AMD Readies X870E Chipset to Launch Alongside First Ryzen 9000 "Granite Ridge" CPUs

[AusWolf]

i've been looking at this but still to expensive need it to hit the CAD $350 range which probably won't happen until the new boards launch

View attachment 333309

X670 is unreasonably expensive. I'd rather look at B650 unless you absolutely can't live without the extra PCI-e lanes.

 

[AnotherReader]

Even without reviews, we can conclude that Zen 5 won't be worth an upgrade from Zen 4, unless one likes to tinker with new stuff. +10-15% IPC sounds cool, but isn't worth spending money on, in my opinion.

Edit: The last really good single generational upgrade was between Zen 2 and 3, as it was a completely new architecture with new core and cache layout. Since then, AMD has really been "Intel-ing".

The Intel model used to work well until Haswell and AMD's version of it is improved; generation to generation increases are greater than both Intel's peaks and their norm. Even Zen 4 is a significant increase over Zen 3. Remember that it isn't just IPC; clocks matter and Zen 4 is the biggest clock speed increase for AMD since they switched to the Ryzen cores.

As for this "new" chipset, I would have preferred it if they had changed the link between the chipset and the CPU to use PCIe 5 instead of PCIe 4. Then the chipset could have provided 16 lanes of PCIe 3 for up to 4 PCIe 3 m.2 slots or whatever else the motherboard maker wanted to use.

 

[trsttte]

X670 is unreasonably expensive. I'd rather look at B650 unless you absolutely can't live without the extra PCI-e lanes.

You don't even really get more pcie lanes, you get some extra usb and sata ports and an extra m.2. I mean technically that is extra lanes but whatever, x670 is kind of a scam imo, a good b650 board is much more interesting, you can have pretty much all the same features without the stupid price. Only problem is market segmentation and how most b650 also suck.

I would have preferred it if they had changed the link between the chipset and the CPU to use PCIe 5 instead of PCIe 4.

Got to save something for zen6 and x970/b950

 

[JoeTheDestroyer]

What board did you settle on?

Asrock X670E PG Lightning. Had it for nearly a year now, no complaints.

 

[Wirko]

It's a shame that AMD and Asmedia aren't willing to sell these Prom 21 chips as separate products. There aren't many non-enterprise PCIe switches in existence.

 

[Tek-Check]

No, you have an Intel JHL8540 Thunderbolt 4 chip that is slower than the ASM4242 when it comes to USB4 data transfers due to it being PCIe 3.0 rather than PCIe 4.0.

I have something to clarify here. Please have a look at the specification of the controller below. Is this how TB5 and ASM4242 should roughly look like?
On ASM4242, PCIe data transfer is not that much faster than on TB4, ~3.8 GB/s vs. ~3 GB/s, as measured with several SSDs. In theory, over 40 Gbps port, USB4 PCIe data speed could be ~4.5 GB/s, but it will take the implementation od USB4 v2.0 ports at 80 Gbps to fully unlock the speed of PCIe link at 64 Gbps.

 

[LabRat 891]

X670 is unreasonably expensive. I'd rather look at B650 unless you absolutely can't live without the extra PCI-e lanes.

Re-realizing this, is making me de-regret sticking to AM4. I use every single lane and slot my X570 provides.
It was both notable, and disappointing how much 'expansion' has been removed entirely or relegated to (few, Gen4/Gen5) M.2 slots.

It's a shame that AMD and Asmedia aren't willing to sell these Prom 21 chips as separate products. There aren't many non-enterprise PCIe switches in existence.

I can think of at least 3 major providers, and have ran-across 3 'no names'. However, (IIRC) only 2 (ASM and PLX) are in active R&D on Gen4->.
Still, agreed.
TBH, I wish enthusiasts were allowed to 'piecemeal' our platforms. Ex. those PROM21 x4 Add-In cards ASRock showed off, really tickled my fancy.

 

[R0H1T]

Even without reviews, we can conclude that Zen 5 won't be worth an upgrade from Zen 4, unless one likes to tinker with new stuff.

Depends on what you're doing with it, for professional setups(TR?) with work counting/billing for every hour this will be a lot of help. There should also be some major changes on the NPU/XPU front which "could" be worth the extra price to some potential buyers. Overall though, with zen4 probably already topping out clocks on normal Si based chips, I would be surprised if AMD got more than 15~20% uplift in overall performance, whether it be ST &/or MT benches. I suspect the biggest changes/uplift would be down to some new instructions or even IGP.

 

[AusWolf]

Depends on what you're doing with it, for professional setups(TR?) with work counting/billing for every hour this will be a lot of help. There should also be some major changes on the NPU/XPU front which "could" be worth the extra price to some potential buyers. Overall though, with zen4 probably already topping out clocks on normal Si based chips, I would be surprised if AMD got more than 15~20% uplift in overall performance, whether it be ST &/or MT benches. I suspect the biggest changes/uplift would be down to some new instructions or even IGP.

