Jump to content
Toggle menu
Toggle preferences menu
Toggle personal menu
Not logged in
Your IP address will be publicly visible if you make any edits.

Every platform report: boards and servers people ran CMP 170HX cards in, with the settings that mattered. Click a column heading to sort; long values are cut to one line, hover for the full text or open the report.

Filter by card variant: 8 GB only · 10 GB only · tested with both. Most results (Gen2, BAR1, P2P) differ between the two variants, so check the variant before copying settings.


On a phone this table shows the platform, card variant, the Gen2, BAR1 and P2P results and the status. Open a report for the rest, or turn the phone sideways.
PlatformVariantCPUGen2ReBAR / BAR1P2PCards per hostStatusTested by
ASRock Rack ROMED8-2TSingle-socket AMD EPYC (SP3), for example EPYC 7543Reportedfnerdman
ASUS P9X79 WSIntel LGA2011 (X79)Unlock works after the MMIOH change; BAR1 size not reportedReportedfdelfini
ASUS Pro WS WRX80E-SAGE SE WIFIAMD Threadripper Pro (sWRX8)BAR1 cannot be assigned ("no space"); stays 64 MBFixed by the curiouspixel P2P patch in the Morrowmake cmpunlocker fork (works without the 64 GB BAR1)Reportedtrustinter
ASUS Pro WS WRX90E-SAGE SEAMD Threadripper Pro 7000 (sTR5)BAR1 cannot be assigned (same "no space" problem as the WRX80E)Fixed by the curiouspixel P2P patch in the Morrowmake cmpunlocker fork, without the 64 GB BAR1Reportedghostinthedatacenter_87281
ASUS X99-E WSIntel Xeon E5-2695 v4one slot never trains x16, another later dropped to x8Reportedmatstatman
Dell OptiPlex 7040mIntel Core i5 ("5 core" as posted)(OcuLink from M.2)n/a (one card)Reportedglobalapathy
Dell PowerEdge R720Dual Intel Xeon E5 v1/v2 (LGA2011, DDR3)Theory
Dell PowerEdge R730Dual Intel Xeon E5 v3/v4Reportednathang6866
Dell PowerEdge R740Dual Intel Xeon ScalableReportedchonsworth
Dell Precision T5810Intel Xeon E5 v3/v4 (single socket)Reportedkellertreppe123
Gigabyte G292-Z20Single-socket AMD EPYCEach GPU pair sits on its own PCIe switch: good P2P within a pair, slower between pairsProven1a2b3c0967
Gigabyte G481-HA0Dual Intel Xeon Gold 6152All cards at x16 (generation not stated)Reportedredpandamining
Gigabyte MZ31-AR0Single-socket AMD EPYC (SP3)two cards limited to x8 by slot wiringReportedfinishtilt14
Gigabyte MZ32-AR08 GBSingle-socket AMD EPYC Rome or Milan (7K62 suggested)Reportedzorg33
Huananzhi H12D-8DSingle-socket AMD EPYC 7002/7003 (7K62 suggested)Works (Miséricorde)Reportedwakeupmbrc1
Huananzhi X99-F88 GBIntel Xeon E5 v3/v4 (LGA2011-3), for example E5-2696 v3Reportedwakeupmbrc1
Hyve Edge Metal G10AMD EPYCReportedport00
Minisforum MS-01Intel (MS-01 mobile CPU, model not given)BAR1 stuck at 64 MB on 4 cards; 2 of 4 reached 64 GB after MMIOH changes; MMIOH size prevents all 4replacedriver47: no transfers with the Morrowmake P2P fork, even with 2 cards at 64 GB BAR1; Con3po: mailbox P2P with 8 cards on MS-01sReportedreplacedriver47
Supermicro H12DSG-O-CPU8 GBDual AMD EPYC RomeGen2 x16 on all 8 cards64 GB BAR1 on all 8 cards (firmware sizes 64 MB, the patched driver resizes to 65536 MB)Working on all 8 cards, across both sockets: <code>'"`UNIQ--nowiki-000000FE-QINU`"'</code> 504/504 with the Morrowmake P2P build; <code>'"`UNIQ--nowiki-000000FF-QINU`"'</code> = 1 for every pairReporteddarkmaniac7
Supermicro SYS-4028GR-TRT2Dual Intel Xeon E5 v3/v4 (E5-2673 v4 used), DDR4 up to 2400 MHzAll 10 GPU slots on CPU1 behind two PEX9797 (PCIe 3.0) switches, five slots each: in-switch P2P expected good, cross-switch traffic goes through the CPUProvenmarkshowa
Supermicro X11SPA-TF10 GBIntel Xeon Scalable or Xeon W-320010 GB cards at Gen3 on this Gen3 board (after the 10 GB Gen3 unlock)Reporteddjpandaman
Tyan S8030GM2NE10 GBAMD EPYC 7R13 (48 cores)Reporteddjpandaman