lanes. Well, that obviously didn't happen. We get 16 total, 4 of which are used for the upstream to the CPU, so 12 available.

Well, that's kind of a bummer.

Then MSI comes along ad displays something like this?

IMG_20190526_111502_575px.jpg


The claim is:

upload_2019-5-27_21-31-53.png


How on earth do you do that with only 12 lanes? Are they adding their own PCIe/PLX switch downstream to the chipset?
 
...

I expect Intel's 5GHz parts to edge ahead of any of the AMD CPUs detailed in single-core, and a 10-core Intel part to be faster than any of them overall, based on AMD's numbers. Intel can adjust their lineup so that they're more directly competitive by the time these CPUs are released.
.

And how do you expect this? Assuming similar IPC, that 10 core Intel part would have to running at 5.5 ghz or so to match the 3900x in MT.
 
cageymaru I'd actually update the timestamp to 36m30s, which is when he alludes to OCing being similar to AM3+.
I mean if Buildzoid is excited about it, I am even more interested now :p

Is he referring to CPU-NB or HTT? Would be nice to be challenged for once.
 
It does not matter which one I run it on DX12 or Vulkan it is 1080p Ultra plus 110% to load the gpu's up some .. but there stock clocked so RX 570 XFire can do it easy to the RTX 2070 on 10 year old platform in MP on line as where we need to get back to and control pricing some .

also I may try to install a 3rd RX 570 if my power supply can take it .

 
Last edited:
I have to admit, I am still thoroughly confused about the x570 chipset and its lanes.

Prior to launch, rumours were saying 40 lanes. Well, that obviously didn't happen. We get 16 total, 4 of which are used for the upstream to the CPU, so 12 available.

Well, that's kind of a bummer.

Then MSI comes along ad displays something like this?

-

The claim is:


-
How on earth do you do that with only 12 lanes? Are they adding their own PCIe/PLX switch downstream to the chipset?


I thought it was 24 lanes?
 
I thought it was 24 lanes?

CPU has 24 lanes.
- 16x goes to GPU1 slot
- 4x can be used as either M.2 or PCIe slot direct to CPU on motherboard
- 4x go to chipset.

Chipset has 16 lanes in total. 4x are used by the uplink to the CPU. This leaves 12x to share the 4x bandwidth to the CPU. these 12x are shared between any on board devices (LAN, Audio, etc) and any m.2 or PCIe slots that come from the chipset.

So, in theory you wind up with 20x usable lanes off of the CPU, and 12x usable lanes off of the chipset, but the 12x off of th echipset share 4x worth of bandwidth going to the CPU. This is not quite as bad as it might sound, because it's 4x of Gen4 bandwdith. This is equivalent to 8x Gen3, 16x Gen2 or 32x Gen1
 
  • Like
Reactions: GHRTW
like this
How on earth do you do that with only 12 lanes? Are they adding their own PCIe/PLX switch downstream to the chipset?

I see:
  • 16x to GPU
  • 4x to CPU NVMe
  • 4x to chipset
That's what was speculated beforehand; so yeah, 16 + 4 = 20, which means that if there are 40 total exposed, then the four to the chipset need to be broken out to 20. And that would definitely mean some sort of PCIe switching, if that is indeed the case.

What's notably missing here is the breakout of PCIe lanes for the chipset.
 
CPU has 24 lanes.
- 16x goes to GPU1 slot
- 4x can be used as either M.2 or PCIe slot direct to CPU on motherboard
- 4x go to chipset.

Chipset has 16 lanes in total. 4x are used by the uplink to the CPU. This leaves 12x to share the 4x bandwidth to the CPU. these 12x are shared between any on board devices (LAN, Audio, etc) and any m.2 or PCIe slots that come from the chipset.

So, in theory you wind up with 20x usable lanes off of the CPU, and 12x usable lanes off of the chipset, but the 12x off of th echipset share 4x worth of bandwidth going to the CPU. This is not quite as bad as it might sound, because it's 4x of Gen4 bandwdith. This is equivalent to 8x Gen3, 16x Gen2 or 32x Gen1

That's plenty for what that board claims.

from the CPU x16 lanes go to the GPU, split by 2 for 2 way sli (x8 per card), or by 4 for 4 way crossfire (x4), x4 lanes go to one of the m.2 slots most likely, then the other x4 are interconnect for the chipset.
Then the chipset has 16 lanes, minus 4 to the CPU, so you have x12. Probably x4 to each of the other 2 m.2 slots, leaving x4 for *ahem* other things like maybe one of the LAN ports, wifi, BT,
SATA and USB won't take up lanes from the chipset,

and no they aren't shared between all onboard devices. some go directly the chipset, some directly to CPU, and others don't interface on PCI-E
 
Last edited by a moderator:
I see:
  • 16x to GPU
  • 4x to CPU NVMe
  • 4x to chipset
That's what was speculated beforehand; so yeah, 16 + 4 = 20, which means that if there are 40 total exposed, then the four to the chipset need to be broken out to 20. And that would definitely mean some sort of PCIe switching, if that is indeed the case.

What's notably missing here is the breakout of PCIe lanes for the chipset.


Maybe they are switching the primary 16x Gen 4 lanes to two 16x Gen 3 lanes? That would make sense considering most GPU's on the market are still Gen 3, and I'd imagine there would be little to no benefit from moving to Gen4 at this point.
 
Maybe they are switching the primary 16x Gen 4 lanes to two 16x Gen 3 lanes? That would make sense considering most GPU's on the market are still Gen 3, and I'd imagine there would be little to no benefit from moving to Gen4 at this point.