The IGP is part of the IO die. As long as that remains the same, the IGP remains the same, too. IPC uplifts are nice, but not necessary on a consumer level. Ryzen 7000, or even 5000 is plenty fast already. As for the NPU... is there even a single application that uses it?

 

[R0H1T]

The IGP's probably RDNA 3.5 based on the rumors for Strix Halo, just guesstimating so don't quote me on that. Like I said for someone using it for their work (i.e. printing money) it could make sense if their work is (extremely) time sensitive. For XPU/NPU it was used for copilot(?) IIRC as well as camera & maybe background noise(isolation) as well. Definitely for cam so it would be helpful in similar tasks.

 

[Tek-Check]

You loose 8 lanes Gen4 just for the connection of the chipsets. Plus it makes no sense to connect more than one SSD

You loose or you invest lanes for expanded connectivity. Glass is both half empty and half full, depending on perception and point of view.

Chipset silicon is much cheaper this way, and all additional lanes on the second chipset make a difference between B650 and X670. It's really simple - either less connectivity or more connectivity.

If anyone wants even more, which is rarely the case, there are boards maximizing the connectivity and x4 AICs with USB4, SATA, NVMe, etc. Research also shows that the usage of secondary x8 GPU slot is low. And Gigabyte got rid off it. You can love then ir hate them for that, but it's a clear signal for their buyers - if you use only one x16 slot for GPU, here are boards for you. Asrock and Asus more often than not have two x16 slots, running at x8 idmf both are occupied, for those who need more connectivity.

It's about choice And there's plenty of it.

The IGP's probably RDNA 3.5 based on the rumors for Strix Halo, just guesstimating so don't quote me on that. Like I said for someone using it for their work (i.e. printing money) it could make sense if their work is (extremely) time sensitive. For XPU/NPU it was used for copilot(?) IIRC as well as camera & maybe background noise(isolation) as well. Definitely for cam so it would be helpful in similar tasks.

Yes. Ryzen 9000 will have a substantial update on iGPU, with at least RDNA 3 engine, probably 2CUs, plus we expect XDNA cores on the silicon too, in readiness for Windows 12 Copilot and other tasks.

iGPU update is another improvement on I/O die, so I don't understand when people say "it's the same I/O". Well, clearly it's not.

The only thing AMD could have provided better was integrated USB4 on die rather than discrete solution via ASM4242. They have had integrated USB4 since Rembrandt APU, it has finally arrived on desktop APUs too with Phoenix, so it's time that 'regular' desktop CPUs also integrate USB4. Arrow Lake S will have TB4 integrated, the first time Intel does this on desktop CPUs.

 

[AusWolf]

Yes. Ryzen 9000 will have a substantial update on iGPU, with at least RDNA 3 engine, probably 2CUs, plus we expect XDNA cores on the silicon too, in readiness for Windows 12 Copilot and other tasks.

iGPU update is another improvement on I/O die, so I don't understand when people say "it's the same I/O". Well, clearly it's not.

The only thing AMD could have provided better was integrated USB4 on die rather than discrete solution via ASM4242. They have had integrated USB4 since Rembrandt APU, it has finally arrived on desktop APUs too with Phoenix, so it's time that 'regular' desktop CPUs also integrate USB4. Arrow Lake S will have TB4 integrated, the first time Intel does this on desktop CPUs.

So are they redesigning the IO die on Zen 5 then? I didn't think they were, but I'm honestly confused at this point.

 

[Tek-Check]

So are they redesigning the IO die on Zen 5 then? I didn't think they were, but I'm honestly confused at this point.

Why confused? It's a proper new generation and not a refresh.

Even if they etch I/O die again on 6nm, its going to have a different design due to feature upgrades. A million dollar question is which features, apart from new iGPU and a bit faster memory controller, would be included. Media engine could also get a new codec capability alongside RDNA upgrade, so AV1 encoding should come to desktop x86 CPU for the first time.

We haven't seen any reliable leaks dissecting I/O. Leakers clearly have no idea either and AMD has been guarding information quite well.

 

[R0H1T]

Some of these chips will also go into laptops (8945hx?) so the I/O die would almost certainly need to be completely revamped. Now whether we'll also see zen5c on regular desktop/laptop chips is what I'm really waiting for!

 

[AusWolf]

Why confused? It's a proper new generation and not a refresh.

Even if they etch I/O die again on 6nm, its going to have a different design due to feature upgrades. A million dollar question is which features, apart from new iGPU and a bit faster memory controller, would be included. Media engine could also get a new codec capability alongside RDNA upgrade, so AV1 encoding should come to desktop x86 CPU for the first time.

We haven't seen any reliable leaks dissecting I/O. Leakers clearly have no idea either and AMD has been guarding information quite well.

Proper new generation on the compute die doesn't necessarily bring new IO with it automatically. I thought this to be (one of) the main advantage(s) of the modular/chiplet design in the first place - a lot less spent on R&D.

I guess we'll see when it comes out.