Not impossible, but this switching would likely require more hardware, and hardware that has lately been priced higher than the highest cost boards released with X570.
 
Not impossible, but this switching would likely require more hardware, and hardware that has lately been priced higher than the highest cost boards released with X570.
Isn't this board supposed to be like $800?
 
Isn't this board supposed to be like $800?

I heard ~US$600 for one of the top-end boards, Gigabyte I think, from GN. That's a lot, but PLX switches are priced exclusively for enterprise these days, and for enterprise, that's peanuts.

And I'm making the assumption that AMD didn't build their own discrete PCIe switch because they'd likely run into patent issues (which is what I assume is keeping PLX switch prices protected), and that adding such switching to the CPU would exceed the sockets capabilities.
 
I heard ~US$600 for one of the top-end boards, Gigabyte I think, from GN. That's a lot, but PLX switches are priced exclusively for enterprise these days, and for enterprise, that's peanuts.

And I'm making the assumption that AMD didn't build their own discrete PCIe switch because they'd likely run into patent issues (which is what I assume is keeping PLX switch prices protected), and that adding such switching to the CPU would exceed the sockets capabilities.

Well we have seen some leaked documents that show switching.
AMD-X570-Chipset-Block-Diagram.jpg

And

X570_PCIe_Lanes.jpg


Both are unnoficial leaks from before computex so take with a grain of salt
 
Seems pretty cheap to me, but then maybe these aren't the right chips?

From that page, this guy at ~US$100 seems like it might fit as an example. That's the only one with >20 PCIe lanes. And it only supports PCIe 2.0.

Both are unnoficial leaks from before computex so take with a grain of salt

These do indeed portray PCIe switching, but not really what we're looking for; they show switching between a 4x slot, M.2 slot, and SATA ports.

What was being leaked / hoped for was additional chipset PCIe slots on the switch, but more so the ability to take the split 16 CPU lanes for GPUs and convert 8x PCIe 4.0 lanes to 16x PCIe 3.0 lanes, and that functionality is definitely missing.
 
From that page, this guy at ~US$100 seems like it might fit as an example. That's the only one with >20 PCIe lanes. And it only supports PCIe 2.0.



These do indeed portray PCIe switching, but not really what we're looking for; they show switching between a 4x slot, M.2 slot, and SATA ports.

What was being leaked / hoped for was additional chipset PCIe slots on the switch, but more so the ability to take the split 16 CPU lanes for GPUs and convert 8x PCIe 4.0 lanes to 16x PCIe 3.0 lanes, and that functionality is definitely missing.
There's this guy from broadcom with 24 gen3 lanes for a little over $100. They aren't cheap, for sure, but on a $300-500 board I'd expect an expensive chip or two, and they might get a better deal elsewhere (or direct from the manufacturer).
 
That's plenty for what that board claims.

from the CPU x16 lanes go to the GPU, split by 2 for 2 way sli (x8 per card), or by 4 for 4 way crossfire (x4), x4 lanes go to one of the m.2 slots most likely, then the other x4 are interconnect for the chipset.
Then the chipset has 16 lanes, minus 4 to the CPU, so you have x12. Probably x4 to each of the other 2 m.2 slots, leaving x4 for *ahem* other things like maybe one of the LAN ports, wifi, BT,
SATA and USB won't take up lanes from the chipset,

and no they aren't shared between all onboard devices. some go directly the chipset, some directly to CPU, and others don't interface on PCI-E


Well, as long as I get enough lanes to use:

- 16x for GPU
- Two 4x NVMe drives
- 8x Gen2 Dual SFP+ 10gig NIC

I'll be happy. That's all I really need.
 
With 3200c14, yes they were ~ to Skylake....Lock the clocks and run the tests yourself and you will see. It's been mentioned in MANY MANY TIMES but th IDF ignores it.

Probably got lost because it wasn't enough to make the chip competitive in per-core performance, due to the clock speeds being so much lower.
 
some of the manufactures are splitting the 16 lanes from the cpu to dual x16 slots to support SLI/CFX still so it'll be x16/0 x8/x8 depending on your use case. then you have 4 to one m.2 then 8 from the chipset to the other two m.2 slots and 4 split across the x1 and remaining 2 x16 slots running at x4
This Twitter post from ASRock has a video. X570AQUA pron. :)




GN did a video talking about it and other asrock boards.
 
No, it is 10% faster in Strange Brigade.

Average performance is still unknown.

The fact they lacked the confidence to not resort to a single, cherrypicked, relatively obscure title that's either super-AMD optimized, or just weirdly Nvidia-bugged, does provide a huge clue about where average performance is not going to be. On the bright side they still have some time to optimize drivers.
 
hopefully e3 is where they show off navi in more than one game but we'll see..
 
frame rates

Frametimes- the distinction is important, as 30FPS can be smoother than 100FPS- and 100FPS made to feel like 5FPS. Measuring frames in a second is almost useless if you're not also paying attention to the rate of delivery :D.
 
Frametimes- the distinction is important, as 30FPS can be smoother than 100FPS- and 100FPS made to feel like 5FPS. Measuring frames in a second is almost useless if you're not also paying attention to the rate of delivery :D.
I feel a smidge mis-quoted but yeah that too.
 
Easy to do if you are GPU limited.... There isn’t enough data there to go on.
Not in that game. Pubg equal performance is pretty impressive considering how far behind a 2700x is compared to a 9900k in it
 
Back
Top