 

[Arco]

About the costs of 670 and 670E. I just went Asrock. Not sure of the prices now but it was around $250 for my 670E Steel Legend. Decent motherboard that fits my uses. Other companies jack up the prices. Like so much that you get a 650E for the same price you can buy a 670E from Asrock.

 

[TheLostSwede]

I have something to clarify here. Please have a look at the specification of the controller below. Is this how TB5 and ASM4242 should roughly look like?
On ASM4242, PCIe data transfer is not that much faster than on TB4, ~3.8 GB/s vs. ~3 GB/s, as measured with several SSDs. In theory, over 40 Gbps port, USB4 PCIe data speed could be ~4.5 GB/s, but it will take the implementation od USB4 v2.0 ports at 80 Gbps to fully unlock the speed of PCIe link at 64 Gbps.

View attachment 333357

USB4 should have an actual USB4 data mode as well, that isn't PCIe. See https://en.wikipedia.org/wiki/USB4#Data_transfer_modes
3.8 Gbps is about what ASMedia has managed to squeeze out of their implementation, so that's correct.
USB4 doesn't support native networking, it was a stipulation by a certain company.

Some test results from ASMedia at Computex last year.

 

[csendesmark]

It would be nice to see the 10Gbps LAN became standard... beside USB4
Since 1Gbps LAN became standard beside the good old USB v2.0 in the early 2000-s

 

[TheLostSwede]

About the costs of 670 and 670E. I just went Asrock. Not sure of the prices now but it was around $250 for my 670E Steel Legend. Decent motherboard that fits my uses. Other companies jack up the prices. Like so much that you get a 650E for the same price you can buy a 670E from Asrock.

Pricing was meant to be much lower originally, but everyone jacked them up due to increased shipping cost, increased material costs and what not. From Computex to launch that year, most boards went up by US$50 compared to the pricing I was given at Computex by one manufacturer.

It would be nice to see the 10Gbps LAN became standard... beside USB4
Since 1Gbps LAN became standard beside the good old USB v2.0 in the early 2000-s

5 Gbps is likely to be be next step. The simple reason for this is power efficiency. 2.5 and 5 Gbps Ethernet is a lot less power hungry compared to 10 Gbps so far.
Then there's cost as well and with Realtek now having a low-cost 5 Gbps part which will be sub $5 vs Marvell/Aquantia at $20-25 for a 10 Gbps chip... You do the math.

 

[AusWolf]

It would be nice to see the 10Gbps LAN became standard... beside USB4
Since 1Gbps LAN became standard beside the good old USB v2.0 in the early 2000-s

It would be nice to see even 100 Mbps broadband become standard across the UK first.

 

[Tek-Check]

USB4 should have an actual USB4 data mode as well, that isn't PCIe. See https://en.wikipedia.org/wiki/USB4#Data_transfer_modes

I looked at it yesterday. It reads that P2P IP networking is natively supported. On Windows 11 forum there are specific screenshots of IP networking drivers in device manager. It shows 20 Gbps of network traffic.

I was also puzzled by actual USB4 data mode. The wording of the spec of ASM4242 on their website suggests that the controller provides USB4 Gen 3 x 2 bandwidth for data traffic, mostly tunnelled, including up to USB 3.2 20 Gbps.

There is no mention that USB4 native data traffic is supported at 40 Gbps, unless I missed something obvious.

 

[Darkholm]

Please correct me if I'm mistaken.
New X870E is X670E with USB4 controller. B850 is B650E and B840 is B650. And we will have a small 20% price hike so if you would like post code debug display on-board which costs 3 cents, you will have to buy at least B850 board for lets say €400 as a midrange.
I think I will stay with AM4 and B550-E for a long long time.

 

[Tigerfox]

PCIe x1 quad SATA add-in cards are dirt cheap and AFAIK you can get x4 models so that the drives each get decent bandwidth, but again... this requires a board with an x4 slot.

Exactly and since I wanted to keep my two x4-slots free for soundcard, 10GbE or TB4/USB4, i bought a M.2-AIC with ASM1166 (6xSATA@Gen3x2). Coincidentally, my board has one M.2-slot which ist Gen3x2 or SATA (why on earth did they use it for that instead of 2xSATA?!?).

The problem is space on the board rather than lanes. With a modern GPU, that blocks 3-5 slots, there isn't much room left for PCIe-slots usable for AIC/soundcards/Storage-controllers/10GbE/videocapture/USB4/TB4.

 

[Knight47]

If 4TB and 8TB nvme were cheaper and available ya but I can buy three 2TB for the price of one 4TB drive. I have 3/4 slots filled.

I have all m.2 slots filled, two 2TB and two 4TB drives, fortunately I can use two more NVMe with PCIe adapters. I miss the old days when boards had 8+ Sata ports.
I could get Sata drives like 870 Evo 4TB but that costs the same as a 990 Pro 4TB where I live.

 

[Tek-Check]

. I miss the old days when boards had 8+ Sata ports.

ITX boards are, of course, limited by space. Many bigger boards still have 6 or 8 SATA ports